JP2004012351A - Equipment, method, and program for tracking target - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide target tracking equipment which uses not only information on positions but information on Doppler speeds (the rates of change in distances with respect to time), and performs tracking processing by the use of a postponed determination type correlation system. <P>SOLUTION: Gate computing portions 4a, 4b, observed value selecting portions 5a, 5b, a hypothesis reducing portion 15, smoothing portions 17a, 17b, and a track determining portion 16, are provided. The portions 4a, 4b and the portions 5a, 5b extract observation information having possibilities of correlation concerning the Doppler speeds of the targets to be tracked and the target positions at the present time, with respect to the tracks of the targets found until that time. Using the observation information extracted by these, the portion 15 forms groups of hypotheses which represent combinations of tracks of the targets presumed at each point of observation time, performs evaluation of certainty for each hypothesis in the groups, and reduces the number of hypotheses in the groups on the basis of the evaluation result. The portions 17a, 17b perform smoothing for various motion factors including the Doppler speeds of the targets, with respect to each track in the groups. The portion 16 selects a hypothesis which represents the motion factors at the present time of the targets most exactly from the groups. <P>COPYRIGHT: (C)2004,JPO

Description

【００１９】
但し、上記式（１）で示すクラスタ内ゲート内判定行列の各行は、現時刻ｔ_ｋの観測値ｚ_ｋ，ｊ（ｊ＝１，２，・・・，ｍ_ｋ）に対応する。また、各列は、現時刻ｔ_ｋでのクラスタ内の航跡Ｔ_ｔ（ｔ＝０，１，２，・・・，Ｎ_ｋ）に対応する。行列の各要素は、行に対応する観測値が列に対応する航跡のゲート内にあるか否かを表している。具体的には、以下のように設定される。
【００２０】
先ず、ｔ＝０の列は、観測値が誤信号による値である場合を示している。実際に全ての観測値は誤信号である可能性があるとして、ω_ｊ，０＝１のように設定する。
【００２１】
次に、１≦ｔ≦Ｎ_ｋ−１の各列は、クラスタ内の既存の航跡Ｔ_ｔに対応している。ここで、観測値ｚ_ｋ，ｊ（ｊ＝１，２，・・・，ｍ_ｋ）が、既存の航跡Ｔ_ｔ（ｔ＝０，１，２，・・・，Ｎ_ｋ）のゲートに含まれる場合、ω_ｊ，ｔ＝１のように設定する。また、観測値ｚ_ｋ，ｊが、既存の航跡Ｔ_ｔのゲートに含まれない場合には、ω_ｊ，ｔ＝０のように設定する。
【００２２】
さらに、Ｎ_ｋ−１＋１≦ｔ≦Ｎ_ｋの各列は、現時刻ｔ_ｋで新たに現れたクラスタ内の航跡に対応する。これら列は、全ての観測値がそれぞれ新たに発見された目標である可能性を表すものである。ここで、１個の観測値が１本の新たな航跡に対応するように要素を設定する。つまり、ｊ＝ｔ−Ｎ_ｋ−１の場合には、ω_ｊ，ｔ＝１のように設定する。反対に、ｊ≠ｔ−Ｎ_ｋ−１の場合には、ω_ｊ，ｔ＝０のように設定する。
【００２３】
続いて、航跡相関行列算出部１１１は、クラスタ内ゲート内判定行列から全てのクラスタ内航跡相関行列を算出する（ステップＳＴ１０５）。ここで、クラスタ内航跡相関行列とは、実際に仮説として取り得るクラスタ内の観測値と航跡との相関関係を示すものである。一般に、１つのクラスタ内ゲート内判定行列から複数のクラスタ内航跡相関行列が生成される。現時刻ｔ_ｋにおける、ｓ＝１，２，・・・，ＳのＳ個のクラスタ内航跡相関行列は、クラスタ内ゲート内判定行列から下記式（２）に示すように定義される。
【数６】

【００２４】
ここで、クラスタ内航跡相関行列の各行は、クラスタ内ゲート内判定行列の場合と同様に、現時刻ｔ_ｋでのクラスタ内の観測値に対応しており、各列はクラスタ内の航跡に対応している。また、行列の各要素は、行に対応する観測値が列に対応する航跡と相関しているか否かを示すものである。具体的には、以下のように設定される。
【００２５】
先ず、観測値ｚ_ｋ，ｊ（ｊ＝１，２，・・・，ｍ_ｋ）が航跡Ｔ_ｔ（ｔ＝０，１，２，・・・，Ｎ_ｋ）と相関がある場合には、ω_ｊ，ｔ ^ｓ＝１のように設定する。また、観測値ｚ_ｋ，ｊが航跡Ｔ_ｔと相関がないと、ω_ｊ，ｔ ^ｓ＝０のように設定する。
【００２６】
実際に、航跡相関行列算出部１１１がクラスタ内ゲート内判定行列からクラスタ内航跡相関行列を作成する際には、次に示す（ア）から（ウ）までの３つの基準を同時に満たす全ての組み合わせを、各々別個のクラスタ内航跡相関行列として算出する。
【００２７】
（ア）クラスタ内ゲート内判定行列において、１である要素に対応するクラスタ内航跡相関行列の要素のみを１とでき、その他の要素は０とする。
（イ）クラスタ内航跡相関行列のｔ＝０の列以外の全ての列では、高々１つの要素のみを１とし他の要素は０とする。
（ウ）クラスタ内航跡相関行列の全ての行では、必ず１つの要素を１とし他の要素は０とする。
【００２８】
次に、クラスタ内の仮説を更新するため、仮説更新部１１３は、クラスタ内の既存の仮説、即ち前時刻ｔ_ｋ−１で作成された仮説に対し、この仮説と組み合わせ可能な航跡相関行列を選択する。そして、両者の組み合わせから、現時刻ｔ_ｋに対応した新たな仮説を作成する。この操作を繰り返し実行することで、クラスタ内の仮説が更新される（ステップＳＴ１０６）。
【００２９】
ここで、クラスタ内から選択した既存の仮説を親仮説、新たな仮説を子仮説と呼ぶことがある。実際には、航跡相関行列において観測値と相関があるとされている既存航跡が親仮説内に含まれていない場合、両者を組み合わせることはできないものとする。以下に、ステップＳＴ１０６における詳細な動作を説明する。
【００３０】
先ず、仮説更新部１１３は、クラスタ内の既存の仮説を親仮説として１個選択し、この親仮説から派生する子仮説群を導くために、記憶部１１２内のクラスタ内航跡相関行列の１個を選択する。このあと、仮説更新部１１３は、上記親仮説に対してクラスタ内航跡相関行列が組み合わせ可能か否かを判定する。
【００３１】
具体的には、上述のようにして選択されたクラスタ内航跡相関行列において、いずれかの観測値に相関がある既存の航跡が親たな仮説内に含まれていない場合、両者を組み合わせ不可能とする。一方、その他の全ての場合を組み合わせ可能と判定する。
【００３２】
次に、上述のようにして組み合わせ可能と判定された場合に限り、既存の仮説内の既存の航跡に対して、これと相関がある新しい観測値を追加して航跡を伸張すること、新たな航跡を仮説に追加すること、ある観測値を誤信号として取り扱うことによって子仮説を作成する。以上の処理は、ステップＳＴ１０５で算出したクラスタ内航跡相関行列を全て選択し終わるまで繰り返し実行される。
【００３３】
このあと、仮説更新部１１３は、上述した一連の仮説更新処理を終了したか否かを判断しながら、クラスタ内の残余の仮説がなくなるまで上記処理を繰り返す。
【００３４】
なお、多くの場合、１つの既存の仮説が複数のクラスタ内航跡相関行列と組み合わされて、複数の子仮説が作成される。このため、ステップＳＴ１０６の仮説更新処理によってクラスタ内の仮説数は増加する。
【００３５】
次に、仮説縮小部１１４は、ステップＳＴ１０６で作成した現時刻ｔ_ｋの各仮説の信頼度評価値を算出し、全ての仮説の中から上記信頼度が高いものを残して他を削除するなどの方法で、仮説数の縮小を行う（ステップＳＴ１０７）。なお、上記信頼度評価値の具体的な算出方法は後述する。
【００３６】
続いて、航跡平滑部１１６は、クラスタ内に作成された現時刻ｔ_ｋの各航跡に対してカルマンフィルタの平滑処理を施すことによって、位置、速度の平滑化を行う（ステップＳＴ１０８）。具体的には、現時刻ｔ_ｋより新たに開始された航跡に対して、位置と速度の情報で構成される状態ベクトル及びその誤差共分散行列の初期値を設定する。
【００３７】
次に、現時刻ｔ_ｋの新たな観測値が追加された航跡に対し、この観測値を用いて上記状態ベクトル及び誤差共分散行列の平滑値を更新する。このあと、現時刻ｔ_ｋに対応する観測値が存在しないとされた航跡に対して、上記状態ベクトル及び誤差共分散行列の予測値を平滑値に置き換える、いわゆるメモリ追尾処理を実行する。
【００３８】
続いて、目標追尾装置１０１は、システム内クラスタ表に定義された全てのクラスタについてステップＳＴ１０４からステップＳＴ１０８までの処理を実行したか否かを判断する（ステップＳＴ１０９）。ここで、全てのクラスタについて処理が終了した場合には、ステップＳＴ１１０の処理に移行する。
【００３９】
ステップＳＴ１１０において、クラスタ分離部１１７は、クラスタ内の各航跡が有する観測値の構成を調べ、「現時刻ｔ_ｋに至るまでに少なくとも１つの観測値を共有した航跡同士は、同一のクラスタに属する」との条件に照らし、クラスタ内の航跡群が複数のクラスタに分離可能であるか否かを評価する。そして、分離可能である場合に、クラスタの分離と仮説の再構築を行い、クラスタ内仮説状況データ群を更新する。なお、このクラスタ分離処理によって、ある航跡がその開始以来はじめて単独の航跡のみで１つのクラスタを成すと判定された場合、この航跡は確立したものと判定される。
【００４０】
このあと、航跡決定部１１５は、クラスタ内の現時刻ｔ_ｋの仮説中から最善の仮説を１つ選択することによって目標の数及び航跡を決定する（ステップＳＴ１１１）。これら航跡群は、目標表示装置１０２のディスプレイ上に表示される。
【００４１】
以上述べた一連の処理は、目標追尾装置１０１によって追尾終了が検出されるまで繰り返し実行される（ステップＳＴ１１２）。
【００４２】
次に、文献２に従って、ゲート算出部１０３、航跡平滑部１１６における追尾フィルタ処理、観測値選択部１０４によるゲート判定処理、仮説縮小部１１４による仮説の信頼度評価値の算出処理について説明する。これらの処理の中心となる追尾フィルタ処理は、いわゆるカルマンフィルタを利用して以下のように実施される。
【００４３】
先ず、時刻ｔ_ｋにおける位置と速度により構成される追尾目標の状態ベクトルｘ_ｋと、位置情報により構成される観測値ベクトルｚ_ｋを、下記式（３）、（４）でそれぞれ定義する。ここで、Ｔは、行列の転置を表す記号である。また、ｘ_ｋ，ｙ_ｋ，ｚ_ｋは、目標の位置座標真値を表しており、ｖｘ_ｋ，ｖｙ_ｋ，ｖｚ_ｋは、目標の速度成分真値を表している。さらに、ｘ^ｏ _ｋ，ｙ^ｏ _ｋ，ｚ^ｏ _ｋは、目標の観測位置座標を表し、レーダによる距離、仰角、方位角の極座標系での観測結果を、北基準直交座標系に座標変換した値である。
【００４４】
【数７】

【００４５】
ここで、追尾目標の状態ベクトルの時間的推移を表す運動モデルを、下記式（５）で定義する。また、レーダの観測モデルを下記式（６）で定義する。なお、Φ_ｋは状態遷移行列、Ｈ_ｋは観測行列である。また、ｗ_ｋは、平均ゼロ、共分散行列Ｑ_ｋの正規性白色雑音に従う駆動雑音ベクトルであり、ｖ_ｋは平均ゼロ、共分散行列Ｒ_ｋの正規性白色雑音に従う観測雑音ベクトルである。
【数８】

【００４６】
上記式（５）の運動モデル式において、例えば目標の運動を等速直線運動とみなしてモデル化すれば、状態遷移行列を下記式（７）で定義することができる。ここで、Ｉ_３，Ｏ_３は、それぞれ３行３列の単位行列と零行列である。このとき、駆動雑音ベクトルｗ_ｋは、目標運動を等速直線とみなしたことによる打ち切り誤差項を表現したものと捉えることができる。また、上記式（６）の観測モデル式における観測行列Ｈ_ｋは、状態ベクトルから位置情報を抽出するための行列であり、下記式（８）で定義される。
【数９】

【００４７】
上記定義に基づいて、ゲート算出部１０３は、クラスタ内の各航跡に対して下記式（９）、（１０）に示すカルマンフィルタの予測処理を実行する。ここで、ｘ_{ｋ−１｜ｋ−１}は、時刻ｔ_ｋ−１の状態ベクトルの平滑値を表している。また、ｘ_{ｋ｜ｋ−１}は、時刻ｔ_ｋの状態ベクトルの予測値を表している。さらに、Ｐ_{ｋ−１｜ｋ−１}，Ｐ_{ｋ｜ｋ−１}は、それぞれ平滑値ｘ_{ｋ−１｜ｋ−１}、予測値ｘ_{ｋ｜ｋ−１}の誤差共分散行列を表している。
【数１０】

【００４８】
続いて、ゲート算出部１０３は、下記式（１１）、（１２）に従って、航跡の現時刻ｔ_ｋに対する予測位置ｚ_{ｋ｜ｋ−１}と残差共分散行列Ｓ_ｋを算出する。ここで、予測位置ｚ_{ｋ｜ｋ−１}はゲートの中心位置に相当し、残差共分散行列Ｓ_ｋはゲート領域の広がりを特徴づける諸元である。
【数１１】

【００４９】
一方、観測値選択部１０４は、上述のようにして算出された予測位置ｚ_{ｋ｜ｋ−１}、残差共分散行列Ｓ_ｋを使用して、下記式（１３）を満たす観測値ｚ_ｋを、当該航跡のゲート内の観測値であると判定する。但し、ｄはゲートサイズを決めるパラメータである。
【数１２】

【００５０】
次に仮説縮小部１１４による各子仮説の信頼度評価値の算出方法を説明する。この評価値は、次の手順によって算出される。先ず、各子仮説Ｈ（ｉ）の尤度γ_ｉを下記式（１４）を用いて計算する。
【数１３】

【００５１】
ここで、β_ｐ（ｉ）は、子仮説Ｈ（ｉ）を生んだ親仮説の前時刻の信頼度評価値である。Ｎ_ＤＴは、子仮説内でいずれかの観測値により伸張された航跡の数であり、Ｎ_ＭＴは、子仮説内でいずれの観測値にも対応しないものとされた既存航跡の数を示している。Ｎ_ＮＴは、子仮説内で新たに開始された航跡の数であり、Ｎ_ＦＴは、子仮説内で誤信号として扱われた観測値の数を表している。また、Ｐ_Ｄは、レーダの探知確率である。Ｐ_Ｇは、ゲート内に観測値が捕捉される確率であり、β_ＮＴは、単位体積あたりの新目標発生数の平均値である。β_ＦＴは、単位体積あたりの誤信号発生数の平均値である。これらはパラメータである。
【００５２】
ｇ_ｊ，ｔは、航跡Ｔ_ｔが観測値ｚ_ｋ，ｊにより伸張された場合の航跡の尤度である。この尤度ｇ_ｊ，ｔを観測値が航跡Ｔ_ｔの予測位置ｚ_{ｋ｜ｋ−１}を中心に共分散行列Ｓ_ｋの３変量ガウス分布で発生すると仮定して、下記式（１５）により計算する。
【数１４】

【００５３】
仮説縮小部１１４は、上記仮説尤度γ_ｉを全ての子仮説について算出した後、下記式（１６）に示すように、これらの値を全仮説に渡って正規化したものを、各子仮説の現時刻ｔ_ｋの信頼度評価値β_ｉとする。
【数１５】

【００５４】
次に、航跡平滑部１１６は、カルマンフィルタの平滑処理式に従い、以下の計算を実行する。即ち、航跡Ｔ_ｔに観測値ｚ_ｋ，ｊが追加され伸張された場合、下記式（１７）、（１８）、（１９）に従って平滑値の更新を実行する。
【数１６】

【００５５】
ここで、ｘ_{ｋ｜ｋ−１}、Ｐ_{ｋ｜ｋ−１}は、それぞれ上記式（９）、（１０）で算出した航跡Ｔ_ｔの状態ベクトル予測値及び予測誤差共分散行列である。また、ｘ_ｋ｜ｋは、現時刻ｔ_ｋに対応する航跡Ｔ_ｔの状態ベクトル平滑値、Ｐ_ｋ｜ｋは、ｘ_ｋ｜ｋの誤差共分散行列である。なお、Ｋ_ｋは、カルマンゲイン行列である。一方、航跡Ｔ_ｔに対応する現時刻の観測値が存在しない場合、下記式（２０）、（２１）に示すメモリ追尾処理を実行する。
【数１７】

【００５６】
【発明が解決しようとする課題】
従来の目標追尾装置は以上のように構成されているので、レーダからの位置の観測情報のみに基づいて航跡の推定を行い、目標の運動諸元の算出を行っているので、誤警報やクラッタなどの誤信号が多く発生する観測環境では、たとえ延期決定型の相関方式であっても、追尾性能に限界が生じるという課題があった。
【００５７】
この発明は上記のような課題を解決するためになされたもので、レーダから得られる位置の情報のみならず、ドップラ速度（距離の時間変化率）の情報を使用して延期決定型の相関方式を用いた追尾処理を実行する目標追尾装置及び方法並びにプログラムを得ることを目的とする。
【００５８】
【課題を解決するための手段】
この発明に係る目標追尾装置は、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度について相関の可能性がある観測情報を抽出するドップラ速度観測値抽出手段と、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における目標位置について相関の可能性がある観測情報を抽出する位置情報観測値抽出手段と、両観測値抽出手段によって抽出された観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成する仮説作成手段と、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減する仮説縮小手段と、仮説群内の各航跡に対して、当該目標のドップラ速度を含む運動諸元にて平滑化を実行する平滑化手段と、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択する仮説選択手段とを備えるものである。
【００５９】
この発明に係る目標追尾装置は、仮説縮小手段が、仮説群中の各仮説の確からしさを評価するにあたり、前時刻までに求めた航跡が現時刻の目標の観測値により伸張された場合における当該目標の位置及びドップラ速度による航跡の尤度を求め、これら航跡の尤度を用いて前時刻までに求めた航跡から導かれる現時刻の仮説の尤度を評価結果として算出するものである。
【００６０】
この発明に係る目標追尾装置は、仮説縮小手段が、前時刻までに求めた航跡Ｔ_ｔ（ｔは仮説内の航跡数）が現時刻ｔ_ｋの目標の観測値ｚ_ｋ，ｊ（ｊは仮説内の観測値数）により伸張された場合における、当該目標の位置情報による航跡の尤度をｇ_ｊ，ｔ、当該目標のドップラ速度による航跡の尤度をｇ^Ｄ _ｊ，ｔ、現時刻の仮説を導く前時刻までに求めた航跡Ｔ_ｔからなる仮説の前時刻の信頼度評価値をβ_ｐ（ｉ）、当該目標についての探知確率をＰ_Ｄ、前時刻までに求めた航跡Ｔ_ｔと現時刻の観測値が目標位置及びドップラ速度について相関の可能性がある観測値情報として抽出される確率をＰ_Ｇ、観測空間の単位体積当たりの新たな目標の発生数の平均値をβ_ＮＴ、観測空間の単位体積当たりの誤信号発生数の平均値をβ_ＦＴ、観測値ｚ_ｋ，ｊに付随するドップラ速度の計測値をＤｒ^ｏ _ｋ，ｊ、現時刻の仮説内でいずれかの観測値により伸張された航跡の数をＮ_ＤＴ、現時刻の仮説内でいずれの観測値にも対応しないものとされた既存の航跡の数をＮ_ＭＴ、現時刻の仮説内で新たに開始された航跡の数Ｎ_ＮＴ、現時刻の仮説内で誤信号として扱われた観測値の数をＮ_ＦＴとし、観測値が航跡Ｔ_ｔの予測位置ｚ_{ｋ｜ｋ−１}を中心に共分散行列Ｓ_ｋの３変量ガウス分布で発生し、観測値のドップラ速度が航跡Ｔ_ｔのドップラ速度の予測値Ｄｒ_{ｋ｜ｋ−１}を中心に残差共分散行列Ｓ^Ｄ _ｋの１変量ガウス分布で発生するものと仮定して、前時刻までに求めた航跡Ｔ_ｔから導かれる現時刻の仮説の尤度γ_ｉを、上記式から算出するものである。
【００６１】
この発明に係る目標追尾装置は、仮説縮小手段が、前時刻までに求められた仮説の尤度に当該目標についての探知確率を乗算する探知確率乗算部と、探知確率乗算部の乗算結果に当該目標の観測値における目標位置に関する尤度を乗算する位置情報尤度乗算部と、位置情報尤度乗算部の乗算結果に当該目標の観測値におけるドップラ速度の尤度を乗算するドップラ速度尤度乗算部とを有する尤度乗算ブロック部と、前時刻までに求められた仮説の尤度に、観測空間の単位体積当たりの新たな目標の発生数の平均値である新目標発生率を乗算する新目標発生率乗算部と、前時刻までに求められた仮説の尤度に、当該仮説内の航跡に相関する観測値が位置及びドップラ速度について相関の可能性がある観測値情報として抽出されない度合いを示すゲート内非探知確率を乗算するゲート内非探知確率乗算部と、前時刻までに求めた仮説内の各航跡に対する現時刻の観測値の割り当て状態に応じて、探知確率乗算部、ゲート内非探知確率乗算部、及び新目標発生率乗算部のいずれか一に上記仮説の尤度を送出する航跡状態判定部と、前時刻までに求めた仮説の尤度に、観測空間の単位体積当たりの誤信号発生数の平均値である誤信号発生率を乗算する誤信号発生率乗算部と、前時刻までに求めた仮説内で観測値が誤信号と扱われている場合、前時刻までに求められた仮説の尤度を上記誤信号発生率乗算部に送る観測値状態判定部と、仮説作成手段が前時刻までに求めた仮説群を構成する全ての仮説の尤度の合計値を算出する仮説尤度集計部と、仮説尤度集計部が算出した合計値で仮説群内の各仮説の尤度を除算し、当該除算結果を各仮説の信頼度評価値として算出する仮説信頼度算出部と、仮説信頼度算出部が算出した仮説の信頼度評価値に基づいて仮説群中の仮説数を削減する仮説削減部とを有してなるものである。
【００６２】
この発明に係る目標追尾装置は、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度と目標位置とで規定される観測値ベクトルについて相関の可能性がある観測情報を抽出する観測値選択手段と、観測値抽出手段によって抽出された観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成する仮説作成手段と、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減する仮説縮小手段と、仮説群内の各航跡に対して、当該目標のドップラ速度と目標位置とで規定される観測値ベクトルにて平滑化を実行する平滑化手段と、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択する仮説選択手段とを備えるものである。
【００６３】
この発明に係る目標追尾装置は、仮説縮小手段が、仮説群中の各仮説の確からしさを評価するにあたり、目標の位置ベクトルとドップラ速度とを合成した観測値ベクトルの尤度を求め、これら航跡の尤度を用いて前時刻までに求めた航跡から導かれる現時刻の仮説の尤度を評価結果として算出するものである。
【００６４】
この発明に係る目標追尾装置は、仮説縮小手段が、前時刻までに求めた航跡Ｔ_ｔ（ｔは仮説内の航跡数）が現時刻ｔ_ｋの目標の観測値ｚ_ｋ，ｊ（ｊは仮説内の観測値数）により伸張された場合における、当該目標の観測位置及びドップラ速度計測値からなる観測値ベクトルｚ^Ｍ _ｋ，ｊの尤度をｇ^Ｍ _ｊ，ｔ、現時刻の仮説を導く前時刻までに求めた航跡Ｔ_ｔからなる仮説の前時刻の信頼度評価値をβ_ｐ（ｉ）、当該目標についての探知確率をＰ_Ｄ、前時刻までに求めた航跡Ｔ_ｔと現時刻の観測値が位置及びドップラ速度について相関の可能性がある観測値情報として抽出される確率をＰ_Ｇ、観測空間の単位体積当たりの新たな目標の発生数の平均値をβ_ＮＴ、観測空間の単位体積当たりの誤信号発生数の平均値をβ_ＦＴ、現時刻の仮説内でいずれかの観測値により伸張された航跡の数をＮ_ＤＴ、現時刻の仮説内でいずれの観測値にも対応しないものとされた既存の航跡の数をＮ_ＭＴ、現時刻の仮説内で新たに開始された航跡の数Ｎ_ＮＴ、現時刻の仮説内で誤信号として扱われた観測値の数をＮ_ＦＴとし、観測値ベクトルが航跡Ｔ_ｔの予測位置ｚ_{ｋ｜ｋ−１}を中心に残差共分散行列Ｓ^Ｍ _ｋの４変量ガウス分布で発生するものと仮定して、前時刻までに求めた航跡Ｔ_ｔから導かれる現時刻の仮説の尤度γ_ｉを、上記式から算出するものである。
【００６５】
この発明に係る目標追尾装置は、仮説縮小手段が、前時刻までに求められた仮説の尤度に当該目標についての探知確率を乗算する探知確率乗算部と、探知確率乗算部の乗算結果に当該目標の観測値における目標位置及びドップラ速度とで規定される観測値ベクトルの尤度を乗算する尤度乗算部とを有する尤度乗算ブロック部と、前時刻までに求められた仮説の尤度に、観測空間の単位体積当たりの新たな目標の発生数の平均値である新目標発生率を乗算する新目標発生率乗算部と、前時刻までに求められた仮説の尤度に、当該仮説内の航跡に相関する観測値が相関の可能性がある観測値情報として抽出されない度合いを示すゲート内非探知確率を乗算するゲート内非探知確率乗算部と、前時刻までに求めた仮説内の各航跡に対する現時刻の観測値の割り当て状態に応じて、探知確率乗算部、ゲート内非探知確率乗算部、及び新目標発生率乗算部のいずれか一に仮説の尤度を送出する航跡状態判定部と、前時刻までに求めた仮説の尤度に、観測空間の単位体積当たりの誤信号発生数の平均値である誤信号発生率を乗算する誤信号発生率乗算部と、前時刻までに求めた仮説内で観測値が誤信号と扱われている場合、前時刻までに求められた仮説の尤度を誤信号発生率乗算部に送る観測値状態判定部と、仮説作成手段が前時刻までに求めた仮説群を構成する全ての仮説の尤度の合計値を算出する仮説尤度集計部と、仮説尤度集計部が算出した合計値で仮説群内の各仮説の尤度を除算し、当該除算結果を各仮説の信頼度評価値として算出する仮説信頼度算出部と、仮説信頼度算出部が算出した仮説の信頼度評価値に基づいて仮説群中の仮説数を削減する仮説削減部とを有してなるものである。
【００６６】
この発明に係る目標追尾方法は、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度について相関の可能性がある観測情報を抽出するドップラ速度観測値抽出ステップと、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における目標位置について相関の可能性がある観測情報を抽出する位置情報観測値抽出ステップと、両観測値抽出ステップにて抽出された観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成する仮説作成ステップと、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減する仮説縮小ステップと、仮説群内の各航跡に対して、当該目標のドップラ速度を含む運動諸元にて平滑化を実行する平滑化ステップと、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択する仮説選択ステップとを備えるものである。
【００６７】
この発明に係る目標追尾方法は、仮説縮小ステップにて、仮説群中の各仮説の確からしさを評価するにあたり、前時刻までに求めた航跡が現時刻の目標の観測値により伸張された場合における当該目標の位置及びドップラ速度による航跡の尤度を求め、これら航跡の尤度を用いて前時刻までに求めた航跡から導かれる現時刻の仮説の尤度を評価結果として算出するものである。
【００６８】
この発明に係る目標追尾方法は、仮説縮小ステップにて、前時刻までに求めた航跡Ｔ_ｔ（ｔは仮説内の航跡数）が現時刻ｔ_ｋの目標の観測値ｚ_ｋ，ｊ（ｊは仮説内の観測値数）により伸張された場合における、当該目標の位置情報による航跡の尤度をｇ_ｊ，ｔ、当該目標のドップラ速度による航跡の尤度をｇ^Ｄ _ｊ，ｔ、現時刻の仮説を導く前時刻までに求めた航跡Ｔ_ｔからなる仮説の前時刻の信頼度評価値をβ_ｐ（ｉ）、当該目標についての探知確率をＰ_Ｄ、前時刻までに求めた航跡Ｔ_ｔと現時刻の観測値が位置及びドップラ速度について相関の可能性がある観測情報として抽出される確率をＰ_Ｇ、観測空間の単位体積当たりの新たな目標の発生数の平均値をβ_ＮＴ、観測空間の単位体積当たりの誤信号発生数の平均値をβ_ＦＴ、観測値ｚ_ｋ，ｊに付随するドップラ速度の計測値をＤｒ^ｏ _ｋ，ｊ、現時刻の仮説内でいずれかの観測値により伸張された航跡の数をＮ_ＤＴ、現時刻の仮説内でいずれの観測値にも対応しないものとされた既存の航跡の数をＮ_ＭＴ、現時刻の仮説内で新たに開始された航跡の数Ｎ_ＮＴ、現時刻の仮説内で誤信号として扱われた観測値の数をＮ_ＦＴとし、観測値が航跡Ｔ_ｔの予測位置ｚ_{ｋ｜ｋ−１}を中心に共分散行列Ｓ_ｋの３変量ガウス分布で発生し、観測値のドップラ速度が航跡Ｔ_ｔのドップラ速度の予測値Ｄｒ_{ｋ｜ｋ−１}を中心に残差共分散行列Ｓ^Ｄ _ｋの１変量ガウス分布で発生するものと仮定して、前時刻までに求めた航跡Ｔ_ｔから導かれる現時刻の仮説の尤度γ_ｉを、上記式から算出するものである。
【００６９】
この発明に係る目標追尾方法は、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度と目標位置とで規定される観測値ベクトルについて相関の可能性がある観測情報を抽出する観測値選択ステップと、観測値抽出ステップにて抽出された観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成する仮説作成ステップと、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減する仮説縮小ステップと、仮説群内の各航跡に対して、当該目標のドップラ速度と目標位置とで規定される観測値ベクトルにて平滑化を実行する平滑化ステップと、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択する仮説選択ステップとを備えるものである。
【００７０】
この発明に係る目標追尾方法は、仮説縮小ステップにて、仮説群中の各仮説の確からしさを評価するにあたり、目標の位置ベクトルとドップラ速度とを合成した観測値ベクトルの尤度を求め、これら航跡の尤度を用いて前時刻までに求めた航跡から導かれる現時刻の仮説の尤度を評価結果として算出するものである。
【００７１】
この発明に係る目標追尾方法は、仮説縮小ステップにて、前時刻までに求めた航跡Ｔ_ｔ（ｔは仮説内の航跡数）が現時刻ｔ_ｋの目標の観測値ｚ_ｋ，ｊ（ｊは仮説内の観測値数）により伸張された場合における、当該目標のドップラ速度計測値を含む観測値ベクトルｚ^Ｍ _ｋ，ｊの尤度をｇ^Ｍ _ｊ，ｔ、現時刻の仮説を導く前時刻までに求めた航跡Ｔ_ｔからなる仮説の前時刻の信頼度評価値をβ_ｐ（ｉ）、当該目標についての探知確率をＰ_Ｄ、前時刻までに求めた航跡Ｔ_ｔと現時刻の観測値が位置及びドップラ速度について相関の可能性がある観測値情報として抽出される確率をＰ_Ｇ、観測空間の単位体積当たりの新たな目標の発生数の平均値をβ_ＮＴ、観測空間の単位体積当たりの誤信号発生数の平均値をβ_ＦＴ、現時刻の仮説内でいずれかの観測値により伸張された航跡の数をＮ_ＤＴ、現時刻の仮説内でいずれの観測値にも対応しないものとされた既存の航跡の数をＮ_ＭＴ、現時刻の仮説内で新たに開始された航跡の数Ｎ_ＮＴ、現時刻の仮説内で誤信号として扱われた観測値の数をＮ_ＦＴとし、観測値ベクトルが航跡Ｔ_ｔの予測位置ｚ_{ｋ｜ｋ−１}を中心に残差共分散行列Ｓ^Ｍ _ｋの４変量ガウス分布で発生するものと仮定して、前時刻までに求めた航跡Ｔ_ｔから導かれる現時刻の仮説の尤度γ_ｉを、上記式から算出するものである。
【００７２】
この発明に係るプログラムは、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度について相関の可能性がある観測情報を抽出するドップラ速度観測値抽出手段、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における目標位置について相関の可能性がある観測情報を抽出する位置情報観測値抽出手段、両観測値抽出手段によって抽出された観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成する仮説作成手段、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減する仮説縮小手段、仮説群内の各航跡に対して、当該目標のドップラ速度を含む運動諸元にて平滑化を実行する平滑化手段、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択する仮説選択手段としてコンピュータを機能させるものである。
【００７３】
この発明に係るプログラムは、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度と目標位置とで規定される観測値ベクトルについて相関の可能性がある観測情報を抽出する観測値選択手段、観測値抽出手段によって抽出された観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成する仮説作成手段、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減する仮説縮小手段、仮説群内の各航跡に対して、当該目標のドップラ速度と目標位置とで規定される観測値ベクトルにて平滑化を実行する平滑化手段、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択する仮説選択手段としてコンピュータを機能させるものである。
【００７４】
【発明の実施の形態】
以下、この発明の実施の一形態を説明する。
実施の形態１．
図１はこの発明の実施の形態１による目標追尾装置の構成を示す図である。図において、１は実施の形態１による目標追尾装置であって、後述する各構成部を具現化するプログラムを実行するコンピュータ装置から構成される。２は空間中の目標の位置ベクトルとドップラ速度を観測して観測値を得るレーダとしての目標観測装置である。３はディスプレイ上に航跡を表示して追尾している目標の航跡などをユーザに提示する目標表示装置である。４ａは上述した従来の目標追尾装置１０１によるものと同様に動作するゲート算出部であって、既存のクラスタ群に含まれる各航跡に対して現時刻の目標の位置情報に関する観測値の存在期待領域となるゲートを算出する。４ｂは既存のクラスタ群に含まれる各航跡に対して現時刻の目標のドップラ速度計測値に関する観測値の存在期待領域となるゲートを算出するゲート算出部である。
【００７５】
５ａは上述した従来の目標追尾装置１０１によるものと同様に動作する観測値選択部（位置情報観測値抽出手段）であって、観測値選択部５ｂによって選択された現時刻における全ての目標の観測値のうちからゲート算出部４ａが算出した各航跡のゲートに含まれる観測値を選択する。５ｂは目標観測装置２から目標追尾装置１に入力された現時刻の全ての観測値のうち各航跡ごとにドップラ速度によるゲート内に含まれるドップラ速度計測値を有する観測値を選択する観測値選択部（ドップラ速度観測値抽出手段）である。６は目標追尾装置１内のクラスタの状態を示したシステム内クラスタ表を格納する記憶部である。
【００７６】
７はクラスタ新設・統合部（仮説作成手段）であって、観測値選択部５ａからの観測値とシステム内クラスタ表における既存のクラスタとの関係に基づいて既存のクラスタを統合し、また新しいクラスタを作成する。後述するように、航跡の推定では、空間内の全ての航跡を互いにゲートが重なり合う航跡群ごとに一括して相関の仮説を構成して処理する。このような仮説作成処理の単位となる航跡群をクラスタと呼んでいる。また、仮説とは、クラスタ内のいくつかの航跡を選択して構成される航跡の集合である。さらに、各仮説は、クラスタ内のいずれの航跡の組み合わせが正しいかを判断させるものである。
【００７７】
８はクラスタ新設・統合部７によって作成されるクラスタ内観測ベクトル表を格納する記憶部である。クラスタ内観測ベクトル表は、観測値選択部５ａから入力した観測値を各クラスタごとの観測値群として格納する。９はクラスタ内仮説状況データ群を格納する記憶部であって、クラスタ内仮説表、仮説内航跡表及びクラスタ内航跡−観測ベクトル表を格納する。９ａは記憶部９中でクラスタ内にある全ての仮説を示したクラスタ内仮説表として機能する記憶領域である。９ｂは記憶部９中で各仮説ごとに仮説内にある全ての航跡を示した仮説内航跡表として機能する記憶領域である。９ｃは記憶部９中でクラスタ内にある全ての航跡とこれらを構成する観測値との関係を示したクラスタ内航跡−観測ベクトル表として機能する記憶領域である。
【００７８】
１０はゲート内判定行列算出部（仮説作成手段）であって、クラスタ内観測ベクトル表及びクラスタ内仮説状況データ群からのデータを入力してクラスタ内ゲート内判定行列を算出する。１１はクラスタ内の現時刻の観測値と航跡との関係を示すクラスタ内ゲート内判定行列を格納する記憶部である。１２はクラスタ内ゲート内判定行列を入力してクラスタ内航跡相関行列を算出する航跡相関行列算出部（仮説作成手段）である。１３はクラスタ内で仮説の拡張可能性を示すクラスタ内航跡相関行列を格納する記憶部である。
【００７９】
１４は仮説更新部（仮説作成手段）で、前時刻までの観測値による仮説の状況とクラスタ内航跡相関行列から現時刻に入力した観測値とに対応する新たな仮説を作成する。１５は仮説縮小部（仮説縮小手段）であって、クラスタ内の各仮説の確からしさ（信頼度）の評価を行いその評価結果より一部の仮説を削除することで仮説数を縮小する。１６はクラスタ内に複数の仮説が存在する場合にその中から最善の仮説を１つ選択して目標の数とその航跡を決定する航跡決定部（仮説選択手段）である。
【００８０】
１７ａは上述した従来の目標追尾装置１０１の航跡平滑部と同様に動作する位置・速度平滑部（平滑化手段）であって、クラスタ内の各航跡に対しカルマンフィルタの平滑処理を施すことによって、位置や速度の平滑化を行う。１７ｂはドップラ速度平滑部（平滑化手段）であって、クラスタ内の各航跡に対してドップラ速度の平滑化を行う。１８はクラスタ分離部であって、仮説縮小後のクラスタを分離できるか否かを調べると共に、分離後のそれぞれのクラスタの仮説を再構成する。
【００８１】
次に概要について説明する。
この実施の形態１による目標追尾装置１では、従来の目標追尾装置１０１における位置情報に対するゲートに加え、ドップラ情報に対するゲートを設けている。これにより、各航跡に対して相関し得る観測値をより厳しく選別することができ、誤相関を減らすことができる。このため、従来の目標追尾装置１０１で実施されていた位置情報を使用した追尾フィルタ処理に加えて、以下に説明するようなドップラ速度情報を対象とした追尾フィルタ処理を別個に設ける。
【００８２】
先ず、ドップラ速度を追尾処理するため、上記式（３）とは異なる別の状態ベクトルｘ^Ｄ _ｋを下記式（２２）のように定義する。ここで、Ｄｒ_ｋは、目標のドップラ速度（距離の時間変化率）であり、Ｄｒ_ｋドットはドップラ速度Ｄｒ_ｋの一階の時間微分である。また、レーダの観測値に付随してくるドップラ速度の計測結果をＤｒ^ｏ _ｋとする。
【数１８】

【００８３】
さらに、目標のドップラ速度の時間的推移を表すドップラ速度の運動モデルを下記式（２３）で、また、ドップラ速度の観測モデル（ドップラ速度の計測結果Ｄｒ^ｏ _ｋ）を下記式（２４）で定義する。ここで、Φ^Ｄ _ｋは、上記式（２２）の状態ベクトルｘ^Ｄ _ｋに対する状態遷移行列である。Ｈ^Ｄ _ｋはドップラ速度の観測値に対する観測行列である。また、ｗ^Ｄ _ｋは上記式（２２）の状態ベクトルに対する駆動雑音ベクトルであって、平均ゼロ、共分散行列Ｑ^Ｄ _ｋの正規性白色雑音に従うものと仮定する。ｖ^Ｄ _ｋはスカラの観測雑音であり、平均ゼロ、分散Ｒ^Ｄ _ｋの正規性白色雑音に従うものと仮定する。
【数１９】

【００８４】
例えば、目標のドップラ速度の運動を時間に対して線形的であると仮定すれば、上記式（２３）における状態遷移行列は、下記式（２５）で定義することができる。このとき、駆動雑音ベクトルｗ^Ｄ _ｋは、ドップラ速度の運動を線形的とみなしたことによる打ち切り誤差項を表現したものと捉えることができる。また、観測行列Ｈ^Ｄ _ｋは、上記式（２２）の状態ベクトルｘ^Ｄ _ｋからドップラ速度の項を抽出するためのもので、下記式（２６）で定義することができる。
【数２０】

【００８５】
次に動作について説明する。
図２は図１中の目標追尾装置の動作を示すフロー図であり、この図に沿って説明する。
先ず、時刻ｔ_ｋに目標観測装置２から観測値を入力する（ステップＳＴ１）と、目標追尾装置１内のゲート算出部４ｂは、既存のクラスタ群に含まれる全ての航跡に対して、上記式（２２）の状態ベクトルｘ^Ｄ _ｋについての下記式（２７）、（２８）に示すカルマンフィルタの予測処理を施す（ステップＳＴ２）。ここで、ｘ^Ｄ _{ｋ−１｜ｋ−１}は、時刻ｔ_ｋ−１の状態ベクトルの平滑値である。ｘ^Ｄ _{ｋ｜ｋ−１}は、時刻ｔ_ｋに対する状態ベクトルの予測値を表している。また、Ｐ^Ｄ _{ｋ−１｜ｋ−１}、Ｐ^Ｄ _{ｋ｜ｋ−１}はそれぞれｘ^Ｄ _{ｋ−１｜ｋ−１}、ｘ^Ｄ _{ｋ｜ｋ−１}の誤差共分散行列である。
【数２１】

【００８６】
さらに、ゲート算出部４ｂは、下記式（２９）、（３０）に従って、各航跡の現時刻ｔ_ｋに対するドップラ速度予測値Ｄｒ_{ｋ｜ｋ−１}と、残差分散Ｓ^Ｄ _ｋを算出する。このＤｒ_{ｋ｜ｋ−１}はドップラ速度ゲートの中心値を表し、Ｓ^Ｄ _ｋはドップラ速度ゲートの広がりを特徴づける諸元である。
【数２２】

【００８７】
次に、観測値選択部５ｂは、ゲート算出部４ｂが算出した各航跡のドップラ速度ゲート内にいずれの観測値が存在するかを調べることにより、各航跡に相関し得る観測値の最初の選別を行う（ステップＳＴ３、ドップラ速度観測値抽出ステップ）。具体的には、各航跡に対して下記式（３１）を満たすドップラ速度の計測値を有する観測値を、その航跡に相関し得る観測値として選び出す。但し、ｄ_Ｄはドップラ速度ゲートのサイズを決定するパラメータである。
【数２３】

【００８８】
上述のようにしてドップラ速度に関する処理が完了すると、ゲート算出部４ａは、全ての航跡に対して、上記式（３）の位置と速度の状態ベクトルについてカルマンフィルタの予測処理を施すことにより、現時刻ｔ_ｋにおける観測値の位置情報に関する存在期待領域であるゲートを算出する（ステップＳＴ４、位置情報観測値抽出ステップ）。この処理は、従来の目標追尾装置１０１の場合と同様である。
【００８９】
続いて、位置情報による観測値選択部５ａは、ドップラ速度情報による観測値選択部５ｂが各航跡ごとに選択した観測値に各々対して、各航跡の位置情報に関するゲート内にも存在するか否かを調べることで、各航跡に相関し得る観測値を最終的に決定する（ステップＳＴ５）。なお、この位置情報による観測値選択部５ａによるゲート内判定の方法は、従来の目標追尾装置１０１の場合と同様である。
【００９０】
このように、この実施の形態１による目標追尾装置１では、各航跡に相関し得る観測値を選択する場合、ドップラ速度に関するゲート内に存在することと、位置情報に関するゲート内に存在することの両方の条件を課すことによって、従来の目標追尾方式の場合に比べてより厳しい選別を行っている。
【００９１】
以下、ステップＳＴ６からステップＳＴ１０までの、クラスタ新設・統合部７、ゲート内判定行列算出部１０、航跡相関行列算出部１２、仮説更新部１４、仮説縮小部１５による処理は、従来の目標追尾装置１０１による図９に示すステップＳＴ１０３からステップＳＴ１０７までの処理と同様である。また、ステップＳＴ１１における位置・速度平滑部１７ａによる処理は、従来の目標追尾装置１０１内の航跡平滑部１１６による図９に示すステップＳＴ１０８の処理と同様である。これらの処理ステップの説明は省略する。
【００９２】
仮説縮小処理が完了すると、ドップラ速度平滑部１７ｂは、カルマンフィルタの平滑処理式に従って、クラスタ内の各航跡に対して上記式（２２）のドップラ速度の状態ベクトルについての平滑処理を実行する（ステップＳＴ１２、平滑化ステップ）。具体的には、航跡Ｔ_ｔが現時刻ｔ_ｋの新たな観測値の追加によって伸張された場合、この観測値に付随するドップラ速度の計測結果Ｄｒ^ｏ _ｋを使用して、下記式（３２）、（３３）、（３４）に従って平滑値の更新を実行する。
【数２４】

【００９３】
ここで、ｘ^Ｄ _{ｋ｜ｋ−１}、Ｐ^Ｄ _{ｋ｜ｋ−１}は、上記式（２７）、（２８）で算出した航跡Ｔ_ｔの状態ベクトル予測値及び予測誤差共分散行列である。ｘ^Ｄ _ｋ｜ｋはこの航跡の時刻ｔ_ｋの状態ベクトル平滑値であり、Ｐ^Ｄ _ｋ｜ｋはその誤差共分散行列である。また、Ｋ^Ｄ _ｋはカルマンゲイン行列である。
【００９４】
以降のステップＳＴ１３からステップＳＴ１６までの処理は、従来の目標追尾装置１０１による図９に示すステップＳＴ１０９からステップＳＴ１１２までの処理と同様である。よって、説明を省略する。
【００９５】
以上のように、この実施の形態１によれば、レーダからの位置の観測情報に加え、ドップラ速度の観測情報を活用することを目的として、位置、速度の平滑化を行うための追尾フィルタ処理とドップラ速度の平滑化を行うための追尾フィルタ処理とを実施し、各航跡に対して位置のゲートのみならずドップラ速度のゲート判定を行うので、各航跡に相関し得る観測値が厳しく選別され、航跡に対する観測値の誤相関が減少して追尾性能を向上させることができる。また、各航跡に相関可能な観測値の数が減少するため、結果的に装置内に作成される仮説の数が減少し、装置全体の処理効率も向上させることができる。
【００９６】
実施の形態２．
図３はこの発明の実施の形態２による目標追尾装置における仮説縮小部の構成を示す図である。図において、１５ａは実施の形態２における仮説縮小部（仮説縮小手段）であって、後述するような従来の目標追尾装置１０１とは異なる機能を有し、上記実施の形態１の目標追尾装置に適用することができる。また、仮説縮小部１５ａは、実施の形態２による目標追尾装置として機能するコンピュータ装置に実行されるプログラムとして具現化することができる。１９は親仮説信頼度設定器（親仮説信頼度設定部）で、仮説尤度の初期値として親仮説の前時刻における信頼度を設定する。２０は航跡状態判定器（航跡状態判定部）であって、仮説内の各航跡に対する現時刻の観測値の割り当て状態を調べ、この結果に応じて探知確率乗算器２２、ゲート内非探知確率乗算器２４、新目標発生率乗算器２１のいずれかに尤度計算値を送出する。
【００９７】
２１は仮説尤度計算値に新目標発生率を乗算する新目標発生率乗算器（新目標発生率乗算部）で、２２は仮説尤度計算値に探知確率を乗算する探知確率乗算器（探知確率乗算部、尤度乗算ブロック部）である。また、２３ａは位置情報による尤度乗算器（尤度乗算部、尤度乗算ブロック部）であって、仮説尤度計算値に観測値の位置の尤度を乗算する。２３ｂはドップラ速度情報による尤度乗算器（尤度乗算部、尤度乗算ブロック部）で、仮説尤度計算値に観測値のドップラ速度の尤度を乗算する。２４は仮説尤度計算値にゲート内非探知確率を乗算するゲート内非探知確率乗算器（ゲート内非探知確率乗算部）である。
【００９８】
２５は観測値状態判定器（観測値状態判定部）であって、クラスタ内の各観測値に対する仮説内での扱われ方を調べ、観測値が誤信号と扱われている場合に限り、仮説尤度計算値を誤信号発生率乗算器２６に送る。２６は仮説尤度計算値に誤信号発生率を乗算する誤信号発生率乗算器（誤信号発生率乗算部）である。また、２７はクラスタ内の各仮説の尤度計算結果を記憶しておく仮説尤度記憶部である。２８はクラスタ内の全仮説についての仮説尤度の合計値を算出する仮説尤度集計器（仮説尤度集計部）である。２９は仮説信頼度算出器（仮説信頼度算出部）であって、クラスタ内の各仮説の仮説尤度を上記全仮説に対する合計値で除算することで各仮説の信頼度評価値を算出する。３０は仮説削減器（仮説削減部）で、仮説の信頼度評価値に従って信頼度の低い仮説を削除することによって仮説数を削減する。
【００９９】
次に概要について説明する。
この実施の形態２による仮説縮小部１５ａは、各子仮説の信頼度評価値を位置の観測情報のみでなく、ドップラ速度の観測情報を使用して算出することを特徴とする。具体的には、各子仮説Ｈ（ｉ）の尤度γ_ｉを、従来の目標追尾装置１０１の場合の上記式（１４）に代わって、下記式（３５）を用いて算出する。
【数２５】

【０１００】
ここで、上記式（３５）は、従来の目標追尾装置１０１の場合の上記式（１４）と異なって、右辺第２項の積演算にｇ^Ｄ _ｊ，ｔが追加されている。ｇ^Ｄ _ｊ，ｔは、航跡Ｔ_ｔが観測値ｚ_ｋ，ｊにより伸張された場合のドップラ速度情報による航跡の尤度である。また、このｇ^Ｄ _ｊ，ｔは、観測値のドップラ速度が航跡Ｔ_ｔの予測値Ｄｒ_{ｋ｜ｋ−１}を中心に分散Ｓ^Ｄ _ｋの１変量ガウス分布で発生するものと仮定して、下記式（３６）により計算する。但し、Ｄｒ^ｏ _ｋ，ｊは、観測値ｚ_ｋ，ｊに付随するドップラ速度の計測結果である。なお、各仮説の信頼度評価値は、従来の目標追尾装置１０１の場合と同様に、上記式（３５）による仮説尤度を上記式（１６）によって正規化したものとする。
【数２６】

【０１０１】
次に動作について説明する。
図４は図３中の仮説縮小部の動作を示すフロー図であり、この図に沿って実施形態２による仮説縮小処理を詳細に説明する。
先ず、親仮説信頼度設定器１９は、クラスタ内の仮説Ｈ（ｉ）を１つ選択して、この仮説の親仮説の信頼度β_ｐ（ｉ）を仮説尤度計算値γ_ｉの初期値として設定する（ステップＳＴ１ａ）。この仮説尤度計算値は、航跡状態判定器２０に送出される。
【０１０２】
次に、航跡状態判定器２０は、親仮説信頼度設定器１９が選択した仮説内の航跡を順次選択（ステップＳＴ２ａ）して、現時刻の観測値の割り当て状態を調べる（ステップＳＴ３ａ、ステップＳＴ５ａ）。即ち、選択した航跡が下記（１）から（３）までのいずれであるかを判断する。
【０１０３】
（１）現時刻に開始された新航跡
（２）前時刻の既存航跡に現時刻の新たな観測値が追加された航跡（伸張された航跡）
（３）前時刻の既存航跡に現時刻の観測値がないとして保持された航跡
【０１０４】
航跡状態判定器２０は、上述した判断結果が上記（１）の場合には新目標発生率乗算器２１に、上記（２）の場合には探知確率乗算器２２に、また、上記（３）の場合にはゲート内非探知確率乗算器２４に、仮説尤度計算値γ_ｉを送出する。
【０１０５】
ステップＳＴ３ａにて現時刻に開始された新航跡であると判断されると、新目標発生率乗算器２１は、航跡状態判定器２０から入力した仮説尤度計算値γ_ｉに、観測空間の単位体積当たりの新たな目標の発生数の平均値である新目標発生率β_ＮＴを乗算して、乗算結果を再び航跡状態判定器２０に返す（ステップＳＴ４ａ）。
【０１０６】
また、ステップＳＴ３ａにて現時刻に開始された新航跡でなく、ステップＳＴ５ａにて前時刻の既存航跡に現時刻の新たな観測値が追加された航跡（伸張された航跡）であると判断されると、探知確率乗算器２２は、航跡状態判定器２０から入力した仮説尤度計算値γ_ｉに探知確率Ｐ_Ｄを乗算し、乗算結果を位置情報による尤度乗算器２３ａに送る（ステップＳＴ６ａ）。
【０１０７】
位置情報による尤度乗算器２３ａでは、探知確率乗算器２２から入力した乗算結果値に、上記式（１５）で算出される位置情報による尤度ｇ_ｊ，ｔを乗算し、乗算結果をドップラ速度情報による尤度乗算器２３ｂに送る（ステップＳＴ７ａ）。さらに、ドップラ速度情報による尤度乗算器２３ｂは、尤度乗算器２３ｂから入力した乗算結果値に、上記式（３６）で算出されるドップラ速度情報による尤度ｇ^Ｄ _ｊ，ｔを乗算する（ステップＳＴ８ａ）。この乗算結果は、再び航跡状態判定器２０に送り返される。
【０１０８】
また、ステップＳＴ３ａにて現時刻に開始された新航跡でなく、ステップＳＴ５ａにて新たな観測値で伸張された航跡でなく、前時刻の既存航跡に現時刻の観測値がないものとして保持された航跡であると判断されると、ゲート内非探知確率乗算器２４は、航跡状態判定器２０から入力した仮説尤度計算値γ_ｉに対して、ゲート内に航跡の観測値が探知されない確率に相当する（１−Ｐ_ＤＰ_Ｇ）（ここでは、この値をゲート内非探知確率と呼ぶ）を乗算し、乗算結果を再び航跡状態判定器に送り返す（ステップＳＴ９ａ）。
【０１０９】
上述した一連の処理は、航跡状態判定器２０が仮説内の全ての航跡を選択し終わるまで、繰り返し実行される（ステップＳＴ１０ａ）。仮説内の全ての航跡について処理が終了すると、航跡状態判定器２０は、仮説尤度計算値を観測値状態判定器２５に送る。
【０１１０】
上記仮説尤度計算値を航跡状態判定器２０から入力すると、観測値状態判定器２５では、クラスタ内の各観測値を順次選択して（ステップＳＴ１１ａ）、着目している仮説内で誤信号として扱われているかどうかを調べる（ステップＳＴ１２ａ）。このとき、選択した観測値が誤信号と扱われていれば、仮説尤度計算値を誤信号発生率乗算器２６に送る。
【０１１１】
誤信号発生率乗算器２６では、観測値状態判定器２５から入力した仮説尤度計算値に誤信号発生率β_ＦＴを乗算し、乗算結果を再び観測値状態判定器２５に返す（ステップＳＴ１３ａ）。観測値状態判定器２５は、親仮説信頼度設定器１９が選択した仮説内の全ての観測値を選択し終わるまで、上記操作を繰り返し実行する（ステップＳＴ１４ａ）。
【０１１２】
また、親仮説信頼度設定器１９による初期値設定から観測値状態判定器２５による繰り返し処理までを終了すると、仮説尤度計算値には、着目した仮説に対する上記式（３５）による仮説尤度γ_ｉの計算結果が得られることになる。この計算結果は、仮説尤度記憶部２７に保存される。
【０１１３】
一方、仮説縮小部１５ａでは、上記仮説尤度を計算し、計算結果を仮説尤度記憶部２７に保存する一連の処理を、クラスタ内の全ての仮説について順次実行する（ステップＳＴ１５ａ）。ここで、クラスタ内の全ての仮説について仮説尤度の計算が終了すると、仮説尤度集計器２８は、仮説尤度記憶部２７に保存された前仮説の尤度を読み出して、その合計値を算出する（ステップＳＴ１６ａ）。この結果は、仮説信頼度算出器２９に送られる。仮説信頼度算出器２９は、上記式（１６）に従って、仮説尤度集計器２８から入力した合計値で仮説尤度記憶部２７から読み出した各仮説の尤度を除算して、各仮説の信頼度評価値β_ｉを計算する（ステップＳＴ１７ａ）。
【０１１４】
最後に、仮説削減器３０は、上述のようにして算出された仮説の信頼度評価値に従って、信頼度の低い仮説を削除するなどの方法によって仮説数を削減する。
【０１１５】
以上のように、この実施の形態２によれば、レーダからの位置の観測情報に加え、ドップラ速度の観測情報を使用してクラスタ内に作成された仮説の信頼度評価値を算出し、当該評価値を用いて仮説数を削減するので、不必要な仮説をより効率的に削除することができる。この結果、誤相関が減少して追尾性能を向上させることができる。また、各時刻で装置内に残る仮説数が減少するため、装置全体の処理効率も向上させることができる。
【０１１６】
実施の形態３．
図５はこの発明の実施の形態３による目標追尾装置の構成を示す図である。図において、１Ａは実施の形態３による目標追尾装置であって、後述する各構成部を具現化するプログラムを実行するコンピュータ装置から構成される。４ｃはゲート算出部であって、既存のクラスタ群に含まれる各航跡に対して、現時刻の位置及びドップラ速度の４次元の観測値ベクトルに対する存在期待領域を算出する。５ｃは位置・ドップラ速度の合成情報による観測値選択部（観測値抽出手段）で、目標観測装置２から目標追尾装置１Ａに入力された現時刻の観測値全体から、各航跡ごとに目標の位置及びドップラ速度に関するゲートに含まれる観測情報を有する観測値を選択する。１７ｃは位置・ドップラ速度の合成情報による航跡平滑部（平滑化手段）であって、クラスタ内の各航跡に対して位置及びドップラ速度の観測情報を使用して、位置、速度の状態ベクトルを更新する。なお、図１と同一構成要素には同一符号を付して重複する説明を省略する。
【０１１７】
次に概要について説明する。
この実施の形態３による目標追尾装置１Ａは、従来の目標追尾装置１０１における位置情報のみに対するゲートに代わり、位置及びドップラ速度の４次元の観測値ベクトルに対するゲートを設けている。これにより、各航跡に対して相関し得る観測値がより厳しく選別されて、誤相関を減らすことができる。このため、従来の目標追尾装置１０１で実施されていた位置情報のみを使用した追尾フィルタ処理に代わり、位置とドップラ速度の４次元の観測情報を使用した別の追尾フィルタ処理を以下のように構成する。
【０１１８】
先ず、従来の目標追尾装置１０１の場合と同様に、時刻ｔ_ｋにおける位置と速度により構成される追尾目標の状態ベクトルｘ_ｋを上記式（３）で定義する。また、追尾目標の状態ベクトルｘ_ｋの時間的推移を表す運動モデルを上記式（５）で定義する。上記式（５）の運動モデルの設定方法の一例は、上記実施の形態１で説明したように、例えば目標の運動を等速直線運動とみなして、上記式（７）で状態遷移行列Φ_ｋを定義する。
【０１１９】
一方、時刻ｔ_ｋのレーダによる観測値ベクトルは、下記式（３７）で定義する。ここで、ｘ^ｏ _ｋ，ｙ^ｏ _ｋ，ｚ^ｏ _ｋは、目標の観測位置座標を表している。また、Ｄｒ^ｏ _ｋは、ドップラ速度の計測結果を表している。下記式（３７）は、観測値ベクトル対して目標の観測位置とドップラ速度の計測結果とを含めている点が、従来の目標追尾装置１０１における上記式（６）と異なっている。
【数２７】

【０１２０】
また、レーダの観測モデルを下記式（３８）、（３９）、（４０）の非線形関数で定義する。但し、ｖ^Ｍ _ｋは、平均ゼロ、共分散行列Ｒ^Ｍ _ｋの４次元の正規性白色雑音に従う観測雑音ベクトルである。また、ｒ_ｋは目標の距離を表している。
【数２８】

【０１２１】
上述したような観測モデルを設定して、この実施の形態３では、非線形モデルに適用可能な拡張カルマンフィルタを用いて、追尾フィルタ処理を実行する。このとき、上記非線形関数ｈ（ｘ）の線形一次係数行列Ｈ^Ｍ（ｘ_ｋ）が必要となる。この行列は、下記式（４１）〜（４７）によって計算することができる。
【数２９】

【０１２２】
なお、線形一次係数値行列は、状態ベクトルの真値ｘ_ｋにおいて定義されるが、実際の追尾フィルタ処理では真値が得られないため、予測値ｘ_{ｋ｜ｋ−１}における値を使用する。
【０１２３】
次に動作について説明する。
図６は図５中の目標追尾装置の動作を示すフロー図であり、この図に沿って動作の詳細を説明する。
先ず、目標観測装置２から時刻ｔ_ｋにおける観測値を入力（ステップＳＴ１ｂ）すると、目標追尾装置１Ａ内のゲート算出部４ｃは、既存のクラスタ群に含まれる全ての航跡に対して、上記式（３）の状態ベクトルｘ_ｋについての予測処理を施す。ここで、目標の運動モデルは、従来の目標追尾装置１０１の場合と同様に定義している。このため、当該予測処理は、従来の場合と同様に上記式（９）、（１０）による通常のカルマンフィルタの予測処理式で実行される。
【０１２４】
続いて、ゲート算出部４ｃは、各航跡について、上記式（３７）の４次元の観測値ベクトルの現時刻ｔ_ｋに対する予測値ｚ^Ｍ _{ｋ｜ｋ−１}と残差共分散行列Ｓ^Ｍ _ｋを、下記式（４８）、（４９）に従って算出する（ステップＳＴ２ｂ）。但し、ｘ_{ｋ｜ｋ−１}、Ｐ_{ｋ｜ｋ−１}は、上記式（９）、（１０）による通常のカルマンフィルタの予測処理で算出した状態ベクトル予測値及びその誤差共分散行列である。また、観測値ベクトルの予測値ｚ^Ｍ _{ｋ｜ｋ−１}は、位置・ドップラ速度合成ゲートの中心を表し、残差共分散行列Ｓ^Ｍ _ｋはゲートの広がりを特徴づける諸元である。
【数３０】

【０１２５】
次に、位置・ドップラ速度の合成情報による観測値選択部５ｃは、各航跡の位置・ドップラ速度合成ゲート内にいずれの観測値が存在するかを調べることにより、各航跡に相関し得る観測値を決定する（ステップＳＴ３ｂ）。具体的には、各航跡に相関し得る観測値として、下記式（５０）の条件を満たす観測値を選び出す。但し、ｄ_Ｍは、位置・ドップラ速度合成ゲートのサイズを決定するパラメータである。
【数３１】

【０１２６】
以下、ステップＳＴ４ｂからステップＳＴ８ｂまでの、クラスタ新設・統合部７、ゲート内判定行列算出部１０、航跡相関行列算出部１２、仮説更新部１４、及び、仮説縮小部１５による処理は、従来の目標追尾装置１０１による図９に示すステップＳＴ１０３からステップＳＴ１０７までの処理と同様である。よって、説明を省略する。
【０１２７】
続いて、位置・ドップラ速度の合成情報による航跡平滑部１７ｃは、拡張カルマンフィルタの平滑処理式に従って、上記式（３７）の４次元の観測値ベクトルを使用して、状態ベクトル平滑値及びその誤差共分散行列の更新を行う（ステップＳＴ９ｂ）。具体的には、航跡が現時刻ｔ_ｋの新たな観測値の追加によって伸張された場合、この観測値によって下記式（５１）〜（５３）に従って平滑値の更新を行う。
【数３２】

【０１２８】
ここで、ｘ_ｋ｜ｋは、上記式（３）の状態ベクトルに対する時刻ｔ_ｋの平滑値であり、Ｐ_ｋ｜ｋは、ｘ_ｋ｜ｋの誤差共分散行列である。また、Ｋ^Ｍ _ｋは、拡張カルマンフィルタのゲイン行列である。
【０１２９】
以降、ステップＳＴ１０ｂからステップＳＴ１３ｂまでの処理は、従来の目標追尾装置１０１による図９に示すステップＳＴ１０９からステップＳＴ１１２までの処理と同様である。よって、説明を省略する。
【０１３０】
以上のように、この実施の形態３によれば、レーダからの位置の観測情報に加え、ドップラ速度の観測情報を活用することを目的として、位置及びドップラ速度で構成される４次元の観測値ベクトルを使用して、目標の状態ベクトル平滑値の更新を行なうための追尾フィルタ処理と、位置及びドップラ速度の４次元の観測値ベクトルに対してゲート判定とを行うので、各航跡に相関し得る観測値が厳しく選別され、航跡に対する観測値の誤相関が減少して追尾性能を向上させることができる。また、各航跡に相関可能な観測値の数が減少することから装置内に作成される仮説の数が減少して装置全体の処理効率も向上させることができる。
【０１３１】
また、４次元の観測値ベクトルを定義することにより、位置とドップラ速度との観測誤差や推定誤差の分布の相関を考慮したゲート形成が可能となり、処理負荷が増大するものの、上記実施の形態１の目標追尾装置１と比較してより高い精度の追尾処理を期待することができる。
【０１３２】
実施の形態４．
図７はこの発明の実施の形態４による目標追尾装置における仮説縮小部の構成を示す図である。図において、１５ｂは実施の形態４の仮説縮小部（仮説縮小手段）であって、後述するような従来の目標追尾装置１０１とは異なる機能を有し、上記実施の形態３の目標追尾装置に適用することができる。また、仮説縮小部１５ｂは、実施の形態３による目標追尾装置として機能するコンピュータ装置に実行されるプログラムとして具現化することができる。２３ｃは位置とドップラ速度の合成情報による尤度乗算器（尤度乗算部、尤度乗算ブロック部）であって、仮説尤度計算値に位置及びドップラ速度で構成される４次元の観測値ベクトルの尤度を乗算する。なお、図３と同一構成要素には同一符号を付して重複する説明を省略する。
【０１３３】
次に概要について説明する。
この実施の形態４による仮説縮小部１５ｂは、各子仮説の信頼度評価値を位置及びドップラ速度の観測情報を使用して算出することを特徴とする。具体的には、各子仮説Ｈ（ｉ）の尤度γ_ｉを、従来の目標追尾装置１０１の場合の上記式（１４）に代わって、下記式（５４）で計算する。
【数３３】

【０１３４】
ここで、上記式（５４）は、従来の目標追尾装置１０１の場合の上記式（１４）と異なって、右辺第２項の積演算におけるｇ_ｊ，ｔがｇ^Ｍ _ｊ，ｔに変更された部分が異なっている。このｇ^Ｍ _ｊ，ｔは、航跡Ｔ_ｔが観測値ｚ_ｋ，ｊにより伸張された場合の上記式（３７）の観測値ベクトルの尤度である。この観測値ベクトルが上記式（４８）の予測値ｚ^Ｍ _{ｋ｜ｋ−１}を中心に、上記式（４９）の残差共分散行列Ｓ^Ｍ _ｋの４変量ガウス分布で発生するものと仮定し、下記式（５５）を用いて算出する。但し、ｚ^Ｍ _ｋ，ｊは、観測値ｚ_ｋ，ｊに対応する上記式（３７）の観測値ベクトルである。なお、各仮説の信頼度評価値は、従来の目標追尾装置１０１の場合と同様に、上記式（５４）による仮説尤度を上記式（１６）によって正規化したものとする。
【数３４】

【０１３５】
次に動作について説明する。
ここでは、上述した実施の形態と異なる仮説縮小部１５ｂの動作を説明する。仮説縮小部１５ｂでは、上記実施の形態２における尤度乗算器２３ａ，２３ｂが、位置及びドップラ速度の合成情報による尤度乗算器２３ｃに置き換わっている。この尤度乗算器２３ｃは、探知確率乗算器２２から入力した仮説尤度計算値に上記式（５５）で算出される位置及びドップラ速度の４次元観測値ベクトルによる尤度ｇ^Ｍ _ｊ，ｔを乗算して、乗算結果を航跡状態判定器２０に返す。他の部位の動作手順は、上記実施の形態２の場合と同様であり、説明を省略する。
【０１３６】
以上のように、この実施の形態４によれば、位置及びドップラ速度の４次元観測値ベクトルによる尤度を用いてクラスタ内に作成された仮説の信頼度評価値を算出し、当該評価値を用いて仮説数を削減するので、位置とドップラ速度との観測誤差や推定誤差の分布の相関を考慮した尤度の計算が可能となり、処理負荷は増大するものの、上記実施の形態２の仮説縮小部１５ａと比較してより高い精度の追尾処理を期待することができる。
【０１３７】
【発明の効果】
以上のように、この発明によれば、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度について相関の可能性がある観測情報を抽出すると共に、現時刻における目標位置について相関の可能性がある観測情報を抽出し、当該抽出した観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成し、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減すると共に、仮説群内の各航跡に対して、当該目標のドップラ速度を含む運動諸元にて平滑化を実行し、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択するので、各航跡に相関し得る観測値が厳しく選別され、航跡に対する観測値の誤相関が減少して追尾性能を向上させることができるという効果がある。
【０１３８】
この発明によれば、仮説縮小手段が、仮説群中の各仮説の確からしさを評価するにあたり、前時刻までに求めた航跡が現時刻の目標の観測値により伸張された場合における当該目標の位置及びドップラ速度による航跡の尤度を求め、これら航跡の尤度を用いて前時刻までに求めた航跡から導かれる現時刻の仮説の尤度を評価結果として算出するので、不必要な仮説をより効率的に削除することができ、誤相関が減少して追尾性能を向上させることができるという効果がある。また、各時刻で装置内に残る仮説数が減少するため、装置全体の処理効率も向上させることができるという効果がある。
【０１３９】
この発明によれば、追尾すべき目標の前時刻までに求めた航跡に対して、現時刻における当該目標のドップラ速度と目標位置とで規定される観測値ベクトルについて相関の可能性がある観測情報を抽出し、当該抽出した観測情報を用いて、当該目標の各観測時点で推定される航跡の組み合わせを示す仮説群を作成し、仮説群内の各仮説に対して、当該目標の観測位置及びドップラ速度に関する確からしさの評価を行い、当該評価結果に基づいて仮説群中の仮説数を削減すると共に、仮説群内の各航跡に対して、当該目標のドップラ速度と目標位置とで規定される観測値ベクトルにて平滑化を実行し、仮説群から現時刻における目標の運動諸元を最も的確に表現する仮説を選択するので、各航跡に相関し得る観測値が厳しく選別され、航跡に対する観測値の誤相関が減少して追尾性能を向上させることができるという効果がある。
【０１４０】
この発明によれば、仮説縮小手段が、仮説群中の各仮説の確からしさを評価するにあたり、目標の位置ベクトルとドップラ速度とを合成した観測値ベクトルの尤度を求め、これら航跡の尤度を用いて前時刻までに求めた航跡から導かれる現時刻の仮説の尤度を評価結果として算出するので、位置とドップラ速度との観測誤差や推定誤差の分布の相関を考慮した尤度の計算が可能となり、処理負荷は増大するものの、より高い精度の追尾処理を期待することができるという効果がある。
【図面の簡単な説明】
【図１】この発明の実施の形態１による目標追尾装置の構成を示す図である。
【図２】図１中の目標追尾装置の動作を示すフロー図である。
【図３】この発明の実施の形態２による目標追尾装置における仮説縮小部の構成を示す図である。
【図４】図３中の仮説縮小部の動作を示すフロー図である。
【図５】この発明の実施の形態３による目標追尾装置の構成を示す図である。
【図６】図５中の目標追尾装置の動作を示すフロー図である。
【図７】この発明の実施の形態４による目標追尾装置における仮説縮小部の構成を示す図である。
【図８】従来の目標追尾装置の構成を示すブロック図である。
【図９】図８中の目標追尾装置の動作を示すフロー図である。
【符号の説明】
１，１Ａ　目標追尾装置、２　目標観測装置、３　目標表示装置、４ａ，４ｂ，４ｃ　ゲート算出部、５ａ　観測値選択部（位置情報観測値抽出手段）、５ｂ観測値選択部（ドップラ速度観測値抽出手段）、５ｃ　観測値選択部（観測値抽出手段）、６　記憶部（システム内クラスタ表）、７　クラスタ新設・統合部（仮説作成手段）、８　記憶部（クラスタ内観測ベクトル）、９　記憶部（クラスタ内仮説状況データ群）、９ａ　記憶領域（クラスタ内仮説表）、９ｂ　記憶領域（仮説内航跡表）、９ｃ　記憶領域（クラスタ内航跡−観測ベクトル表）、１０　ゲート内判定行列算出部（仮説作成手段）、１１　記憶部（クラスタ内ゲート内判定行列）、１２　航跡相関行列算出部（仮説作成手段）、１３　記憶部（クラスタ内航跡相関行列）、１４　仮説更新部（仮説作成手段）、１５，１５ａ，１５ｂ　仮説縮小部（仮説縮小手段）、１６　航跡決定部（仮説選択手段）、１７ａ　位置・速度平滑部（平滑化手段）、１７ｂ　ドップラ速度平滑部（平滑化手段）、１７ｃ　航跡平滑部（平滑化手段）、１８　クラスタ分離部、１９親仮説信頼度設定器（親仮説信頼度設定部）、２０　航跡状態判定器（航跡状態判定部）、２１　新目標発生率乗算器（新目標発生率乗算部）、２２　探知確率乗算器（探知確率乗算部、尤度乗算ブロック部）、２３ａ，２３ｂ，２３ｃ　尤度乗算器（尤度乗算部、尤度乗算ブロック部）、２４　ゲート内非探知確率乗算器（ゲート内非探知確率乗算部）、２５　観測値状態判定器（観測値状態判定部）、２６　誤信号発生率乗算器（誤信号発生率乗算部）、２７　仮説尤度記憶部、２８　仮説尤度集計器（仮説尤度集計部）、２９　仮説信頼度算出器（仮説信頼度算出部）、３０　仮説削減器（仮説削減部）。[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a target tracking device, a method, and a program for estimating tracks of a plurality of targets based on radar observation information.
[0002]
[Prior art]
For example, Japanese Patent Application Laid-Open No. 8-271617 (hereinafter referred to as Document 1) discloses a tracking device or a tracking method for estimating a wake of each target based on observation information of a position by a radar and calculating a movement parameter of the target. Call) or D. B. Reid, "AnAlgorithm for Tracking, Multiple Targets", IEEE Transactions on Automatic Control, AC-24, pp. 843-845, @ 1979. (Hereinafter referred to as Document 2).
[0003]
In these documents, the position of the current time is predicted using a tracking filter for each existing wake obtained up to the previous time. Further, an expected existence area (hereinafter, this area is referred to as a gate) for the observed value of the current time of each track is calculated, and correlation processing between the observed value acquired in the gate and each track is performed. Thus, the target track at the current time is estimated.
[0004]
Here, in practice, an erroneous signal other than the target may be obtained from the radar, or conversely, the radar may fail to observe the target and may not obtain the target observation value. Also, if multiple targets are densely packed in a narrow area, observations of other tracked targets can be obtained in the gate of one track, or observations of new targets that have not been detected so far. In some cases.
[0005]
Even in such a situation, in order to correctly track a plurality of targets, in the above-mentioned document, a plurality of hypotheses are established regarding the correlation between each observation value and which track, and the tracking process is continued for each hypothesis. However, a postponement-decision type system is used in which a high probability is selected after the next time.
[0006]
For example, FIG. 8 is a block diagram showing the configuration of a conventional target tracking device in which the function of the tracking filter processing shown in Document 2 is explicitly added to the configuration of the target tracking device shown in Document 1. In the figure, reference numeral 100 denotes a target observation device, which observes a position vector of a target in space using a sensor such as a radar, obtains an observation value, and sends it to the target tracking device 101. A target tracking device 101 tracks a target based on an observation value obtained from the target observation device 100. Reference numeral 102 denotes a target display device, which displays a wake on a display and presents the state of the target being tracked to the user.
[0007]
A gate calculation unit 103 calculates a gate at the current time for each track included in each existing cluster. Here, one time-series data obtained by selecting at most one observation value from each sampling time (in the case of zero, it is assumed that no observation value of the wake was detected at that time). Are called wakes. Generally, multiple tracks are calculated for one target.
[0008]
An observation value selection unit 104 selects an observation value included in the gate of each track calculated by the gate calculation unit 103 from all target observation values at the current time acquired by the target observation device 100. Reference numeral 105 denotes a storage unit that stores an in-system cluster table indicating the state of clusters in the target tracking device 101.
[0009]
Reference numeral 106 denotes a cluster new establishment / integration unit that integrates existing clusters based on the relationship between the observation values from the observation value selection unit 104 and the existing clusters in the in-system cluster table, and creates a new cluster. As will be described later, in the trajectory estimation, all trajectories in the space are collectively constructed and processed for each trajectory group whose gates overlap each other. A wake group that is a unit of such a hypothesis creation process is called a cluster. The hypothesis is a set of tracks formed by selecting some tracks in the cluster. Further, each hypothesis is to determine which combination of tracks in the cluster is correct.
[0010]
A storage unit 107 stores an intra-cluster observation vector table created by the cluster new / integration unit 106. The intra-cluster observation vector table stores the observation values input from the observation value selection unit 104 as an observation value group for each cluster.
[0011]
Reference numeral 108 denotes a storage unit that stores a group of hypothesis situation data in a cluster, and stores a hypothesis table in a cluster, a trajectory table in a hypothesis and a trajectory-observation vector table in a cluster. Reference numeral 108a denotes a storage area that functions as an intra-cluster hypothesis table showing all the hypotheses in the cluster in the storage unit 108. Reference numeral 108b denotes a storage area which functions as a hypothesis track table indicating all tracks within the hypothesis for each hypothesis in the storage unit 108. Reference numeral 108c denotes a storage area that functions as a track-intracluster-observation vector table that indicates a relationship between all tracks in the cluster and observation values constituting the tracks in the storage unit 108.
[0012]
Reference numeral 109 denotes an intra-gate decision matrix calculation unit which receives an intra-cluster observation vector table and data from the intra-cluster hypothesis situation data group and calculates an intra-cluster gate intra-gate decision matrix. Reference numeral 110 denotes a storage unit for storing an intra-cluster gate in-gate determination matrix indicating the relationship between the observed value at the current time in the cluster and the wake. Reference numeral 111 denotes a wake correlation matrix calculation unit that inputs an intra-cluster gate determination matrix and calculates an intra-cluster wake correlation matrix. Reference numeral 112 denotes a storage unit that stores an intra-cluster wake correlation matrix indicating the possibility of expanding a hypothesis in a cluster.
[0013]
A hypothesis updating unit 113 creates a new hypothesis corresponding to the observation value input at the current time from the state of the hypothesis based on the observation value up to the previous time and the intra-cluster wake correlation matrix. Reference numeral 114 denotes a hypothesis reduction unit that evaluates the likelihood (reliability) of each hypothesis in the cluster, and reduces the number of hypotheses by deleting some of the hypotheses from the evaluation result. Reference numeral 115 denotes a track determining unit that selects one of the best hypotheses from among a plurality of hypotheses in the cluster and determines the number of targets and the track of the target.
[0014]
Reference numeral 116 denotes a track smoothing unit that performs a Kalman filter smoothing process on each track in the cluster to smooth the position and speed. Reference numeral 117 denotes a cluster separation unit that checks whether clusters after hypothesis reduction can be separated and, if separation is possible, reconstructs a hypothesis of each cluster after separation.
[0015]
Next, the operation will be described.
FIG. 9 is a flowchart showing the operation of the target tracking device in FIG. 8, and description will be made along this diagram.
First, time t_kWhen the observation value is input from the target observation device 100 to the target tracking device 100 (step ST100), the gate calculation unit 103 in the target tracking device 101 performs a Kalman filter prediction process on all the tracks included in the existing cluster group, and t_kIs calculated (step ST101). Thereafter, the observation value selection unit 104 checks which observation value exists in the gate of each track, and determines an observation value that can be correlated with each track (step ST102).
[0016]
Next, the cluster new / integration unit 106 newly establishes a new cluster composed of isolated observation values that cannot be correlated with the wakes in any of the clusters and defines the new cluster in the in-system cluster table. If a certain observation value exists in the gates of a plurality of tracks included in different clusters, these clusters are integrated as one cluster and redefined in the in-system cluster table (step ST103). At this time, the observation value selection unit 104 determines, for each of the clusters defined in the cluster table in the system, the entirety of the observation values that can be correlated with any of the wakes in this cluster. Write to value table.
[0017]
Subsequently, the target tracking device 101 executes a series of processes from step ST104 to step ST107 in order to update the state of each cluster defined in the in-system cluster table. First, the in-gate determination matrix calculation unit 109 calculates the current time t in the cluster._kThen, an intra-cluster in-gate decision matrix is calculated which indicates whether or not all the observation values can be correlated with the wake (step ST104).
[0018]
Here, the intra-cluster gate in-gate decision matrix is represented by the current time t_kThe number of observations in a cluster at m_k, Current time t_kExisting track in the cluster at_k-1The number of tracks in the cluster that has already been created by N_k-1(N_k-1= 0), the current time t_kNumber of tracks in cluster at N_k= N_k-1+ M_kIn this case, it is defined as the following equation (1).
(Equation 5)

[0019]
However, each row of the intra-cluster gate in-gate decision matrix represented by the above equation (1) is the current time t_kObservation z_{k, j}(J = 1, 2,..., M_k). Further, each column indicates the current time t._kTrack T in the cluster at_t(T = 0, 1, 2,..., N_k). Each element of the matrix indicates whether the observation corresponding to the row is within the wake gate corresponding to the column. Specifically, it is set as follows.
[0020]
First, the column of t = 0 shows a case where the observed value is a value due to an erroneous signal. Assuming that in fact all observations could be false signals, ω_{j, 0}= 1.
[0021]
Next, 1 ≦ t ≦ N_k-1Are the existing tracks T in the cluster._tIt corresponds to. Where the observed value z_{k, j}(J = 1, 2,..., M_k) Is the existing track T_t(T = 0, 1, 2,..., N_k) When included in the gate_{j, t}= 1. Also, the observed value z_{k, j}But the existing wake T_tIf not included in the gate of_{j, t}= 0.
[0022]
Furthermore, N_k-1+ 1 ≦ t ≦ N_kIs the current time t_kCorresponds to the track in the cluster that has newly appeared. These columns represent the possibility that all observations are each newly discovered targets. Here, the elements are set so that one observation value corresponds to one new track. That is, j = t−N_k-1In the case of ω_{j, t}= 1. Conversely, j ≠ t−N_k-1In the case of ω_{j, t}= 0.
[0023]
Subsequently, wake correlation matrix calculation section 111 calculates all intra-cluster wake correlation matrices from the intra-cluster gate-in-gate determination matrix (step ST105). Here, the intra-cluster wake correlation matrix indicates the correlation between the observed value in the cluster and the wake that can be actually assumed as a hypothesis. In general, a plurality of intra-cluster wake correlation matrices are generated from one intra-cluster in-gate decision matrix. Current time t_k, S in-cluster wake correlation matrices of s = 1, 2,..., S are defined from the intra-cluster gate in-gate decision matrix as shown in the following equation (2).
(Equation 6)

[0024]
Here, each row of the intra-cluster wake correlation matrix represents the current time t_k, And each column corresponds to a track in the cluster. Each element of the matrix indicates whether or not the observation value corresponding to the row is correlated with the track corresponding to the column. Specifically, it is set as follows.
[0025]
First, the observed value z_{k, j}(J = 1, 2,..., M_k) Is the wake T_t(T = 0, 1, 2,..., N_k) Is correlated with ω_{j, t} ^s= 1. Also, the observed value z_{k, j}Is the wake T_tIs not correlated with_{j, t} ^s= 0.
[0026]
Actually, when the wake correlation matrix calculation section 111 creates the intra-cluster wake correlation matrix from the intra-cluster gate judgment matrix, all combinations that simultaneously satisfy the following three criteria (A) to (C) are used. Are calculated as separate intra-cluster wake correlation matrices.
[0027]
(A) In the intra-cluster gate decision matrix, only the element of the intra-cluster wake correlation matrix corresponding to the element of 1 can be set to 1 and the other elements are set to 0.
(A) In all columns other than the column of t = 0 of the intra-cluster wake correlation matrix, only one element at most is set to 1 and the other elements are set to 0.
(C) In every row of the intra-cluster wake correlation matrix, one element is always 1 and the other elements are 0.
[0028]
Next, in order to update the hypothesis in the cluster, the hypothesis updating unit 113 updates the existing hypothesis in the cluster, that is, the previous time t._k-1A wake correlation matrix that can be combined with the hypothesis created in (1) is selected. Then, from the combination of the two, the current time t_kCreate a new hypothesis corresponding to. By repeatedly performing this operation, the hypothesis in the cluster is updated (step ST106).
[0029]
Here, an existing hypothesis selected from within a cluster may be called a parent hypothesis, and a new hypothesis may be called a child hypothesis. Actually, if an existing track that is considered to be correlated with the observed value in the track correlation matrix is not included in the parent hypothesis, it is assumed that the two cannot be combined. Hereinafter, the detailed operation in step ST106 will be described.
[0030]
First, the hypothesis updating unit 113 selects one existing hypothesis in the cluster as a parent hypothesis, and selects one of the intra-cluster wake correlation matrices in the storage unit 112 to derive a group of child hypotheses derived from the parent hypothesis. Select Thereafter, the hypothesis updating unit 113 determines whether the intra-cluster wake correlation matrix can be combined with the parent hypothesis.
[0031]
More specifically, in the track correlation matrix within the cluster selected as described above, if an existing track having a correlation with any observation value is not included in the parent hypothesis, the two cannot be combined. And On the other hand, it is determined that all other cases can be combined.
[0032]
Next, only when it is determined that the combination can be performed as described above, the existing track in the existing hypothesis is extended by adding a new observation value correlated with the existing track, and a new track is added. Create child hypotheses by adding wakes to hypotheses and treating certain observations as false signals. The above process is repeatedly executed until all the intra-cluster wake correlation matrices calculated in step ST105 have been selected.
[0033]
Thereafter, the hypothesis updating unit 113 repeats the above-described processing while determining whether or not the above-described series of hypothesis updating processing is completed, until there is no remaining hypothesis in the cluster.
[0034]
In many cases, one existing hypothesis is combined with a plurality of intra-cluster wake correlation matrices to create a plurality of child hypotheses. Therefore, the number of hypotheses in the cluster increases by the hypothesis update processing in step ST106.
[0035]
Next, the hypothesis reducing unit 114 calculates the current time t created in step ST106._kIs calculated, and the number of hypotheses is reduced by a method such as deleting the others with high reliability among all the hypotheses (step ST107). A specific method of calculating the reliability evaluation value will be described later.
[0036]
Subsequently, the track smoothing unit 116 calculates the current time t created in the cluster._kBy performing a Kalman filter smoothing process on each of the wakes, the position and speed are smoothed (step ST108). Specifically, the current time t_kFor a newly started track, an initial value of a state vector composed of position and velocity information and its error covariance matrix is set.
[0037]
Next, the current time t_kFor the wake to which the new observation value has been added, the state vector and the smoothed value of the error covariance matrix are updated using this observation value. After this, the current time t_kA so-called memory tracking process of replacing the predicted value of the state vector and the error covariance matrix with a smoothed value is performed on the wake determined to have no observation value corresponding to.
[0038]
Subsequently, the target tracking device 101 determines whether or not the processing from step ST104 to step ST108 has been executed for all clusters defined in the in-system cluster table (step ST109). Here, when the processing has been completed for all clusters, the processing shifts to the processing of step ST110.
[0039]
In step ST110, the cluster separation unit 117 checks the configuration of the observation value of each track in the cluster, and determines “the current time t_kThe wakes that share at least one observation value up to belong to the same cluster ”are evaluated in order to determine whether the wake groups in the cluster can be separated into a plurality of clusters. If it can be separated, cluster separation and hypothesis reconstruction are performed, and the intra-cluster hypothesis status data group is updated. If it is determined by this cluster separation processing that a certain track only forms a single cluster with only a single track since its start, this track is determined to be established.
[0040]
Thereafter, the track determination unit 115 determines the current time t in the cluster._kThe number of targets and wakes are determined by selecting one best hypothesis from among the hypotheses (step ST111). These wake groups are displayed on the display of the target display device 102.
[0041]
The above-described series of processing is repeatedly executed until tracking end is detected by the target tracking device 101 (step ST112).
[0042]
Next, a description will be given of a tracking filter process in the gate calculation unit 103 and the wake smoothing unit 116, a gate determination process by the observation value selection unit 104, and a hypothesis reliability evaluation value calculation process by the hypothesis reduction unit 114 according to Literature 2. Tracking filter processing, which is the center of these processes, is performed as follows using a so-called Kalman filter.
[0043]
First, time t_kTracking target state vector x composed of position and velocity at_kAnd an observation vector z composed of position information_kIs defined by the following equations (3) and (4), respectively. Here, T is a symbol representing the transposition of a matrix. Also, x_k, Y_k, Z_kRepresents the true position coordinate value of the target, and vx_k, Vy_k, Vz_kRepresents the true value of the target speed component. Furthermore, x^o _k, Y^o _k, Z^o _kRepresents a target observation position coordinate, and is a value obtained by converting a result of observation in a polar coordinate system of a distance, an elevation angle, and an azimuth angle by a radar into a north reference rectangular coordinate system.
[0044]
(Equation 7)

[0045]
Here, a motion model representing a temporal transition of the state vector of the tracking target is defined by the following equation (5). The radar observation model is defined by the following equation (6). Note that Φ_kIs the state transition matrix, H_kIs the observation matrix. Also, w_kIs the mean zero, the covariance matrix Q_kThe driving noise vector according to the normality white noise of_kIs mean zero, covariance matrix R_kIs an observation noise vector according to the normal white noise of
(Equation 8)

[0046]
In the motion model equation of the above equation (5), if the target motion is modeled as a constant velocity linear motion, for example, the state transition matrix can be defined by the following equation (7). Where I₃, O₃Are the unit matrix and the zero matrix of 3 rows and 3 columns, respectively. At this time, the driving noise vector w_kCan be regarded as expressing the terminating error term caused by regarding the target motion as a constant velocity straight line. Further, the observation matrix H in the observation model equation of the above equation (6)_kIs a matrix for extracting position information from the state vector, and is defined by the following equation (8).
(Equation 9)

[0047]
Based on the above definition, the gate calculation unit 103 performs a Kalman filter prediction process shown in the following equations (9) and (10) for each track in the cluster. Where x_{k-1 | k-1}Is the time t_k-1Represents a smoothed value of the state vector. Also, x_{k | k-1}Is the time t_kRepresents the predicted value of the state vector. Furthermore, P_{k-1 | k-1}, P_{k | k-1}Is the smoothed value x_{k-1 | k-1}, Predicted value x_{k | k-1}Represents an error covariance matrix.
(Equation 10)

[0048]
Subsequently, the gate calculation unit 103 calculates the current time t of the wake according to the following equations (11) and (12)._kPredicted position z for_{k | k-1}And the residual covariance matrix S_kIs calculated. Here, the predicted position z_{k | k-1}Corresponds to the center position of the gate, and the residual covariance matrix S_kAre specifications characterizing the spread of the gate region.
[Equation 11]

[0049]
On the other hand, the observation value selection unit 104 calculates the predicted position z calculated as described above._{k | k-1}, The residual covariance matrix S_k, The observed value z that satisfies the following equation (13):_kIs determined to be the observation value in the gate of the wake. Here, d is a parameter for determining the gate size.
(Equation 12)

[0050]
Next, a method of calculating the reliability evaluation value of each child hypothesis by the hypothesis reduction unit 114 will be described. This evaluation value is calculated by the following procedure. First, the likelihood γ of each child hypothesis H (i)_iIs calculated using the following equation (14).
(Equation 13)

[0051]
Where β_{p (i)}Is the reliability evaluation value at the time before the parent hypothesis that generated the child hypothesis H (i). N_DTIs the number of tracks stretched by any observation in the child hypothesis, and N_MTIndicates the number of existing tracks that did not correspond to any observations in the child hypothesis. N_NTIs the number of newly started tracks in the child hypothesis, N_FTRepresents the number of observations treated as false signals in the child hypothesis. Also, P_DIs the radar detection probability. P_GIs the probability that an observation will be captured in the gate, and β_NTIs the average value of the number of new target generations per unit volume. β_FTIs the average value of the number of erroneous signals generated per unit volume. These are parameters.
[0052]
g_{j, t}Is the wake T_tIs the observed value z_{k, j}Is the likelihood of the wake when it is expanded by This likelihood g_{j, t}Is the wake T_tPredicted position z_{k | k-1}With covariance matrix S_kIs calculated according to the following equation (15), assuming that this occurs in a three-variable Gaussian distribution.
[Equation 14]

[0053]
The hypothesis reduction unit 114 calculates the hypothesis likelihood γ_iIs calculated for all the child hypotheses, and as shown in the following equation (16), those values are normalized over all the hypotheses to obtain the current time t of each child hypothesis._kReliability evaluation value β_iAnd
(Equation 15)

[0054]
Next, the wake smoothing unit 116 performs the following calculation according to the Kalman filter smoothing processing equation. That is, the wake T_tTo the observed value z_{k, j}Is added and expanded, the smoothed value is updated according to the following equations (17), (18), and (19).
(Equation 16)

[0055]
Where x_{k | k-1}, P_{k | k-1}Is the wake T calculated by the above equations (9) and (10), respectively._tIs a state vector prediction value and a prediction error covariance matrix. Also, x_{k | k}Is the current time t_kWake T corresponding to_tState vector smoothed value of P_{k | k}Is x_{k | k}Is the error covariance matrix of Note that K_kIs a Kalman gain matrix. On the other hand, wake T_tIf there is no observation value at the current time corresponding to the above, the memory tracking processing shown in the following equations (20) and (21) is executed.
[Equation 17]

[0056]
[Problems to be solved by the invention]
Since the conventional target tracking device is configured as described above, it estimates the wake only based on the observation information of the position from the radar and calculates the target motion data, so false alarms and clutter In an observation environment where many erroneous signals are generated, there is a problem that the tracking performance is limited even if the correlation method is of the postponing type.
[0057]
SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and uses a correlation method of a postponement determination type using not only information of a position obtained from a radar but also information of a Doppler velocity (a time rate of change of a distance). It is an object of the present invention to obtain a target tracking device, a method, and a program for executing a tracking process using a target.
[0058]
[Means for Solving the Problems]
A target tracking device according to the present invention is a Doppler velocity observation value extraction for extracting observation information that may have a correlation with the Doppler velocity of the target at the current time with respect to the track obtained before the time before the target to be tracked. Means, position information observation value extracting means for extracting observation information that may have a correlation with the target position at the current time with respect to the track obtained by the time before the target to be tracked, and both observation value extracting means Using the extracted observation information, a hypothesis creating means for creating a hypothesis group indicating a combination of wakes estimated at each observation point of the target, and, for each hypothesis in the hypothesis group, an observation position of the target and A hypothesis reduction unit that evaluates the likelihood of the Doppler velocity and reduces the number of hypotheses in the hypothesis group based on the evaluation result. Those comprising a smoothing means for performing smoothing in motion specification including degrees, and a hypothesis selecting means for selecting the hypothesis that most accurately represent the goals of the exercise specifications at the current time from the hypotheses.
[0059]
In the target tracking device according to the present invention, when the hypothesis reduction unit evaluates the likelihood of each hypothesis in the hypothesis group, the hypothesis reduction unit determines whether the wake obtained by the previous time is extended by the target observation value at the current time. The likelihood of the wake based on the target position and the Doppler velocity is obtained, and the likelihood of the hypothesis at the current time derived from the wake obtained by the previous time is calculated using the likelihood of the wake as an evaluation result.
[0060]
In the target tracking apparatus according to the present invention, the hypothesis reduction means may determine the wake T_t(T is the number of tracks in the hypothesis) is the current time t_kTarget observations z_{k, j}(J is the number of observations in the hypothesis), and the likelihood of the wake based on the position information of the target is g_{j, t}, The likelihood of the wake by the Doppler velocity of the target is g^D _{j, t}, The track T obtained by the time before deriving the hypothesis of the current time_tIs the reliability evaluation value of the hypothesis consisting of_{p (i)}, The detection probability for the target is P_DThe track T obtained by the previous time_tAnd the probability that the observation value at the current time is extracted as observation value information that may be correlated with the target position and the Doppler velocity is P_G, The average value of the number of new targets generated per unit volume of observation space is β_NT, The average number of erroneous signals generated per unit volume of the observation space is β_FT, Observation z_{k, j}Is the measured value of the Doppler velocity associated with^o _{k, j}, The number of tracks extended by any of the observations within the hypothesis at the current time_DT, The number of existing wakes determined not to correspond to any observation in the hypothesis at the current time_MT, The number of tracks newly started within the hypothesis at the current time, N_NT, The number of observations treated as false signals in the hypothesis at the current time is N_FTAnd the observed value is the wake T_tPredicted position z_{k | k-1}With covariance matrix S_kGenerated by a three-variable Gaussian distribution of_tDoppler velocity predicted value Dr_{k | k-1}With the residual covariance matrix S^D _kTrajectory T obtained up to the previous time, assuming that it occurs in a univariate Gaussian distribution of_tLikelihood γ of the current time hypothesis derived from_iIs calculated from the above equation.
[0061]
In the target tracking device according to the present invention, the hypothesis reducing unit multiplies the likelihood of the hypothesis obtained up to the previous time by the detection probability of the target, and the multiplication result of the detection probability multiplication unit. A position information likelihood multiplication unit that multiplies the likelihood of the target position in the target observation value, and a Doppler velocity likelihood multiplication that multiplies the multiplication result of the position information likelihood multiplication unit by the likelihood of the Doppler velocity in the target observation value A new likelihood multiplication block unit having a new target occurrence rate, which is the average value of the number of new targets generated per unit volume of the observation space, to the likelihood of the hypothesis obtained up to the previous time. The target occurrence rate multiplying unit and the likelihood of the hypothesis obtained up to the previous time indicate the degree to which observation values correlated with the wakes in the hypothesis are not extracted as observation value information that may be correlated with respect to position and Doppler velocity. Show A non-gate detection probability multiplication unit in the gate that multiplies the non-detection probability in the gate, and a detection probability multiplication unit in the gate according to the assignment state of the observation value at the current time to each track in the hypothesis obtained up to the previous time. A wake-state determination unit that sends the likelihood of the above-mentioned hypothesis to one of the detection probability multiplication unit and the new target incidence multiplication unit; and the likelihood of the hypothesis obtained up to the previous time, the per unit volume of the observation space. A false signal rate multiplying unit that multiplies the false signal rate, which is the average value of false signal occurrences, and, if the observed value is treated as a false signal in the hypothesis determined by the previous time, obtains by the previous time An observation value state determination unit that sends the likelihood of the obtained hypothesis to the error signal occurrence rate multiplying unit, and a total value of likelihoods of all the hypotheses constituting the hypothesis group obtained by the previous time by the hypothesis creation unit is calculated. The hypothesis likelihood aggregation unit and the total value calculated by the hypothesis likelihood aggregation unit A hypothesis reliability calculator that divides the likelihood of the hypothesis and calculates the division result as a reliability evaluation value of each hypothesis; and a hypothesis group based on the reliability evaluation value of the hypothesis calculated by the hypothesis reliability calculator. And a hypothesis reduction unit for reducing the number of hypotheses.
[0062]
The target tracking device according to the present invention has a possibility that, with respect to the track obtained by the time before the target to be tracked, the correlation between the observed value vector defined by the Doppler velocity and the target position of the target at the current time is determined. An observation value selection unit that extracts certain observation information, and a hypothesis creation unit that creates a hypothesis group indicating a combination of wakes estimated at each observation point of the target using the observation information extracted by the observation value extraction unit. A hypothesis reducing means for evaluating the likelihood of the target observation position and Doppler velocity for each hypothesis in the hypothesis group, and reducing the number of hypotheses in the hypothesis group based on the evaluation result; For each wake, a smoothing means for performing smoothing with an observation value vector defined by the Doppler velocity of the target and the target position, and the motion parameters of the target at the current time from the hypothesis group Also is intended and a hypothesis selecting means for selecting a hypothesis to express accurately.
[0063]
In the target tracking device according to the present invention, when the hypothesis reducing unit evaluates the likelihood of each hypothesis in the hypothesis group, the likelihood of an observation value vector obtained by combining the target position vector and the Doppler velocity is determined. The likelihood of the hypothesis at the current time derived from the track obtained up to the previous time using the likelihood of the previous time is calculated as an evaluation result.
[0064]
In the target tracking apparatus according to the present invention, the hypothesis reduction means may determine the wake T_t(T is the number of tracks in the hypothesis) is the current time t_kTarget observations z_{k, j}(J is the number of observations in the hypothesis), the observation value vector z composed of the observation position of the target and the measured value of the Doppler velocity when the target is expanded^M _{k, j}The likelihood of g^M _{j, t}, The track T obtained by the time before deriving the hypothesis of the current time_tIs the reliability evaluation value of the hypothesis consisting of_{p (i)}, The detection probability for the target is P_DThe track T obtained by the previous time_tAnd the probability that the observation value at the current time is extracted as observation value information that may have a correlation with respect to the position and the Doppler velocity is P_G, The average value of the number of new targets generated per unit volume of observation space is β_NT, The average number of erroneous signals generated per unit volume of the observation space is β_FT, The number of tracks extended by any of the observations within the hypothesis at the current time_DT, The number of existing wakes determined not to correspond to any observation in the hypothesis at the current time_MT, The number of tracks newly started within the hypothesis at the current time, N_NT, The number of observations treated as false signals in the hypothesis at the current time is N_FTAnd the observation vector is the wake T_tPredicted position z_{k | k-1}With the residual covariance matrix S^M _kTrack T obtained by the previous time, assuming that the_tLikelihood γ of the current time hypothesis derived from_iIs calculated from the above equation.
[0065]
In the target tracking device according to the present invention, the hypothesis reducing unit multiplies the likelihood of the hypothesis obtained up to the previous time by the detection probability of the target, and the multiplication result of the detection probability multiplication unit. A likelihood multiplication block unit having a likelihood multiplication unit that multiplies the likelihood of the observation value vector defined by the target position and the Doppler velocity in the target observation value, and the likelihood of the hypothesis obtained up to the previous time. , A new target incidence rate multiplication unit that multiplies the new target incidence rate, which is the average value of the number of new targets generated per unit volume of the observation space, and the likelihood of the hypothesis obtained up to the previous time, A non-gate detection probability multiplying unit that multiplies the non-detection probability in the gate indicating the degree to which observation values correlated with the wake of the observation are not extracted as observation value information that may be correlated, and each of the hypotheses obtained by the previous time View of the current time for the wake In accordance with the value assignment state, a wake-up state determination unit that sends the likelihood of a hypothesis to any one of the detection probability multiplication unit, the in-gate non-detection probability multiplication unit, and the new target occurrence rate multiplication unit, An error signal rate multiplying unit that multiplies the likelihood of the determined hypothesis by the error signal occurrence rate, which is the average value of the number of error signals generated per unit volume of the observation space, and the observed value within the hypothesis obtained up to the previous time. Is treated as an erroneous signal, the observed value state determination unit that sends the likelihood of the hypothesis obtained up to the previous time to the erroneous signal occurrence rate multiplication unit, and the hypothesis group obtained by the hypothesis creation unit up to the previous time A hypothesis likelihood tallying unit that calculates the total value of the likelihoods of all the constituent hypotheses, and a likelihood of each hypothesis in the hypothesis group by the total value calculated by the hypothesis likelihood tallying unit. A hypothesis reliability calculator that calculates the reliability evaluation value of the hypothesis, and a hypothesis reliability calculator that calculates the hypothesis reliability evaluation value. It is made and a hypothesis reduction section for reducing the number of hypotheses in the hypotheses based on the reliability evaluation value for the hypothesis.
[0066]
A target tracking method according to the present invention includes a Doppler velocity observation value extraction for extracting observation information that may possibly correlate with a Doppler velocity of a target at a current time with respect to a track obtained up to a time before a target to be tracked. Step, a position information observation value extraction step of extracting observation information that may have a correlation with the target position at the current time with respect to the track obtained by the time before the target to be tracked, and both observation value extraction steps. Creating a hypothesis group indicating a combination of wakes estimated at each observation time point of the target using the observation information extracted by the above, and, for each hypothesis in the hypothesis group, an observation position of the target. A hypothesis reduction step of reducing the number of hypotheses in the group of hypotheses based on the evaluation result and a probability of Doppler velocity. It comprises a smoothing step of performing smoothing with motion parameters including the target Doppler velocity, and a hypothesis selecting step of selecting a hypothesis that most accurately represents the target motion parameters at the current time from a group of hypotheses. is there.
[0067]
In the target tracking method according to the present invention, in the hypothesis reduction step, in evaluating the likelihood of each hypothesis in the hypothesis group, when the wake obtained by the previous time is extended by the target observation value at the current time, The likelihood of a wake based on the position of the target and the Doppler velocity is obtained, and the likelihood of the hypothesis at the current time derived from the wake obtained up to the previous time is calculated as an evaluation result using the likelihood of the wake.
[0068]
In the target tracking method according to the present invention, in the hypothesis reduction step, the wake T_t(T is the number of tracks in the hypothesis) is the current time t_kTarget observations z_{k, j}(J is the number of observations in the hypothesis), and the likelihood of the wake based on the position information of the target is g_{j, t}, The likelihood of the wake by the Doppler velocity of the target is g^D _{j, t}, The track T obtained by the time before deriving the hypothesis of the current time_tIs the reliability evaluation value of the hypothesis consisting of_{p (i)}, The detection probability for the target is P_DThe track T obtained by the previous time_tAnd the probability that the observation value at the current time is extracted as observation information that may have a correlation with respect to the position and the Doppler velocity is P_G, The average value of the number of new targets generated per unit volume of observation space is β_NT, The average number of erroneous signals generated per unit volume of the observation space is β_FT, Observation z_{k, j}Is the measured value of the Doppler velocity associated with^o _{k, j}, The number of tracks extended by any of the observations within the hypothesis at the current time_DT, The number of existing wakes determined not to correspond to any observation in the hypothesis at the current time_MT, The number of tracks newly started within the hypothesis at the current time, N_NT, The number of observations treated as false signals in the hypothesis at the current time is N_FTAnd the observed value is the wake T_tPredicted position z_{k | k-1}With covariance matrix S_kGenerated by a three-variable Gaussian distribution of_tDoppler velocity predicted value Dr_{k | k-1}With the residual covariance matrix S^D _kTrajectory T obtained up to the previous time, assuming that it occurs in a univariate Gaussian distribution of_tLikelihood γ of the current time hypothesis derived from_iIs calculated from the above equation.
[0069]
In the target tracking method according to the present invention, with respect to the track obtained before the time before the target to be tracked, there is a possibility that the observation value vector defined by the Doppler velocity of the target and the target position at the current time has a possibility of correlation. An observation value selection step of extracting certain observation information, and a hypothesis creation step of using the observation information extracted in the observation value extraction step to create a hypothesis group indicating a combination of wakes estimated at each observation time of the target. And for each hypothesis in the group of hypotheses, a hypothesis reduction step of evaluating the likelihood of the observation position and Doppler velocity of the target, and reducing the number of hypotheses in the group of hypotheses based on the evaluation result; For each wake within, a smoothing step of performing smoothing with an observation value vector defined by the Doppler velocity of the target and the target position, and In which and a hypothesis selecting step of selecting hypotheses most accurately represent the target of motion specifications.
[0070]
In the target tracking method according to the present invention, in the hypothesis reduction step, in evaluating the likelihood of each hypothesis in the hypothesis group, the likelihood of an observation value vector obtained by combining the target position vector and the Doppler velocity is determined. The likelihood of the hypothesis at the current time derived from the track obtained up to the previous time using the likelihood of the track is calculated as an evaluation result.
[0071]
In the target tracking method according to the present invention, in the hypothesis reduction step, the wake T_t(T is the number of tracks in the hypothesis) is the current time t_kTarget observations z_{k, j}(J is the number of observations in the hypothesis), and the observation value vector z including the measured value of the Doppler velocity of the target in the case where it is expanded by^M _{k, j}The likelihood of g^M _{j, t}, The track T obtained by the time before deriving the hypothesis of the current time_tIs the reliability evaluation value of the hypothesis consisting of_{p (i)}, The detection probability for the target is P_DThe track T obtained by the previous time_tAnd the probability that the observation value at the current time is extracted as observation value information that may have a correlation with respect to the position and the Doppler velocity is P_G, The average value of the number of new targets generated per unit volume of observation space is β_NT, The average number of erroneous signals generated per unit volume of the observation space is β_FT, The number of tracks extended by any of the observations within the hypothesis at the current time_DT, The number of existing wakes determined not to correspond to any observation in the hypothesis at the current time_MT, The number of tracks newly started within the hypothesis at the current time, N_NT, The number of observations treated as false signals in the hypothesis at the current time is N_FTAnd the observation vector is the wake T_tPredicted position z_{k | k-1}With the residual covariance matrix S^M _kTrack T obtained by the previous time, assuming that the_tLikelihood γ of the current time hypothesis derived from_iIs calculated from the above equation.
[0072]
The program according to the present invention is a Doppler velocity observation value extraction means for extracting observation information that may have a correlation with respect to the Doppler velocity of the target at the current time, for a track obtained before the time before the target to be tracked, Position information observation value extraction means for extracting observation information that may be correlated with the target position at the current time for the track obtained before the time before the target to be tracked, and observations extracted by both observation value extraction means Hypothesis creation means for creating a group of hypotheses indicating combinations of wakes estimated at each observation point of the target using the information, and for each hypothesis in the group of hypotheses, certainty regarding the observation position and Doppler velocity of the target. Hypothesis reduction means for reducing the number of hypotheses in the hypothesis group based on the evaluation result, including the Doppler velocity of the target for each track in the hypothesis group Smoothing means for performing smoothing in kinematic specifications, those causing a computer to function as a hypothesis selecting means for selecting the hypothesis that most accurately represent the goals of the exercise specifications at the current time from the hypotheses.
[0073]
The program according to the present invention provides an observation method that has a possibility that the track obtained by the time before the target to be tracked has a correlation with the observation value vector defined by the Doppler velocity and the target position of the target at the current time. A hypothesis creating means for creating a hypothesis group indicating a combination of wakes estimated at each observation point of the target using the observation information extracted by the observation value selecting means for extracting information, and the observation information extracted by the observation value extracting means; For each hypothesis, we evaluate the likelihood of the observation position and Doppler velocity of the target, and based on the evaluation result, reduce the number of hypotheses in the group of hypotheses. A smoothing means for performing smoothing with an observation value vector defined by the Doppler velocity of the target and the target position, and the most accurate movement parameters of the target at the current time from the group of hypotheses. Those causing a computer to function as a hypothesis selecting means for selecting the hypothesis that current.
[0074]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment of the present invention will be described.
Embodiment 1 FIG.
FIG. 1 is a diagram showing a configuration of a target tracking device according to Embodiment 1 of the present invention. In the figure, reference numeral 1 denotes a target tracking device according to the first embodiment, which is composed of a computer device that executes a program that embodies the components described below. Reference numeral 2 denotes a target observation device as a radar that obtains an observation value by observing a position vector and a Doppler velocity of a target in space. Reference numeral 3 denotes a target display device that displays a wake on a display and presents the user with the wake of the target being tracked. Reference numeral 4a denotes a gate calculation unit that operates in the same manner as that of the conventional target tracking device 101 described above, and is a region in which an observation value regarding the position information of the target at the current time is present for each track included in the existing cluster group. Is calculated. Reference numeral 4b denotes a gate calculation unit that calculates a gate that is an expected area for the observed value of the target Doppler velocity measurement value at the current time for each track included in the existing cluster group.
[0075]
Reference numeral 5a denotes an observation value selection unit (position information observation value extraction means) that operates in the same manner as that of the above-described conventional target tracking device 101, and observes all targets at the current time selected by the observation value selection unit 5b. From the values, the observation value included in the gate of each wake calculated by the gate calculation unit 4a is selected. 5b is an observation value selection for selecting an observation value having a Doppler velocity measurement value included in the gate based on the Doppler velocity for each wake among all observation values at the current time inputted from the target observation apparatus 2 to the target tracking apparatus 1. (Doppler velocity observation value extraction means). Reference numeral 6 denotes a storage unit for storing an in-system cluster table indicating the state of clusters in the target tracking device 1.
[0076]
Reference numeral 7 denotes a cluster new / integration unit (hypothesis creation means) which integrates existing clusters based on the relationship between the observation values from the observation value selection unit 5a and the existing clusters in the in-system cluster table, and creates a new cluster. Create As will be described later, in the trajectory estimation, all trajectories in the space are collectively constructed and processed for each trajectory group whose gates overlap each other. A wake group that is a unit of such a hypothesis creation process is called a cluster. The hypothesis is a set of tracks formed by selecting some tracks in the cluster. Further, each hypothesis is to determine which combination of tracks in the cluster is correct.
[0077]
A storage unit 8 stores an intra-cluster observation vector table created by the cluster new establishment / integration unit 7. The intra-cluster observation vector table stores observation values input from the observation value selection unit 5a as an observation value group for each cluster. Reference numeral 9 denotes a storage unit for storing a group of hypothesis status data in a cluster, which stores a hypothesis table in a cluster, a hypothesis track table, and a intra-cluster track-observation vector table. Reference numeral 9a denotes a storage area that functions as an intra-cluster hypothesis table indicating all the hypotheses in the cluster in the storage unit 9. Reference numeral 9b denotes a storage area that functions as a hypothesis-in-track table showing all the tracks in the hypothesis for each hypothesis in the storage unit 9. Reference numeral 9c denotes a storage area that functions as a track-intracluster-observation vector table indicating the relationship between all tracks in the cluster in the storage unit 9 and the observed values constituting the tracks.
[0078]
Reference numeral 10 denotes an intra-gate decision matrix calculation unit (hypothesis creation means) which inputs an intra-cluster observation vector table and data from the intra-cluster hypothesis status data group to calculate an intra-cluster gate intra-gate decision matrix. Reference numeral 11 denotes a storage unit for storing an intra-cluster gate in-gate determination matrix indicating the relationship between the observation value at the current time in the cluster and the wake. Reference numeral 12 denotes a wake correlation matrix calculation unit (hypothesis creation means) for inputting an intra-cluster gate judgment matrix and calculating an intra-cluster wake correlation matrix. Reference numeral 13 denotes a storage unit for storing an intra-cluster wake correlation matrix indicating the possibility of expanding a hypothesis in a cluster.
[0079]
Reference numeral 14 denotes a hypothesis updating unit (hypothesis creation means) which creates a new hypothesis corresponding to the state of the hypothesis based on the observation values up to the previous time and the observation value input at the current time from the intra-cluster wake correlation matrix. Reference numeral 15 denotes a hypothesis reduction unit (hypothesis reduction means) that evaluates the certainty (reliability) of each hypothesis in a cluster and reduces the number of hypotheses by deleting some of the hypotheses from the evaluation result. Reference numeral 16 denotes a trajectory determination unit (hypothesis selecting means) for selecting one of the best hypotheses from among a plurality of hypotheses in the cluster and determining the number of targets and the trajectory thereof.
[0080]
Reference numeral 17a denotes a position / velocity smoothing unit (smoothing means) that operates in the same manner as the above-described track tracking unit of the conventional target tracking device 101, and performs a Kalman filter smoothing process on each track in the cluster to obtain a position. And speed smoothing. Reference numeral 17b denotes a Doppler velocity smoothing unit (smoothing means), which smoothes the Doppler velocity for each track in the cluster. Reference numeral 18 denotes a cluster separation unit which checks whether or not the clusters after the hypothesis reduction can be separated and reconstructs the hypothesis of each cluster after the separation.
[0081]
Next, an outline will be described.
In the target tracking device 1 according to the first embodiment, a gate for Doppler information is provided in addition to a gate for position information in the conventional target tracking device 101. As a result, observation values that can be correlated with each wake can be more strictly selected, and cross-correlation can be reduced. Therefore, in addition to the tracking filter processing using the position information performed by the conventional target tracking apparatus 101, a tracking filter processing for Doppler velocity information as described below is separately provided.
[0082]
First, to track the Doppler velocity, another state vector x different from the above equation (3) is used.^D _kIs defined as in the following equation (22). Where Dr_kIs the target Doppler velocity (distance rate of change of distance), and Dr_kDot is Doppler speed Dr_kIs the first-order time derivative of. In addition, the measurement result of the Doppler velocity accompanying the observation value of the radar is Dr.^o _kAnd
(Equation 18)

[0083]
Further, a motion model of the Doppler velocity representing a temporal transition of the target Doppler velocity is represented by the following equation (23), and an observation model of the Doppler velocity (Doppler velocity measurement result Dr)^o _k) Is defined by the following equation (24). Where Φ^D _kIs the state vector x in the above equation (22).^D _kIs a state transition matrix for. H^D _kIs the observation matrix for the Doppler velocity observations. Also, w^D _kIs a driving noise vector for the state vector of the above equation (22), and has a mean zero and a covariance matrix Q^D _kSuppose that the normal white noise of v^D _kIs the observation noise of the scalar, with mean zero and variance R^D _kSuppose that the normal white noise of
[Equation 19]

[0084]
For example, assuming that the motion of the target Doppler velocity is linear with respect to time, the state transition matrix in the above equation (23) can be defined by the following equation (25). At this time, the driving noise vector w^D _kCan be regarded as expressing a truncation error term caused by regarding the motion of the Doppler velocity as linear. Also, the observation matrix H^D _kIs the state vector x in the above equation (22).^D _kTo extract the term of the Doppler velocity from the following equation, and can be defined by the following equation (26).
(Equation 20)

[0085]
Next, the operation will be described.
FIG. 2 is a flowchart showing the operation of the target tracking device in FIG. 1, and description will be made with reference to FIG.
First, time t_kWhen the observation value is input from the target observation device 2 to the target tracking device 2 (step ST1), the gate calculation unit 4b in the target tracking device 1 determines the state of the above equation (22) for all the tracks included in the existing cluster group. Vector x^D _kIs subjected to Kalman filter prediction processing shown in the following equations (27) and (28) (step ST2). Where x^D _{k-1 | k-1}Is the time t_k-1Is the smoothed value of the state vector. x^D _{k | k-1}Is the time t_kRepresents the predicted value of the state vector with respect to. Also, P^D _{k-1 | k-1}, P^D _{k | k-1}Is x^D _{k-1 | k-1}, X^D _{k | k-1}Is the error covariance matrix of
(Equation 21)

[0086]
Further, the gate calculation unit 4b calculates the current time t of each track according to the following equations (29) and (30)._kDoppler velocity predicted value Dr_{k | k-1}And the residual variance S^D _kIs calculated. This Dr_{k | k-1}Represents the center value of the Doppler velocity gate, and S^D _kAre the characteristics that characterize the spread of the Doppler velocity gate.
(Equation 22)

[0087]
Next, the observation value selection unit 5b checks which observation value is present in the Doppler velocity gate of each wake calculated by the gate calculation unit 4b, thereby first selecting observation values that can be correlated with each wake. (Step ST3, Doppler velocity observation value extraction step). Specifically, an observation value having a Doppler velocity measurement value that satisfies the following equation (31) for each track is selected as an observation value that can be correlated with the track. Where d_DIs a parameter that determines the size of the Doppler velocity gate.
[Equation 23]

[0088]
When the processing related to the Doppler velocity is completed as described above, the gate calculation unit 4a performs the Kalman filter prediction processing on the position and velocity state vectors of the above equation (3) for all wakes, thereby obtaining the current time. t_k(Step ST4, position information observation value extraction step). This process is the same as that of the conventional target tracking device 101.
[0089]
Subsequently, the observation value selection unit 5a based on the position information determines whether the observation value selected by the observation value selection unit 5b based on the Doppler velocity information for each track also exists in the gate related to the position information of each track. By observing this, an observation value that can be correlated with each track is finally determined (step ST5). The method of determining the inside of the gate by the observation value selecting unit 5a based on the position information is the same as that of the conventional target tracking device 101.
[0090]
As described above, in the target tracking device 1 according to the first embodiment, when an observation value that can be correlated with each wake is selected, it is determined that the observation value exists in the gate related to the Doppler velocity and that it exists in the gate related to the position information. By imposing both conditions, stricter selection is performed than in the case of the conventional target tracking method.
[0091]
Hereinafter, the processing from step ST6 to step ST10 by the cluster new / integration unit 7, the in-gate decision matrix calculation unit 10, the wake correlation matrix calculation unit 12, the hypothesis update unit 14, and the hypothesis reduction unit 15 is performed by a conventional target tracking device. This is the same as the processing from step ST103 to step ST107 shown in FIG. The processing by the position / velocity smoothing unit 17a in step ST11 is the same as the processing in step ST108 shown in FIG. 9 by the wake smoothing unit 116 in the conventional target tracking device 101. The description of these processing steps is omitted.
[0092]
When the hypothesis reduction processing is completed, the Doppler velocity smoothing unit 17b executes the smoothing processing on the Doppler velocity state vector of the above equation (22) for each track in the cluster according to the Kalman filter smoothing equation (step ST12). , Smoothing step). Specifically, the wake T_tIs the current time t_kIs expanded by the addition of a new observation, the measurement result Dr of the Doppler velocity accompanying this observation^o _kIs used to update the smoothed value according to the following equations (32), (33), and (34).
(Equation 24)

[0093]
Where x^D _{k | k-1}, P^D _{k | k-1}Is the wake T calculated by the above equations (27) and (28)._tIs a state vector prediction value and a prediction error covariance matrix. x^D _{k | k}Is the time t of this track_kIs the state vector smoothed value of^D _{k | k}Is the error covariance matrix. Also, K^D _kIs the Kalman gain matrix.
[0094]
The subsequent processing from step ST13 to step ST16 is the same as the processing from step ST109 to step ST112 shown in FIG. 9 by the conventional target tracking device 101. Therefore, description is omitted.
[0095]
As described above, according to the first embodiment, tracking filter processing for performing position and velocity smoothing for the purpose of utilizing Doppler velocity observation information in addition to position observation information from radar. And tracking filter processing for smoothing the Doppler velocity, and not only the position gate but also the Doppler velocity gate judgment is performed for each track, so observation values that can correlate with each track are strictly selected. In addition, it is possible to improve the tracking performance by reducing the cross-correlation of the observation value with respect to the wake. Further, since the number of observation values that can be correlated with each track is reduced, the number of hypotheses created in the apparatus is reduced as a result, and the processing efficiency of the entire apparatus can be improved.
[0096]
Embodiment 2 FIG.
FIG. 3 is a diagram showing a configuration of a hypothesis reduction unit in the target tracking device according to the second embodiment of the present invention. In the figure, reference numeral 15a denotes a hypothesis reduction unit (hypothesis reduction unit) according to the second embodiment, which has a function different from that of a conventional target tracking device 101 as described later. Can be applied. The hypothesis reduction unit 15a can be embodied as a program executed by a computer functioning as the target tracking device according to the second embodiment. Reference numeral 19 denotes a parent hypothesis reliability setting device (parent hypothesis reliability setting unit) which sets the reliability of the parent hypothesis at the previous time as an initial value of the hypothesis likelihood. Reference numeral 20 denotes a track state determination unit (track state determination unit) which checks the assignment state of the observation value at the current time to each track in the hypothesis, and according to the result, a detection probability multiplier 22 and a gate non-detection probability multiplication. The likelihood calculation value is sent to one of the multiplier 24 and the new target incidence rate multiplier 21.
[0097]
21 is a new target occurrence rate multiplier (new target occurrence rate multiplication unit) that multiplies the hypothesis likelihood calculation value by the new target occurrence rate. A probability multiplication unit and a likelihood multiplication block unit). Reference numeral 23a denotes a likelihood multiplier based on position information (likelihood multiplication unit, likelihood multiplication block unit), which multiplies the calculated likelihood likelihood by the likelihood of the position of the observed value. 23b is a likelihood multiplier (likelihood multiplication unit, likelihood multiplication block unit) based on Doppler velocity information, which multiplies the calculated hypothesis likelihood by the likelihood of the Doppler velocity of the observed value. Reference numeral 24 denotes an in-gate non-detection probability multiplier (in-gate non-detection probability multiplication unit) that multiplies the hypothesis likelihood calculation value by the in-gate non-detection probability.
[0098]
Reference numeral 25 denotes an observation value state determiner (observation value state determination unit) which checks how each observation value in the cluster is handled in a hypothesis, and only if the observation value is treated as an erroneous signal, the hypothesis The likelihood calculation value is sent to the error signal occurrence rate multiplier 26. Reference numeral 26 denotes an erroneous signal occurrence rate multiplier (an erroneous signal occurrence rate multiplier) that multiplies the hypothesis likelihood calculation value by the erroneous signal occurrence rate. Reference numeral 27 denotes a hypothesis likelihood storage unit that stores the likelihood calculation result of each hypothesis in the cluster. Reference numeral 28 denotes a hypothesis likelihood counting unit (hypothesis likelihood counting unit) that calculates the total value of the hypothesis likelihoods for all the hypotheses in the cluster. 29 is a hypothesis reliability calculator (hypothesis reliability calculation unit), which calculates the reliability evaluation value of each hypothesis by dividing the hypothesis likelihood of each hypothesis in the cluster by the total value of all the hypotheses. A hypothesis reducer (hypothesis reduction unit) 30 reduces the number of hypotheses by deleting hypotheses with low reliability according to the reliability evaluation value of the hypothesis.
[0099]
Next, an outline will be described.
The hypothesis reducing unit 15a according to the second embodiment is characterized in that the reliability evaluation value of each child hypothesis is calculated using not only the position observation information but also the Doppler velocity observation information. Specifically, the likelihood γ of each child hypothesis H (i)_iIs calculated using the following equation (35) instead of the above equation (14) in the case of the conventional target tracking device 101.
(Equation 25)

[0100]
Here, the above equation (35) is different from the above equation (14) in the case of the conventional target tracking device 101 in that the product operation of the second term on the right side is g^D _{j, t}Has been added. g^D _{j, t}Is the wake T_tIs the observed value z_{k, j}Is the likelihood of a wake based on Doppler velocity information when it is expanded by Also, this g^D _{j, t}Means that the Doppler velocity of the observed value is the wake T_tPredicted value Dr_{k | k-1}S centered on^D _kIs calculated according to the following equation (36), assuming that the signal is generated by a univariate Gaussian distribution. However, Dr^o _{k, j}Is the observed value z_{k, j}Is a measurement result of the Doppler velocity accompanying the. Note that the reliability evaluation value of each hypothesis is obtained by normalizing the hypothesis likelihood by the above equation (35) by the above equation (16), similarly to the case of the conventional target tracking device 101.
(Equation 26)

[0101]
Next, the operation will be described.
FIG. 4 is a flowchart showing the operation of the hypothesis reduction unit in FIG. 3, and the hypothesis reduction processing according to the second embodiment will be described in detail with reference to FIG.
First, the parent hypothesis reliability setting unit 19 selects one hypothesis H (i) in the cluster and determines the reliability β of the parent hypothesis of this hypothesis._{p (i)}Is the calculated likelihood likelihood γ_i(Step ST1a). The calculated value of the hypothesis likelihood is sent to the wake state determiner 20.
[0102]
Next, the track state determination unit 20 sequentially selects the tracks within the hypothesis selected by the parent hypothesis reliability setting unit 19 (step ST2a), and checks the assignment state of the observation value at the current time (step ST3a, step ST5a). ). That is, it is determined whether the selected track is one of the following (1) to (3).
[0103]
(1) New track started at the current time
(2) Tracks with new observations at the current time added to existing tracks at the previous time (stretched tracks)
(3) Tracks that were retained because there were no observations at the current time in existing tracks at the previous time
[0104]
The wake state determination unit 20 determines whether the above-mentioned determination result is the new target occurrence rate multiplier 21 when the determination result is (1), the detection probability multiplier 22 when the determination result is (2), and the (3) In the case of, the in-gate non-detection probability multiplier 24 provides the hypothesis likelihood calculation value γ_iIs sent.
[0105]
If it is determined in step ST3a that the track is a new track started at the current time, the new target occurrence rate multiplier 21 calculates the hypothesis likelihood calculation value γ input from the track state determiner 20._iThe new target incidence β, which is the average value of the number of new targets generated per unit volume of the observation space,_NTAnd the result of the multiplication is returned to the wake condition determiner 20 again (step ST4a).
[0106]
In step ST3a, it is determined that the current track is not the new track started at the current time, but is the track (extended track) in which a new observation value at the current time is added to the existing track at the previous time in step ST5a. Then, the detection probability multiplier 22 calculates the hypothesis likelihood calculation value γ input from the wake state determination unit 20._iDetection probability P_DAnd sends the multiplication result to the likelihood multiplier 23a based on the position information (step ST6a).
[0107]
The likelihood multiplier 23a based on the position information adds the likelihood g based on the position information calculated by the above equation (15) to the multiplication result value input from the detection probability multiplier 22._{j, t}And sends the multiplication result to a likelihood multiplier 23b based on Doppler velocity information (step ST7a). Further, the likelihood multiplier 23b based on the Doppler velocity information adds the likelihood g based on the Doppler velocity information calculated by the above equation (36) to the multiplication result value input from the likelihood multiplier 23b.^D _{j, t}(Step ST8a). The result of the multiplication is sent back to the wake condition determiner 20 again.
[0108]
Also, it is held that the new track started at the current time in step ST3a, the track expanded by the new observation value in step ST5a, and the existing track at the previous time have no observation value at the current time in step ST5a. If it is determined that the wake is a wake, the in-gate non-detection probability multiplier 24 calculates the hypothesis likelihood calculation value γ input from the wake state determiner 20._iCorresponds to the probability that the observed value of the wake is not detected in the gate (1-P_DP_G) (Here, this value is referred to as the in-gate non-detection probability), and the result of the multiplication is sent back to the wake state determiner again (step ST9a).
[0109]
The above-described series of processing is repeatedly executed until the wake state determiner 20 finishes selecting all the wakes in the hypothesis (step ST10a). When the processing has been completed for all the tracks in the hypothesis, the track state determination unit 20 sends the calculated value of the hypothesis likelihood to the observation value state determination unit 25.
[0110]
When the calculated hypothesis likelihood value is input from the wake state determination unit 20, the observation value state determination unit 25 sequentially selects each observation value in the cluster (step ST11a) and generates an erroneous signal in the focused hypothesis. It is checked whether it is handled (step ST12a). At this time, if the selected observation value is treated as an erroneous signal, the calculated hypothesis likelihood value is sent to the erroneous signal occurrence rate multiplier 26.
[0111]
The erroneous signal occurrence rate multiplier 26 adds the erroneous signal occurrence rate β to the calculated hypothesis likelihood value input from the observation value state determiner 25._FTAnd returns the multiplication result to the observed value state determiner 25 again (step ST13a). The observation value state determination unit 25 repeatedly executes the above operation until the parent hypothesis reliability setting unit 19 finishes selecting all the observation values in the selected hypothesis (step ST14a).
[0112]
When the processing from the initial value setting by the parent hypothesis reliability setting unit 19 to the repetition processing by the observation value state determination unit 25 is completed, the calculated hypothesis likelihood value includes the hypothesis likelihood γ according to the above equation (35) for the focused hypothesis._iWill be obtained. This calculation result is stored in the hypothesis likelihood storage unit 27.
[0113]
On the other hand, the hypothesis reducing unit 15a calculates the above-mentioned hypothesis likelihood, and sequentially executes a series of processes for storing the calculation result in the hypothesis likelihood storage unit 27 for all the hypotheses in the cluster (step ST15a). Here, when the calculation of the hypothesis likelihood is completed for all the hypotheses in the cluster, the hypothesis likelihood totalizer 28 reads the likelihood of the previous hypothesis stored in the hypothesis likelihood storage unit 27, and calculates the total value thereof. It is calculated (step ST16a). This result is sent to the hypothesis reliability calculator 29. The hypothesis reliability calculator 29 divides the likelihood of each hypothesis read from the hypothesis likelihood storage unit 27 by the total value input from the hypothesis likelihood totalizer 28 according to the above equation (16), and calculates the reliability of each hypothesis. Degree evaluation value β_iIs calculated (step ST17a).
[0114]
Finally, the hypothesis reducer 30 reduces the number of hypotheses by a method such as deleting hypotheses with low reliability according to the reliability evaluation value of the hypothesis calculated as described above.
[0115]
As described above, according to the second embodiment, the reliability evaluation value of the hypothesis created in the cluster is calculated using the Doppler velocity observation information in addition to the position observation information from the radar. Since the number of hypotheses is reduced using the evaluation value, unnecessary hypotheses can be more efficiently deleted. As a result, cross-correlation is reduced, and tracking performance can be improved. Further, since the number of hypotheses remaining in the device at each time is reduced, the processing efficiency of the entire device can be improved.
[0116]
Embodiment 3 FIG.
FIG. 5 is a diagram showing a configuration of a target tracking device according to Embodiment 3 of the present invention. In the figure, reference numeral 1A denotes a target tracking device according to the third embodiment, which is composed of a computer device that executes a program that embodies the components described below. Reference numeral 4c denotes a gate calculation unit that calculates a presence expectation area for a four-dimensional observation value vector of the current time position and the Doppler velocity for each track included in the existing cluster group. Reference numeral 5c denotes an observation value selection unit (observation value extraction means) based on the combined information of the position and the Doppler velocity. The target position is set for each track based on the entire observation value at the current time input from the target observation device 2 to the target tracking device 1A. And the observation value having the observation information included in the gate regarding the Doppler velocity. Reference numeral 17c denotes a track smoothing unit (smoothing means) based on the combined information of the position and the Doppler velocity, and updates the state vector of the position and the velocity using the observation information of the position and the Doppler velocity for each track in the cluster. I do. The same components as those in FIG. 1 are denoted by the same reference numerals, and redundant description will be omitted.
[0117]
Next, an outline will be described.
The target tracking device 1A according to the third embodiment is provided with a gate for a four-dimensional observation value vector of a position and a Doppler velocity instead of a gate for only the position information in the conventional target tracking device 101. As a result, observation values that can be correlated with each wake are more strictly selected, and cross-correlation can be reduced. Therefore, instead of the tracking filter processing using only the position information performed by the conventional target tracking apparatus 101, another tracking filter processing using the four-dimensional observation information of the position and the Doppler velocity is configured as follows. I do.
[0118]
First, as in the case of the conventional target tracking device 101, the time t_kTracking target state vector x composed of position and velocity at_kIs defined by the above equation (3). Also, the state vector x of the tracking target_kIs defined by the above equation (5). As described in the first embodiment, an example of the method of setting the motion model of the above equation (5) is, for example, assuming that the motion of the target is a constant velocity linear motion and calculating the state transition matrix Φ in the above equation (7)._kIs defined.
[0119]
On the other hand, time t_kIs defined by the following equation (37). Where x^o _k, Y^o _k, Z^o _kRepresents the target observation position coordinates. Also, Dr^o _kRepresents the measurement result of the Doppler velocity. The following equation (37) is different from the above-mentioned equation (6) in the conventional target tracking apparatus 101 in that the observation value vector includes the target observation position and the Doppler velocity measurement result.
[Equation 27]

[0120]
Further, a radar observation model is defined by a nonlinear function of the following equations (38), (39), and (40). Where v^M _kIs the mean zero, the covariance matrix R^M _kIs an observation noise vector according to the four-dimensional normal white noise. Also, r_kRepresents the target distance.
[Equation 28]

[0121]
In the third embodiment, an observation model as described above is set, and tracking filter processing is executed using an extended Kalman filter applicable to a nonlinear model. At this time, the linear first-order coefficient matrix H of the nonlinear function h (x)^M(X_k)Is required. This matrix can be calculated by the following equations (41) to (47).
(Equation 29)

[0122]
Note that the linear primary coefficient value matrix is a true value x of the state vector._kHowever, since the true value cannot be obtained by the actual tracking filter processing, the predicted value x_{k | k-1}Use the value in.
[0123]
Next, the operation will be described.
FIG. 6 is a flowchart showing the operation of the target tracking device in FIG. 5, and the operation will be described in detail with reference to FIG.
First, time t from the target observation device 2_kIs input (step ST1b), the gate calculation unit 4c in the target tracking device 1A calculates the state vector x of the above equation (3) for all the tracks included in the existing cluster group._kIs subjected to a prediction process. Here, the target motion model is defined similarly to the case of the conventional target tracking device 101. For this reason, the prediction processing is executed by the normal Kalman filter prediction processing equation based on the above equations (9) and (10) as in the conventional case.
[0124]
Subsequently, the gate calculation unit 4c calculates, for each track, the current time t of the four-dimensional observation value vector of the above equation (37)._kPredicted value z for^M _{k | k-1}And the residual covariance matrix S^M _kIs calculated according to the following equations (48) and (49) (step ST2b). Where x_{k | k-1}, P_{k | k-1}Is a state vector predicted value calculated by the normal Kalman filter prediction processing by the above equations (9) and (10) and its error covariance matrix. Also, the predicted value z of the observed value vector^M _{k | k-1}Represents the center of the position / Doppler velocity synthesis gate, and the residual covariance matrix S^M _kAre the characteristics that characterize the spread of the gate.
[Equation 30]

[0125]
Next, the observation value selection unit 5c based on the combined information of the position and the Doppler velocity checks which observation value is present in the position / Doppler velocity composition gate of each wake, thereby obtaining an observation value that can be correlated with each wake. Is determined (step ST3b). Specifically, an observation value that satisfies the condition of the following equation (50) is selected as an observation value that can be correlated with each track. Where d_MIs a parameter for determining the size of the position / Doppler velocity synthesis gate.
(Equation 31)

[0126]
Hereinafter, the processing from step ST4b to step ST8b by the cluster new / integration unit 7, the in-gate decision matrix calculation unit 10, the wake correlation matrix calculation unit 12, the hypothesis update unit 14, and the hypothesis reduction unit 15 is the same as the conventional target. This is the same as the processing from step ST103 to step ST107 shown in FIG. 9 by the tracking device 101. Therefore, description is omitted.
[0127]
Subsequently, the wake smoothing unit 17c based on the combined information of the position and the Doppler velocity uses the four-dimensional observation value vector of the above equation (37) according to the smoothing processing equation of the extended Kalman filter to obtain the state vector smooth value and its error. The variance matrix is updated (step ST9b). Specifically, the wake is the current time t_kIs expanded by the addition of a new observation value, the smoothed value is updated according to the following expressions (51) to (53) based on this observation value.
(Equation 32)

[0128]
Where x_{k | k}Is the time t with respect to the state vector of the above equation (3)._k, And P_{k | k}Is x_{k | k}Is the error covariance matrix of Also, K^M _kIs the gain matrix of the extended Kalman filter.
[0129]
Thereafter, the processing from step ST10b to step ST13b is the same as the processing from step ST109 to step ST112 shown in FIG. 9 by the conventional target tracking device 101. Therefore, description is omitted.
[0130]
As described above, according to the third embodiment, in order to utilize the Doppler velocity observation information in addition to the position observation information from the radar, the four-dimensional observation value composed of the position and the Doppler velocity is used. Since the vector is used to perform a tracking filter process for updating the smoothed value of the target state vector and to perform a gate determination on the position and Doppler velocity four-dimensional observation value vector, it is possible to correlate with each wake. Observed values are strictly selected, and cross-correlation of the observed values with respect to the wake is reduced, so that tracking performance can be improved. Further, since the number of observation values that can be correlated to each track decreases, the number of hypotheses created in the device decreases, and the processing efficiency of the entire device can be improved.
[0131]
Also, by defining a four-dimensional observation value vector, it is possible to form a gate in consideration of the correlation between the distribution of the observation error and the estimation error between the position and the Doppler velocity, and the processing load increases. It is possible to expect a higher-accuracy tracking process as compared with the target tracking device 1.
[0132]
Embodiment 4 FIG.
FIG. 7 is a diagram showing a configuration of a hypothesis reduction unit in the target tracking device according to the fourth embodiment of the present invention. In the figure, reference numeral 15b denotes a hypothesis reduction unit (hypothesis reduction unit) according to the fourth embodiment, which has a function different from that of the conventional target tracking device 101 as described later. Can be applied. The hypothesis reduction unit 15b can be embodied as a program executed by a computer functioning as the target tracking device according to the third embodiment. 23c is a likelihood multiplier (likelihood multiplication unit, likelihood multiplication block unit) based on the combined information of the position and the Doppler velocity, and a four-dimensional observation value vector composed of the position and the Doppler velocity in the hypothesis likelihood calculation value. Is multiplied by the likelihood of Note that the same components as those in FIG. 3 are denoted by the same reference numerals, and redundant description will be omitted.
[0133]
Next, an outline will be described.
The hypothesis reducing unit 15b according to the fourth embodiment is characterized in that the reliability evaluation value of each child hypothesis is calculated using the position and Doppler velocity observation information. Specifically, the likelihood γ of each child hypothesis H (i)_iIs calculated by the following equation (54) instead of the above equation (14) in the case of the conventional target tracking device 101.
[Equation 33]

[0134]
Here, the above equation (54) is different from the above equation (14) in the case of the conventional target tracking device 101 in that g in the product operation of the second term on the right side is_{j, t}Is g^M _{j, t}The changed part is different. This g^M _{j, t}Is the wake T_tIs the observed value z_{k, j}Is the likelihood of the observation value vector of the above equation (37) when the data is expanded by This observed value vector is the predicted value z of the above equation (48).^M _{k | k-1}And the residual covariance matrix S of the above equation (49)^M _kIs calculated using the following equation (55). Where z^M _{k, j}Is the observed value z_{k, j}Is the observation value vector of the above equation (37) corresponding to The reliability evaluation value of each hypothesis is obtained by normalizing the hypothesis likelihood by the above equation (54) by the above equation (16), as in the case of the conventional target tracking device 101.
(Equation 34)

[0135]
Next, the operation will be described.
Here, the operation of the hypothesis reduction unit 15b different from the above-described embodiment will be described. In the hypothesis reduction unit 15b, the likelihood multipliers 23a and 23b in the second embodiment are replaced with likelihood multipliers 23c based on combined information of the position and the Doppler velocity. The likelihood multiplier 23c calculates the likelihood g of the calculated value of the hypothesis likelihood input from the detection probability multiplier 22 using the four-dimensional observation value vector of the position and the Doppler velocity calculated by the above equation (55).^M _{j, t}And the result of the multiplication is returned to the wake-state determination unit 20. The operation procedure of the other parts is the same as that of the second embodiment, and the description is omitted.
[0136]
As described above, according to the fourth embodiment, the reliability evaluation value of the hypothesis created in the cluster is calculated using the likelihood of the position and the Doppler velocity based on the four-dimensional observation value vector, and the evaluation value is calculated. Since the number of hypotheses is reduced by using this method, the likelihood can be calculated in consideration of the correlation between the distribution of the observation error and the estimation error between the position and the Doppler velocity, and the processing load increases, but the hypothesis reduction of the second embodiment is performed. It is possible to expect a higher-accuracy tracking process as compared with the unit 15a.
[0137]
【The invention's effect】
As described above, according to the present invention, observation information that may be correlated with the Doppler velocity of the target at the current time is extracted from the track obtained before the time before the target to be tracked. Observation information that may be correlated with the target position at the time is extracted, and using the extracted observation information, a hypothesis group indicating a combination of wakes estimated at each observation time of the target is created. For each hypothesis, we evaluate the likelihood of the observation position and Doppler velocity of the target, reduce the number of hypotheses in the hypothesis group based on the evaluation result, and for each track in the hypothesis group, Smoothing is performed on the motion parameters including the Doppler velocity of the target, and a hypothesis that most accurately expresses the target's motion parameters at the current time is selected from the hypothesis group. It is strictly selected, there is an effect that cross correlation of the observed values for the track can be improved reduced by tracking performance.
[0138]
According to the present invention, when the hypothesis reducing unit evaluates the likelihood of each hypothesis in the hypothesis group, the position of the target in the case where the track obtained by the previous time is extended by the observation value of the target at the current time And the likelihood of the track based on the Doppler velocity is calculated, and the likelihood of the current time hypothesis derived from the track obtained up to the previous time is calculated as the evaluation result using the likelihood of the track, so that unnecessary hypotheses are calculated. There is an effect that it is possible to efficiently delete the data, reduce the cross-correlation, and improve the tracking performance. Further, since the number of hypotheses remaining in the device at each time decreases, there is an effect that the processing efficiency of the entire device can be improved.
[0139]
According to the present invention, with respect to the track obtained before the time before the target to be tracked, there is a possibility that there is a possibility that the observation value vector defined by the Doppler velocity of the target and the target position at the current time has a correlation. Is extracted, and using the extracted observation information, a hypothesis group indicating a combination of wakes estimated at each observation time of the target is created, and for each hypothesis in the hypothesis group, the observation position of the target and Evaluate the likelihood of the Doppler speed, reduce the number of hypotheses in the hypothesis group based on the evaluation result, and for each track in the hypothesis group, specify the Doppler speed of the target and the target position Smoothing is performed using the observation value vector, and a hypothesis that most accurately expresses the target motion data at the current time is selected from the hypothesis group, so that observation values that can correlate with each wake are strictly selected, and That cross correlation of the observed value there is an effect that it is possible to improve the reduction to tracking performance.
[0140]
According to the present invention, the hypothesis reducing means calculates the likelihood of the observation value vector obtained by combining the target position vector and the Doppler velocity in evaluating the likelihood of each hypothesis in the hypothesis group, and calculates the likelihood of these wakes. Is used to calculate the likelihood of the hypothesis at the current time derived from the track obtained up to the previous time as the evaluation result, so that the likelihood is calculated in consideration of the correlation between the distribution of the observation error and the estimation error between the position and the Doppler velocity. Although the processing load increases, there is an effect that a tracking process with higher accuracy can be expected.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a target tracking device according to a first embodiment of the present invention.
FIG. 2 is a flowchart showing an operation of the target tracking device in FIG. 1;
FIG. 3 is a diagram showing a configuration of a hypothesis reduction unit in a target tracking device according to a second embodiment of the present invention.
FIG. 4 is a flowchart showing an operation of a hypothesis reducing unit in FIG. 3;
FIG. 5 is a diagram showing a configuration of a target tracking device according to a third embodiment of the present invention.
FIG. 6 is a flowchart showing the operation of the target tracking device in FIG. 5;
FIG. 7 is a diagram showing a configuration of a hypothesis reduction unit in a target tracking device according to a fourth embodiment of the present invention.
FIG. 8 is a block diagram showing a configuration of a conventional target tracking device.
9 is a flowchart showing the operation of the target tracking device in FIG.
[Explanation of symbols]
1, 1A target tracking device, 2 target observation device, 3 target display device, 4a, 4b, 4c gate calculation unit, 5a observation value selection unit (position information observation value extraction means), 5b observation value selection unit (Doppler velocity observation value Extraction means), 5c observation value selection section (observation value extraction means), 6 storage section (cluster table in the system), 7 cluster new / integration section (hypothesis creation section), 8 storage section (intra-cluster observation vector), 9 storage Part (hypothesis situation data group in cluster), 9a storage area (hypothesis table in cluster), 9b storage area (hypothesis track table), 9c storage area (intracluster track-observation vector table), 10 gate in-gate judgment matrix calculation section (Hypothesis creation means), 11 storage section (intra-cluster gate in-gate decision matrix), 12 wake correlation matrix calculation section (hypothesis creation means), 13 storage section (intra-cluster wake correlation row) ), 14 hypothesis updating unit (hypothesis creation unit), 15, 15a, 15b hypothesis reduction unit (hypothesis reduction unit), 16 track determination unit (hypothesis selection unit), 17a position / velocity smoothing unit (smoothing unit), 17b Doppler Velocity smoothing unit (smoothing means), 17c wake smoothing unit (smoothing means), 18 cluster separation unit, 19 parent hypothesis reliability setting unit (parent hypothesis reliability setting unit), 20 wake state determination unit (wake state determination unit) ), 21 new target incidence rate multiplier (new target incidence rate multiplier), 22 detection probability multiplier (detection probability multiplier, likelihood multiplication block section), 23a, 23b, 23c likelihood multiplier (likelihood multiplication section) , Likelihood multiplication block unit), 24 non-detection probability in-gate multiplier (non-detection probability in-gate multiplication unit), 25 observation value state judgment unit (observation state judgment unit), 26 false signal occurrence rate multiplier (false signal Outbreak Multiplying unit), 27 the hypothesis likelihood storage unit, 28 hypothesis likelihood aggregation device (hypothesis likelihood counting unit), 29 hypotheses reliability calculator (hypothesis reliability calculation unit), 30 the hypothesis reducer (hypothesis reduction unit).

Claims (16)

Doppler velocity observation value extraction means for extracting observation information that may be correlated with respect to the Doppler velocity of the target at the current time with respect to the track obtained by the time before the target to be tracked,
Position information observation value extraction means for extracting observation information that may be correlated with the target position at the current time for the track obtained by the time before the target to be tracked;
Using the observation information extracted by the two observation value extraction means, a hypothesis creation means for creating a hypothesis group indicating a combination of wakes estimated at each observation time of the target,
For each hypothesis in the group of hypotheses, a hypothesis reduction unit that evaluates the likelihood of the observation position and Doppler velocity of the target, and reduces the number of hypotheses in the group of hypotheses based on the evaluation result.
For each wake in the group of hypotheses, smoothing means for performing smoothing with motion parameters including the Doppler velocity of the target,
A target tracking device comprising: a hypothesis selecting unit that selects a hypothesis that most accurately expresses the target motion data at the current time from the hypothesis group. The hypothesis reduction means evaluates the likelihood of each hypothesis in the hypothesis group, and when the wake obtained by the previous time is extended by the target's observation value at the current time, the wake of the wake according to the position of the target and the Doppler velocity 2. The target tracking device according to claim 1, wherein the likelihood is obtained, and the likelihood of the hypothesis at the current time derived from the wake obtained up to the previous time is calculated as an evaluation result using the likelihood of the wake. Hypothesis reduction means, track T _t obtained up to the previous time _(t is the number of track in the hypothesis) is the observed value z _k of the target at the present time t _{k, j (j} is observations number in the hypothesis) is extended by In this case, the likelihood of the track based on the position information of the target is g _{j, t} , the likelihood of the track based on the Doppler velocity of the target is g ^Dj _{, t} , and the track obtained by the time before the current time hypothesis is derived. The reliability evaluation value of the hypothesis consisting of T _{t at} the previous time is β _{p (i)} , the detection probability of the target is P _D , and the track T _t obtained up to the previous time and the observation value of the current time are the target position and Doppler. probability P _G extracted as observation information of the possibility of correlation for _speed, a new target beta _NT average value of the number of occurrences per unit volume of the observation _space, the false signal number generation per unit volume of the observation space dots associated with the average value to beta _FT, observed value _{z k, j} Dr o ^k a measurement value of the La ^speed, _j, and those that do not correspond to any of the observed value of the number of the track that has been extended by any of the observations in the hypothesis of the current time N _DT, in the hypothesis of the current time is number of N _MT existing _track, the number N _NT of the newly initiated trajectory in hypotheses at the current _time, the number of observations which are treated as erroneous signal within hypotheses at the current time and N _FT, observed value predicted position z _k of the track T _{t |} generated in the 3 bivariate Gaussian distribution of the covariance matrix S _k around a _k-1, the predicted value Dr k of the Doppler velocity of the observation value is the Doppler velocity of the track T _{t | k} Assuming that it occurs in a univariate Gaussian distribution of the residual covariance matrix S ^D _k centered at ₋₁ , the likelihood γ _i of the hypothesis at the current time derived from the track T _t obtained up to the previous time is Calculate from the following formula

3. The target tracking device according to claim 2, wherein: The hypothesis reduction means
A detection probability multiplication unit that multiplies the likelihood of the hypothesis obtained by the previous time by the detection probability of the target; and multiplies the multiplication result of the detection probability multiplication unit by the likelihood related to the target position in the observed value of the target. A position information likelihood multiplication unit, a likelihood multiplication block unit having a Doppler velocity likelihood multiplication unit that multiplies the multiplication result of the position information likelihood multiplication unit by the likelihood of the Doppler velocity in the target observation value,
A new target incidence multiplying unit that multiplies the likelihood of the hypothesis obtained up to the previous time by a new target incidence which is an average value of the number of occurrences of new targets per unit volume of the observation space;
The likelihood of the hypothesis obtained up to the previous time is multiplied by the in-gate non-detection probability indicating the degree to which the observation value correlated with the wake in the hypothesis is not extracted as observation information that may be correlated with respect to position and Doppler velocity. A non-detection probability multiplication unit in the gate,
Any of the detection probability multiplying unit, the non-detection in-gate probability multiplying unit, and the new target occurrence rate multiplying unit according to the assignment state of the observation value at the current time to each track in the hypothesis obtained by the previous time. A wake state determination unit that sends the likelihood of the above hypothesis,
An error signal occurrence rate multiplying unit that multiplies the likelihood of the hypothesis obtained up to the previous time by an error signal occurrence rate that is an average value of the number of error signals generated per unit volume of the observation space,
When the observed value is treated as an erroneous signal in the hypothesis obtained by the previous time, an observation value state determination unit that sends the likelihood of the hypothesis obtained by the previous time to the erroneous signal occurrence rate multiplying unit; ,
A hypothesis likelihood tallying unit that calculates a total value of likelihoods of all hypotheses constituting a hypothesis group obtained by the hypothesis creation unit up to the previous time;
A hypothesis reliability calculation unit that divides the likelihood of each hypothesis in the hypothesis group by the total value calculated by the hypothesis likelihood aggregation unit and calculates the division result as a reliability evaluation value of each hypothesis,
2. The target tracking device according to claim 1, further comprising: a hypothesis reduction unit configured to reduce the number of hypotheses in the hypothesis group based on the reliability evaluation value of the hypothesis calculated by the hypothesis reliability calculation unit. . Observation value selection that extracts observation information that may have a correlation with the observation value vector defined by the Doppler velocity of the target and the target position at the current time for the track obtained before the time before the target to be tracked Means,
Using the observation information extracted by the observation value extraction means, hypothesis creation means for creating a hypothesis group indicating a combination of wakes estimated at each observation time of the target,
For each hypothesis in the group of hypotheses, a hypothesis reduction unit that evaluates the likelihood of the observation position and Doppler velocity of the target, and reduces the number of hypotheses in the group of hypotheses based on the evaluation result.
For each wake in the hypothesis group, smoothing means for performing smoothing with an observation value vector defined by the Doppler velocity of the target and the target position,
A target tracking device comprising: a hypothesis selecting unit that selects a hypothesis that most accurately expresses the target motion data at the current time from the hypothesis group. The hypothesis reducing means calculates the likelihood of the observation value vector obtained by combining the target position vector and the Doppler velocity in evaluating the likelihood of each hypothesis in the hypothesis group, and uses the likelihood of these wakes until the previous time. 6. The target tracking device according to claim 5, wherein the likelihood of the hypothesis at the current time derived from the track obtained in step (b) is calculated as an evaluation result. Hypothesis reduction means, track T t obtained up to the previous time (t is the number of track in the hypothesis) is the observed value z k of the target at the present time t k, j (j is observations number in the hypothesis) is extended by when the observed value vector z M k consisting of the observation position and the Doppler velocity measurement values of the target, the likelihood of j g M j, t, from track T t determined by the time before directing the hypothesis of current time The hypothesis that the reliability evaluation value at the previous time is β p (i) , the detection probability of the target is P D , and the track T t obtained up to the previous time and the observation value at the current time are correlated with respect to the position and the Doppler velocity. The probability of being extracted as possible observation information is P G , the average value of the number of new targets generated per unit volume of the observation space is β NT , and the average value of the number of false signal generations per unit volume of the observation space is β FT , expanded by any observation within the hypothesis at the current time N DT is the number of tracks that have been stretched, N MT is the number of existing tracks that are assumed not to correspond to any observation value in the hypothesis at the current time, and N MT is the number of tracks that are newly started in the hypothesis at the current time. The number N NT , the number of observations treated as erroneous signals in the hypothesis at the current time is N FT , and the observation value vector is a residual covariance matrix around the predicted position z k | k−1 of the wake T t Calculating the likelihood γ i of the hypothesis at the current time derived from the track T t obtained up to the previous time, assuming that it occurs in a four-variable Gaussian distribution of S M k , from the following equation:

7. The target tracking device according to claim 6, wherein: The hypothesis reduction means
A detection probability multiplication unit that multiplies the likelihood of the hypothesis obtained up to the previous time by the detection probability of the target, and a multiplication result of the detection probability multiplication unit specified by a target position and a Doppler velocity in an observation value of the target. A likelihood multiplication block unit having a likelihood multiplication unit that multiplies the likelihood of the observed value vector to be obtained,
A new target incidence multiplying unit that multiplies the likelihood of the hypothesis obtained up to the previous time by a new target incidence which is an average value of the number of occurrences of new targets per unit volume of the observation space;
In-gate non-detection that multiplies the likelihood of the hypothesis obtained up to the previous time by the in-gate non-detection probability that indicates the degree to which observations correlated with the wakes in the hypothesis are not extracted as potentially correlated observation information A probability multiplier,
Any of the detection probability multiplying unit, the non-detection in-gate probability multiplying unit, and the new target occurrence rate multiplying unit according to the assignment state of the observation value at the current time to each track in the hypothesis obtained by the previous time. A wake state determination unit that sends the likelihood of the above hypothesis,
An error signal occurrence rate multiplying unit that multiplies the likelihood of the hypothesis obtained up to the previous time by an error signal occurrence rate that is an average value of the number of error signals generated per unit volume of the observation space,
When the observed value is treated as an erroneous signal in the hypothesis obtained by the previous time, an observation value state determination unit that sends the likelihood of the hypothesis obtained by the previous time to the erroneous signal occurrence rate multiplying unit; ,
A hypothesis likelihood tallying unit that calculates a total value of likelihoods of all hypotheses constituting a hypothesis group obtained by the hypothesis creation unit up to the previous time;
A hypothesis reliability calculation unit that divides the likelihood of each hypothesis in the hypothesis group by the total value calculated by the hypothesis likelihood aggregation unit and calculates the division result as a reliability evaluation value of each hypothesis,
The target tracking device according to claim 5, further comprising: a hypothesis reducing unit configured to reduce the number of hypotheses in the hypothesis group based on the reliability evaluation value of the hypothesis calculated by the hypothesis reliability calculation unit. . A Doppler velocity observation value extraction step of extracting observation information that may have a correlation with respect to the Doppler velocity of the target at the current time with respect to the track obtained by the previous time of the target to be tracked,
A position information observation value extraction step of extracting observation information that may be correlated with the target position at the current time with respect to the track obtained by the time before the target to be tracked,
Using the observation information extracted in both the observation value extraction step, a hypothesis creation step of creating a hypothesis group indicating a combination of wakes estimated at each observation time of the target,
For each of the hypotheses in the hypothesis group, evaluate the likelihood of the observation position and Doppler velocity of the target, and based on the evaluation result, a hypothesis reduction step of reducing the number of hypotheses in the hypothesis group,
For each wake in the hypothesis group, a smoothing step of performing smoothing with motion parameters including the Doppler velocity of the target,
A hypothesis selecting step of selecting a hypothesis that most accurately expresses the target motion data at the current time from the hypothesis group. In the hypothesis reduction step, when evaluating the likelihood of each hypothesis in the group of hypotheses, the trajectory based on the position of the target and the Doppler velocity when the wake obtained by the previous time is extended by the observation value of the target at the current time 10. The target tracking method according to claim 9, wherein the likelihood of a current time hypothesis derived from the track obtained up to the previous time is calculated as an evaluation result using the likelihood of the track using the likelihood of the track. . Expansion in the hypothesis reduction step, the observed value z _k of the goals of the current time t _k before _(the number of wake in the t hypothesis) wake _{T t} determined by _time, by _{j (j} is the observed value of the number of the hypothesis) In this case, the likelihood of the wake based on the position information of the target is determined by g _{j, t} , the likelihood of the wake based on the Doppler velocity of the target is determined by g ^Dj _{, t} , and the time before the current time hypothesis is derived. The reliability evaluation value of the hypothesis consisting of the track T _{t at} the previous time is β _{p (i)} , the detection probability of the target is P _D , and the track T _t obtained up to the previous time and the observation value of the current time are the position and the Doppler. probability P _G extracted as observation information of the possibility of correlation for _speed, a new target beta _NT average value of the number of occurrences per unit volume of the observation _space, the false signal number generation per unit volume of the observation space de accompanying the average value beta _FT, observed value _{z k, j} Dr o ^k a measurement value of the plastic ^speed, _j, and those that do not correspond to any of the observed value of the number of the track that has been extended by any of the observations in the hypothesis of the current time N _DT, in the hypothesis of the current time is number of N _MT existing _track, the number N _NT of the newly initiated trajectory in hypotheses at the current _time, the number of observations which are treated as erroneous signal within hypotheses at the current time and N _FT, observed value predicted position z _k of the track T _{t |} generated in the 3 bivariate Gaussian distribution of the covariance matrix S _k around a _k-1, the predicted value Dr k of the Doppler velocity of the observation value is the Doppler velocity of the track T _{t | k} Assuming that it occurs in a univariate Gaussian distribution of the residual covariance matrix S ^D _k centered at ₋₁ , the likelihood γ _i of the hypothesis at the current time derived from the track T _t obtained up to the previous time is Calculate from the following formula

The target tracking method according to claim 10, wherein: Observation value selection that extracts observation information that may have a correlation with the observation value vector defined by the Doppler velocity of the target and the target position at the current time for the track obtained before the time before the target to be tracked Steps and
A hypothesis creation step of creating a hypothesis group indicating a combination of wakes estimated at each observation time of the target using the observation information extracted in the observation value extraction step,
For each of the hypotheses in the hypothesis group, evaluate the likelihood of the observation position and Doppler velocity of the target, and based on the evaluation result, a hypothesis reduction step of reducing the number of hypotheses in the hypothesis group,
For each track in the hypothesis group, a smoothing step of performing smoothing with an observation value vector defined by the Doppler velocity and the target position of the target,
A hypothesis selecting step of selecting a hypothesis that most accurately expresses the target motion data at the current time from the hypothesis group. In the hypothesis reduction step, in evaluating the likelihood of each hypothesis in the hypothesis group, the likelihood of an observation value vector obtained by combining the target position vector and the Doppler velocity is obtained, and the likelihood of these wakes is used to calculate the previous time. 13. The target tracking method according to claim 12, wherein the likelihood of the hypothesis at the current time derived from the track obtained up to that point is calculated as an evaluation result. Expansion in the hypothesis reduction step, the observed value z k of the goals of the current time t k before (the number of wake in the t hypothesis) wake T t determined by time, by j (j is the observed value of the number of the hypothesis) in the case where the observed value vector z M k consisting of the observation position and the Doppler velocity measurement values of the target, the likelihood of j g M j, t, track T t determined by the time before directing the hypothesis of current time Is the reliability evaluation value at the previous time of the hypothesis consisting of β p (i) , the detection probability of the target is P D , and the track T t obtained up to the previous time and the observation value at the current time are related to the position and Doppler velocity possibility probability P G extracted as observation information there is, the new target beta NT average value of the number of occurrences per unit volume of the observation space, the false signal generation number average value per unit volume of the observation space To β FT , one of the observations within the hypothesis at the current time N DT is the number of stretched tracks, N MT is the number of existing tracks determined not to correspond to any observed value in the hypothesis at the current time, and the track newly started in the hypothesis at the current time. N NT , the number of observations treated as erroneous signals in the hypothesis at the current time is N FT , and the observation value vector is the residual covariance around the predicted position z k | k−1 of the wake Tt The hypothesis likelihood γ i of the current time hypothesis derived from the track T t obtained up to the previous time is calculated from the following equation, assuming that the matrix S M k is generated by a four-variable Gaussian distribution.

14. The target tracking method according to claim 13, wherein: Doppler velocity observation value extraction means for extracting observation information that may have a correlation with the Doppler velocity of the target at the current time with respect to the track obtained by the time before the target to be tracked,
Position information observation value extraction means for extracting observation information that may be correlated with the target position at the current time with respect to the track obtained by the time before the target to be tracked,
Using the observation information extracted by the two observation value extraction means, a hypothesis creation means for creating a hypothesis group indicating a combination of wakes estimated at each observation time of the target,
For each hypothesis in the hypothesis group, a hypothesis reduction unit that evaluates the likelihood of the observation position and Doppler velocity of the target and reduces the number of hypotheses in the hypothesis group based on the evaluation result.
Smoothing means for performing smoothing on each track in the hypothesis group with motion parameters including the Doppler velocity of the target,
A program that causes a computer to function as hypothesis selecting means for selecting a hypothesis that most accurately represents the target motion data at the current time from the group of hypotheses. Observation value selection that extracts observation information that may have a correlation with the observation value vector defined by the Doppler velocity of the target and the target position at the current time for the track obtained before the time before the target to be tracked means,
Using the observation information extracted by the observation value extraction means, a hypothesis creation means for creating a hypothesis group indicating a combination of wakes estimated at each observation time of the target,
For each hypothesis in the hypothesis group, a hypothesis reduction unit that evaluates the likelihood of the observation position and Doppler velocity of the target and reduces the number of hypotheses in the hypothesis group based on the evaluation result.
For each wake in the hypothesis group, a smoothing unit that performs smoothing with an observation value vector defined by the Doppler velocity and the target position of the target,
A program that causes a computer to function as hypothesis selecting means for selecting a hypothesis that most accurately represents the target motion data at the current time from the group of hypotheses.

Images