JP2004151978A - Figure recognition device, figure recognition method and passage control device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a figure recognition device with which a reproducible standing position and the reproducible image of the result thereof can be inputted and collated, permitting the acquisition of a high recognition rate as a result. <P>SOLUTION: In the figure recognition device with which the facial image of a recognition object person (figure) is inputted by a camera, a face is detected from the inputted facial image to recognize the person, on a display part which displays an operation guidance or the like to the recognition object person at the time of the registration and recognition of the recognition object person, guide lines b, c are displayed which guide the object person to an optimum standing position varying with the stature of the recognition object person, and a circle pattern d indicating the position and size of the actually detected face of the recognition object person is brought into contact with the inside of the guide lines b, c and is displayed. <P>COPYRIGHT: (C)2004,JPO

Description

と対角化して、固有ベクトルのΦを求める。この部分空間は、人物の同定を行なうための認識辞書として利用する。この情報をあらかじめ登録しておいて、それを辞書として登録しておけばよい。
また、後で述べるように、認識を行なうための入力データとして部分空間自身を利用してもよい。したがって、部分空間を算出した結果は、認識部１１０および登録情報保持部１０９に送られる。
【００２６】
認識部１１０は、登録情報保持部１０９に蓄えられた認識データ（部分空間）と特徴量抽出部１０８で得られた特徴量（濃淡情報あるいは部分空間情報）とを照合（比較）することにより、カメラ１０１に写っている認識対象者１００が誰であるかを認識、あるいは、該当人物であるかどうかを同定する。人物を認識するためには、どの人物のデータに最も類似しているかを求めればよく、最大類似度をとるデータに対応する人物を認識結果とすればよい。
【００２７】
また、カードや登録番号、暗証番号、鍵などを用いて、顔画像による認識を行なう場合には、それぞれの人物のカードや番号などの個人情報に対応する認識データとの類似度を計算して、設定した閾値と照合し、その閾値を越えた場合に、その人物と同定する。
【００２８】
これらカードや登録番号、暗証番号、鍵など入力する手段が必要であるが、たとえば、カードならＩＣカード、ＩＤカード、無線カードなどを用いることで実現できる。登録番号や暗証番号を用いる場合は、キー入力手段を設ければよい。
【００２９】
認識の方法としては、特徴量抽出部１０８の情報を用いる場合は、部分空間法や複合類似度法などの方法を用いてよい。本実施の形態での認識方法は、たとえば、文献（前田賢一、渡辺貞一：「局所的構造を導入したパターン・マッチング法」、電子情報通信学会論文誌（Ｄ），ｖｏｌ．Ｊ６８−Ｄ，Ｎｏ．３，ｐｐ３４５−３５２（１９８５））に記載されている公知の認識方法である相互部分空間法を用いる。
【００３０】
この方法では、あらかじめ蓄えられた登録情報の中の認識データも、入力されるデータも部分空間として表現され、２つの部分空間のなす「角度」を類似度として定義する。ここで入力される部分空間を入力部分空間という。入力データ列に対して同様に相関行列Ｃｉｎを求め、
【数２】

と対角化し、固有ベクトルΦｉｎを求める。２つのΦｉｎ、Φｄで表わされる部分空間の部分空間類似度（０．０〜１．０）を求め、それを認識するための類似度とする。
【００３１】
認識部１１０は、図４に示すフローチャートのように動作する。まず、認識部１１０は、認識作業を行なうか、同定作業を行なうかによって動作が異なる（ステップＳＴ１）。同定動作を行なう場合は、まず対象とする認識対象者のＩＤコードを読込む（ステップＳＴ２）。次に、対象とするＩＤコードに対応した登録情報（部分空間）を登録情報保持部１０９から読出す（ステップＳＴ３）。
【００３２】
次に、上述したように、部分空間法などにより認識を行なうため、各登録情報の部分空間と、入力ベクトル（特徴量抽出部１０８からの固有ベクトル）、または、入力部分空間との類似度を計算する（ステップＳＴ４）。次に、その類似度をあらかじめ設定されている閾値と比較し（ステップＳＴ５，ＳＴ６）、同定結果を出力する（ステップＳＴ７）。
【００３３】
認識作業を行なう場合、認識対象となるデータを全て登録情報保持部１０９から読出す（ステップＳＴ８）。そして、それぞれの登録情報との類似度を計算する（ステップＳＴ９）。次に、計算された類似度の中から最大のものを選択し（ステップＳＴ１０）、それを認識結果として出力する（ステップＳＴ１２）。
【００３４】
なお、図４の破線で囲んだステップＳＴ１１のように、最大類似度を閾値判定することによって、認識結果が正しいかどうかを検証することもできる（ステップＳＴ１３）。たとえば、類似度があまりに低い場合には、どの認識対象でもないと判断することもできる。
【００３５】
登録情報保持部１０９は、認識対象者を同定するために利用する部分空間（または、相関行列など）や、登録の際の時刻、日時、場所などの状況情報などを蓄積できる。なお、画像入力データや抽出した特徴量などでもよい。
【００３６】
以上までの処理で基本的な人物認識の動作を行なうことが可能となるが、前述したように、認識対象者の立ち位置や姿勢の変化により、認識率の低下を引き起こしてしまう。たとえば、登録時と同じ状態で認識時（照合時）も顔が画像内に存在している場合には、登録時と近い状態で特徴情報が抽出できるために登録辞書との類似度が高くなりやすいが、登録時よりも顔が大きかったり小さかったりした場合や、図には示されていないが顔の向きなどが異なる場合に類似度が低下しやすくなる。
【００３７】
そこで、本実施の形態では、表示制御部１１１および表示部１０４により、たとえば、図５に示すように、撮影されている認識対象者の顔の動画像ａを表示するだけでなく、認識時（および登録時）に、認識対象者をカメラ１０１の視野中央に自然に立たせるためのガイドとして、台形状に配置された２本の直線からな顔入力用のガイド線（２本の直線）ｂ，ｃを常時表示するとともに、このガイド線ｂ，ｃの内側に実際に検出した顔位置やサイズを示すガイドとして、顔入力用の環状パターン、たとえば、円パターンｄを表示する。
なお、図５（ａ）は認識対象者が標準的な身長の場合の表示例を示し、図５（ｂ）は認識対象者が身長が高い場合の表示例を示している。
【００３８】
ガイド線ｂ，ｃは、標準的なデータで得られた、平均的な顔入力用ガイドの円パターンｄ（直径は標準的なデータから推定）に外接した線となる。このガイド線ｂ，ｃが垂直でなく、台形状に斜めの角度を持っているのは、認識対象者の身長によりカメラ１０１に写ることのできる立ち位置が異なるため（身長の高い人物は後ろに立ち、顔は高い位置になり、身長の低い人物は前に立ち、顔は低い位置になる）、顔の写る高さに応じて相対的に円パターンｄの大きさが変化することを示すためである。
【００３９】
この場合、このガイド線ｂ，ｃを表示するか、図６に示すように、このガイド線ｂ，ｃ以外の表示エリア（斜線部分）を表示せず、マスクする（当該エリアを白色あるいはグレー色などで表示する）方法も有効である。この場合、ガイド線ｂ，ｃの少し外側までを表示しないようにすることも次の点で有効である。すなわち、認識対象者が、厳密に顔サイズを特定することは、使い勝手が落ちることや、精度の問題として許容できないことが想定され、顔サイズの多少の変動を許容するためである。
【００４０】
また、このガイド線ｂ，ｃの傾き（上側の幅と下側の幅で定義される）は、平均的なデータで構成されるため、機器としての標準値の他、実際に登録されたデータを基に設定してもよい。
【００４１】
また、顔のサイズを一定にするための顔入力用ガイドとしての円パターンｄは、顔検出部１０７において実際に検出された目、鼻の位置から算出される顔の中心位置と直径で描画される。この表示制御は表示制御部１１１によって行なわれる。
【００４２】
以下、顔認識動作を例にとって、ガイド線ｂ，ｃおよび円パターンｄの表示制御について、図７に示すフローチャートを参照して詳細に説明する。
まず、本装置側では、表示部１０４に図５に示したようなガイド線ｂ，ｃを常に表示する（ステップＳ１１）。
一方、認識対象者１００は、本装置（カメラ１０１）の前に立ち、表示部１０４の画面を目視することにより（ステップＳ２１）、表示部１０４に円パターンｄが表示されているか否かをチェックし（ステップＳ２２）、円パターンｄが表示されていない場合は、顔や立ち位置を動かすことにより、円パターンｄを表示させる（ステップＳ２３）。
【００４３】
一方、本装置側では、顔検出部１０７が、カメラ１０１により取込んだ画像から顔領域を検出し、その中心座標およびサイズ（顔の大きさ）を計算する（ステップＳ１２）。次に、検出した顔のサイズは所定のＹ座標の±１０％か否かをチェックし（ステップＳ１３）、そうでなければステップＳ１２に戻り、上記同様な動作を繰り返す。
【００４４】
ステップＳ１３のチェックの結果、検出した顔のサイズが所定のＹ座標の±１０％であれば、図５に示すように、表示部１０４に表示されているガイド線ｂ，ｃの内側で、当該認識対象者１００の身長に応じた高さ（つまり、目、鼻を検出した高さ）に、検出した顔のサイズに対応した直径の円パターンｄを表示する（ステップＳ１４）。すなわち、標準のガイド線ｂ，ｃに対して、一定の割合（±１０％程度）の顔サイズにならないと、円パターンｄの表示および次に説明する照合処理をしないように制御を行なうものである。
【００４５】
なお、この場合、認識対象者１００に対し、この円パターンｄがガイド線ｂ，ｃに接する位置に立ち位置を変更するよう、文字や音声などにより案内指示することも可能である。
【００４６】
次に、認識部１１０において、前述したように、特徴量抽出部１０８で得られた特徴量と登録情報保持部１０９に蓄えられた認識データとを照合することにより、両者の類似度を求める（ステップＳ１５）。次に、求めた類似度はあらかじめ設定された閾値以上か否かをチェックし（ステップＳ１６）、閾値以上であれば、当該認識対象者１００は本人であると判定し、認識処理を終了する。
【００４７】
ステップＳ１６のチェックの結果、求めた類似度が閾値以上でなければ、当該認識処理を開始してから指定時間以上経過したか否かをチェックし（ステップＳ１７）、指定時間以上経過していなければステップＳ１２に戻って上記同様な動作を繰り返し、指定時間以上経過していれば、ここで認識処理を終了する。
【００４８】
なお、上記説明では、顔認識時を例にとってガイド線ｂ，ｃおよび円パターンｄの表示について説明したが、顔登録時にも同様にガイド線ｂ，ｃおよび円パターンｄの表示を行なうようになっており、その表示制御は顔認識時と同様であるので、説明は省略する。
【００４９】
このように、顔認識時、顔登録時に表示部１０４にガイド線ｂ，ｃおよび円パターンｄを表示することにより、不特定多数の身長や顔の大きさの異なる認識対象者１００に対して、適切で再現性のある位置に顔を提示させることができる。しかも、入力画像を一定にすることができるため、照合スコア（類似度）をあげることができる。
【００５０】
次に、第２の実施の形態について説明する。
第２の実施の形態は、第１の実施の形態で説明した人物認識装置を、通行者の顔画像を認識して通行者の通行を制御する通行制御装置に適用した場合の例である。
【００５１】
図８は、第２の実施の形態に係る通行制御装置の構成を概略的に示すものである。この通行制御装置は、たとえば、重要施設（セキュリティ重視の部屋など）への入退室管理を行なうもので、利用者（通行者）の顔画像を認識して、その認識結果に基づき重要施設の入退室用ドアの開閉制御を行なうものであり、カメラ１０１、照明部１０２、表示部１０４、処理部１０５、および、認識部１１０の認識結果に応じて重要施設２０１の入退室用ドア２０２の開閉制御を行なう通行制御手段としてのドア制御部１１３から構成されている。
なお、図８において、ドア制御部１１３以外は図１の人物認識装置と同じ構成であるので、同一符号を付して、その説明は省略する。
【００５２】
認識部１１０は、たとえば、図４のステップＳＴ６において、求めた類似度が閾値よりも大きかった場合、あるいは、ステップＳＴ１１において、求めた類似度が閾値よりも大きかった場合、ドア制御部１１３に「ドア開」の信号を出力し、求めた類似度が閾値よりも小さかった場合、ドア制御部１１３に「ドア閉」の信号を出力する。
【００５３】
ドア制御部１１３は、認識部１１０から「ドア開」の信号を受取ると、入退室用ドア２０２を開状態に制御して、認識対象者（この場合は通行者）１００の入室を許可し、「ドア閉」の信号を受取ると、入退室用ドア２０２を閉状態に保持して、通行者１００の入室を拒否する。
このように、第２の実施の形態によれば、第１の実施の形態で説明した人物認識装置を用いて通行者の通行を制御することができる。
【００５４】
以上説明したように、上記実施の形態によれば、顔登録時、顔認識時に、認識対象者の身長により変化する最適な位置をあらかじめ表示したガイド線ｂ，ｃと、実際に検出した顔位置やサイズを示す円パターンｄを外接させるようなインタフェースを提供することにより、再現性のある立ち位置およびその結果である再現性のある画像を入力して照合することができる。その結果として、照合スコア（類似度）を高く保つことができる。
【００５５】
また、１対Ｎの照合で、複数の認識対象者に対して照合をかける際に、認識対象者の情報が特定できないため、顔登録時の顔の大きさや位置を正確には表示できないが、本実施の形態によれば、身長が決まった時点での推奨顔サイズが決定し、そのための円パターンｄのサイズを特定することができる。そのため、顔認識時に再現性のある画像を入力できる外、極端な姿勢の強制もないため、認識対象者にとっても使い勝手のよいユーザインタフェイスを提供できる。
【００５６】
なお、前記実施の形態では、ガイド線ｂ，ｃを表示部１０４によって表示する場合について説明したが、必ずしも表示部１０４により表示する必要はなく、別の表示手段で表示するようにしてもよい。たとえば、あらかじめ表面に任意の色でガイド線ｂ，ｃを描画してある透明なアクリル板（あるいはガラス板）を表示部１０４の画面上に配設するなどが考えられる。
また、表示部１０４のガイド線ｂ，ｃの内側に表示する環状パターンｄは円パターンに限らず、他の多角形パターンであってもよい。
【００５７】
【発明の効果】
以上詳述したように本発明によれば、再現性のある立ち位置およびその結果である再現性のある画像を入力して照合することができ、その結果として高い認識率を得ることができる人物認識装置、人物認識方法および通行制御装置を提供できる。
【００５８】
また、本発明によれば、認識時に再現性のある画像を入力できるとともに、極端な姿勢の強制もなく、利用者にとっても使い勝手のよい人物認識装置、人物認識方法および通行制御装置を提供できる。
【図面の簡単な説明】
【図１】本発明の第１の実施の形態に係る人物認識装置の構成を概略的に示すブロック図。
【図２】顔検出部の処理を説明するための図。
【図３】特徴量抽出部の処理を説明するための図。
【図４】認識部の認識処理の流れを説明するためのフローチャート。
【図５】表示部に対するガイド線および円パターンの表示例を示す図。
【図６】表示部に対するガイド線および円パターンの他の表示例を示す図。
【図７】ガイド線および円パターンの表示制御について説明するフローチャート。
【図８】本発明の第２の実施の形態に係る通行制御装置の構成を概略的に示すブロック図。
【符号の説明】
１００…認識対象者（人物）、１０１…カメラ（画像入力手段）、１０２…照明部、１０４…表示部（第１の表示手段、第２の表示手段）、１０５…処理部、１０６…画像入力部（画像入力手段）、１０７…顔検出部（顔検出手段）、１０８…特徴量抽出部（特徴量抽出手段）、１０９…登録情報保持部（登録情報保持手段）、１１０…認識部（認識手段）、１１１…表示制御部（第１の表示手段、第２の表示手段）、１１３…ドア制御部（通行制御手段）、２０１…重要施設、２０２…入退室用ドア、ｂ，ｃ…ガイド線、ｄ…円パターン（環状パターン）。[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a person recognizing device and a person recognizing method for recognizing a face image of a person in, for example, security management.
The present invention also relates to a traffic control device that controls the traffic of passers-by using the above-mentioned person recognition device.
[0002]
[Prior art]
In this type of person recognition device, the recognition rate of a person decreases when the state such as the standing position and posture of the person with respect to a camera for inputting a face image differs between registration and recognition (collation). . This is because the person's standing position was vaguely in front of the camera (or the display for displaying input images and operation guidance), but there was no strict check, and fluctuations at small levels could be suppressed. It is also because it was not.
[0003]
In addition, it is possible to store targets for registration and recognition (collation) for each individual, and in the one-to-one collation in which the person to be recognized is identified before recognition, it is easy to display the face size at the time of recognition at the time of recognition. It is. However, this method cannot be used for one-to-N matching in which the person to be recognized cannot be specified at the start of recognition.
[0004]
Furthermore, it is also possible to display the basic face size and position for one-to-N matching in a circular pattern or the like, but if this position is in the middle, a tall person will squat and a short person will stand out. Therefore, a stable face image cannot be input.
[0005]
[Problems to be solved by the invention]
In general, a person recognizing device using biometric information including a face image can obtain a high matching rate (recognition rate) if the same biometric information can be input at the time of registration and at the time of recognition. Furthermore, since the person recognition device using face images has a very natural shape (because it does not look into the device or touch the hand) compared to those using other biological information, the posture at the time of registration and at the time of recognition It is difficult to stabilize the standing position.
[0006]
Conventionally, since there is no input guide at the time of one-to-N matching, the standing position fluctuates (the standing position is shifted between registration and recognition), which leads to a decrease in recognition accuracy.
[0007]
In the one-to-N matching, since the recognition target cannot be specified in advance, when a certain input guide is displayed, the guide mismatch due to the height of the recognition target occurs as described above, and the accuracy error due to the posture is reduced. Will happen.
[0008]
Therefore, the present invention provides a person recognizing device and a person recognizing method capable of inputting and collating a reproducible standing position and a reproducible image as a result thereof, thereby obtaining a high recognition rate. And a traffic control device.
[0009]
Another object of the present invention is to provide a person recognizing device, a person recognizing method, and a traffic control device that can input a reproducible image at the time of recognition, do not require an extreme posture, and are easy to use for a user. And
[0010]
[Means for Solving the Problems]
The person recognizing device of the present invention comprises: image input means for inputting at least a face image of a person to be recognized; face detecting means for detecting a face area of the person from an image input by the image input means; A feature amount extraction unit that extracts a feature amount of the person's face based on a detection result of the detection unit; a registration information holding unit that holds the feature amount extracted by the feature amount extraction unit as registration information; Recognition means for recognizing the person by comparing the feature amount extracted by the means with the registration information held in the registration information holding means, and an optimal stand which changes with the height of the person with respect to the image input means. First display means for displaying a guide line for guiding the position, and an annular pattern indicating the position and size of the face obtained from the detection result of the face detection means, And and a second display means for displaying inside the guide lines displayed by stage.
[0011]
Further, the person recognition device of the present invention is an image input unit that inputs at least a face image of a person to be recognized, a face detection unit that detects a face area of the person from an image input by the image input unit, A feature amount extraction unit that extracts a feature amount of the face of the person based on a detection result of the face detection unit; a registration information holding unit that holds the feature amount extracted by the feature amount extraction unit as registration information; Recognition means for recognizing the person by comparing the feature amount extracted by the quantity extraction means with the registration information held in the registration information holding means, and an optimal means for changing the image input means depending on the height of the person. First display means for displaying a guide line for guiding a standing position; and an annular pattern indicating the position and size of the face obtained from the detection result of the face detection means. And and a second display means for displaying inside the guide line which is displayed by the display means.
[0012]
Further, the traffic control device of the present invention is a traffic control device that recognizes a traffic and controls the traffic of the traffic, wherein the image input means for inputting at least a face image of the traffic is input by the image input means. A face detection unit that detects the passer's face area from the extracted image; a feature amount extraction unit that extracts a feature amount of the passer's face based on a detection result of the face detection unit; and a feature amount extraction unit. A registration information holding unit for holding the extracted feature amount as registration information; and the passerby comparing the feature amount extracted by the feature amount extraction unit with the registration information held in the registration information holding unit. Recognizing means for recognizing a vehicle, a traffic control means for controlling the traffic of the passer in accordance with a recognition result of the recognizing means, and an optimal value which varies with the height of the passer to the image input means. First display means for displaying a guide line for guiding a standing position, and guide lines for displaying an annular pattern indicating the position and size of the face obtained from the detection result of the face detection means by the first display means And second display means for displaying the image on the inside.
[0013]
Further, the traffic control device of the present invention is a traffic control device for recognizing a passer and controlling the traffic of the passer, wherein the image input means for inputting at least a face image of the passer, and input by the image input means. A face detection unit that detects the passer's face area from the extracted image; a feature amount extraction unit that extracts a feature amount of the passer's face based on a detection result of the face detection unit; and a feature amount extraction unit. A registration information holding unit for holding the extracted feature amount as registration information; and the passerby comparing the feature amount extracted by the feature amount extraction unit with the registration information held in the registration information holding unit. Recognition means for recognizing the image, a traffic control means for controlling the traffic of the passer in accordance with the recognition result of the recognition means, First display means for displaying a guide line for guiding a standing position, and a guide for displaying an annular pattern indicating the position and size of a face obtained from the detection result of the face detection means by the first display means. Second display means for displaying the image inside the line.
[0014]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
First, a first embodiment will be described.
FIG. 1 schematically shows the configuration of the person recognition device according to the first embodiment. This person recognizing device includes a camera 101 as image input means for capturing and inputting a face image (an image including at least a face) of a person to be recognized (hereinafter also referred to as a person) 100, and a recognition target from the vicinity of the camera 101. A lighting unit 102 for irradiating light at a constant illuminance toward at least the face of the person 100, a display unit 104 as a display unit for displaying a face image and various kinds of guidance input to the recognition target person 100, The processing unit 105 is configured to process a face image input from the camera 101 and perform face image recognition processing and the like.
[0015]
In the following description, image input data, extracted feature amounts, subspaces, eigenvectors for configuring the subspaces, correlation matrices, registration time, date and time, location and other status information, passwords, ID codes, etc. Each piece of personal information appears. The term "recognition data" includes a subspace or an eigenvector for configuring the subspace, and the term "registration information" refers to image input data, an extracted feature, a subspace, and a subspace. Eigenvectors, correlation matrices, situation information, and personal information. Therefore, the recognition data is included in the registration information.
[0016]
The camera 101 captures and inputs a face image of the person to be recognized 100 from below at a predetermined angle, and is composed of, for example, a television camera using an image sensor such as a CCD sensor.
[0017]
The lighting unit 102 is a lighting device such as a fluorescent lamp that irradiates light at a constant illuminance from the vicinity of the camera 101 (for example, a side of the camera 101) toward the face of the recognition target 100.
[0018]
The display unit 104 is a monitor that displays a face image of the recognition target person 100 captured by the camera 101, various guidances, processing results, and the like. It is provided at a position corresponding to 100 face portions.
[0019]
The processing unit 105 uses an image input unit 106 as an image input unit and a face image of a person obtained from the image input unit 106 to match a face and face parts such as eyes and nose by matching with a prepared pattern dictionary. A face detection unit 107 as a face detection unit for detecting the position of the image, a feature amount extraction unit 108 as a feature amount extraction unit for extracting a feature amount such as grayscale information or partial space information based on the output of the face detection unit 107, At the time of registration, the feature amount extracted by the feature amount extracting unit 108 is stored as recognition information (registered information). A recognition unit 110 as a recognition unit that recognizes the face of the recognition target person 100 by comparing the amount with recognition data registered in the registration information holding unit 109; Beauty, and a display control unit 111 as display means for controlling the display unit 104.
[0020]
The image input unit 106 receives a face image from the camera 101, A / D converts it, digitizes it, and sends it to the face detection unit 107.
The face detection unit 107 first detects a face area of the recognition target person 100 from the face image input by the image input unit 106. For example, a location having the highest correlation value is determined as a face area by calculating a correlation value while moving a template prepared in advance in an image. In addition, a face detection unit such as a face region extraction method using an eigenspace method or a subspace method may be used.
[0021]
Next, the face detection unit 107 detects the positions of the face parts such as eyes, nose, and mouth from the detected face area. The detection method is described in, for example, literatures (Kazuhiro Fukui, Osamu Yamaguchi: "Face feature point extraction by combination of shape extraction and pattern matching", IEICE Transactions (D), vol. J80-D-II, No. 8, pp. 2170-2177 (1997)).
[0022]
Here, it is also possible to check the direction of the face based on the face detection using the face template and the relative positions of the detection coordinates of the face parts such as eyes and nose. For example, as shown in FIG. 2, the center coordinates of the face detection result are set to the center coordinates (fx, fy) of the face, the face size is set to (fw, fh), and the detection results of the left and right eyes and the nose are set to (EyeLx , EyeLy), (EyeRx, EyeRy), (NoseLx, NoseLy), and (NoseRx, NoseRy), an index indicating the direction of the face can be obtained by the following method.
Face direction (X direction) = ((EyeRx + EyeLx) / 2-fx) / fw
Face direction (Y direction) = ((EyeRy + EyeLy) / 2-fy) / fh
Assuming that the origin of these coordinates is the upper left of the screen, the numerical value becomes a value that becomes larger as the face turns to the lower right.
[0023]
The feature amount extraction unit 108 cuts out the face region into a certain size and shape based on the position of the face part detected by the face detection unit 107, and uses the density information as the feature amount. Here, for example, the gray value of an area of m pixels × n pixels is used as it is as information, and m × n-dimensional information is used as a feature vector.
[0024]
FIG. 3 shows face image data obtained by processing the input image by the feature amount extraction unit 108, and is obtained in time series. A subspace is calculated by obtaining a correlation matrix of a feature vector from these data and obtaining an orthonormal vector by a known KL expansion. The method of calculating the subspace calculates the subspace by obtaining the correlation matrix (or covariance matrix) of the feature vectors and obtaining the orthonormal vector (eigenvector) by KL expansion. In the subspace, k eigenvectors corresponding to the eigenvalues are selected in the descending order of the eigenvalues, and the subspaces are expressed using the eigenvector set.
[0025]
In the present embodiment, the correlation matrix Cd is obtained from the feature vector, and the correlation matrix Cd is obtained.

To obtain the eigenvector Φ. This subspace is used as a recognition dictionary for identifying a person. This information may be registered in advance and registered as a dictionary.
As will be described later, the subspace itself may be used as input data for performing recognition. Therefore, the result of calculating the subspace is sent to the recognition unit 110 and the registration information holding unit 109.
[0026]
The recognition unit 110 compares (recognizes) the recognition data (subspace) stored in the registration information holding unit 109 with the feature amount (shade information or subspace information) obtained by the feature amount extraction unit 108, The recognition target person 100 shown in the camera 101 is recognized or identified as the corresponding person. In order to recognize a person, it is sufficient to determine which person's data is most similar, and the person corresponding to the data having the highest similarity may be used as the recognition result.
[0027]
In addition, when performing recognition using a face image using a card, a registration number, a personal identification number, a key, and the like, the similarity with the recognition data corresponding to personal information such as a card and a number of each person is calculated. Is compared with the set threshold value, and when the threshold value is exceeded, the person is identified.
[0028]
Means for inputting such a card, a registration number, a password, a key, and the like are required. For example, a card can be realized by using an IC card, an ID card, a wireless card, or the like. When a registration number or a password is used, key input means may be provided.
[0029]
When information of the feature amount extraction unit 108 is used as a recognition method, a method such as a subspace method or a composite similarity method may be used. The recognition method in the present embodiment is described in, for example, literatures (Kenichi Maeda, Seiichi Watanabe: “Pattern matching method with local structure introduced”, IEICE Transactions (D), vol. J68-D, No. ., Pp. 345-352 (1985)), which is a known subspace method, which is a known recognition method.
[0030]
In this method, both recognition data and input data in registration information stored in advance are expressed as subspaces, and an “angle” formed by two subspaces is defined as a similarity. The subspace input here is called an input subspace. Similarly, a correlation matrix Cin is obtained for the input data sequence,
(Equation 2)

To obtain the eigenvector Φin. The subspace similarity (0.0 to 1.0) of the subspace represented by the two Φin and Φd is obtained, and is set as the similarity for recognizing it.
[0031]
The recognizing unit 110 operates as shown in the flowchart of FIG. First, the operation of the recognizing unit 110 differs depending on whether a recognition operation or an identification operation is performed (step ST1). When performing the identification operation, first, the ID code of the target recognition target person is read (step ST2). Next, the registration information (partial space) corresponding to the target ID code is read from the registration information holding unit 109 (step ST3).
[0032]
Next, as described above, in order to perform recognition by the subspace method or the like, the similarity between the subspace of each registered information and the input vector (eigenvector from the feature amount extraction unit 108) or the input subspace is calculated. (Step ST4). Next, the similarity is compared with a preset threshold value (steps ST5 and ST6), and an identification result is output (step ST7).
[0033]
When performing the recognition operation, all the data to be recognized are read from the registration information holding unit 109 (step ST8). Then, the degree of similarity with each piece of registration information is calculated (step ST9). Next, the largest one of the calculated similarities is selected (step ST10), and is output as a recognition result (step ST12).
[0034]
As in step ST11 surrounded by a broken line in FIG. 4, it is also possible to verify whether or not the recognition result is correct by determining the maximum similarity as a threshold (step ST13). For example, if the similarity is too low, it can be determined that the target is not any recognition target.
[0035]
The registration information holding unit 109 can accumulate a partial space (or a correlation matrix or the like) used for identifying a recognition target person, status information such as time, date, and place at the time of registration. In addition, image input data or extracted feature amounts may be used.
[0036]
With the above-described processing, it is possible to perform a basic person recognition operation. However, as described above, a change in the standing position or posture of the recognition target person causes a reduction in the recognition rate. For example, if a face is present in the image at the time of recognition (at the time of verification) in the same state as at the time of registration, the similarity with the registered dictionary increases because feature information can be extracted in a state similar to that at the time of registration. The degree of similarity is likely to decrease when the face is larger or smaller than at the time of registration, or when the face direction is not shown in the drawing but is different.
[0037]
Therefore, in the present embodiment, the display control unit 111 and the display unit 104 not only display, for example, a moving image a of the face of the recognition target person being photographed as shown in FIG. And at the time of registration) a face input guide line (two straight lines) b composed of two straight lines arranged in a trapezoidal shape as a guide for allowing the person to be recognized to stand naturally in the center of the field of view of the camera 101. , C are always displayed, and an annular pattern for face input, for example, a circular pattern d, is displayed inside the guide lines b, c as a guide indicating the face position and size actually detected.
FIG. 5A shows a display example in a case where the recognition target person has a standard height, and FIG. 5B shows a display example in a case where the recognition target person is tall.
[0038]
The guide lines b and c are lines circumscribed to the average face input guide circular pattern d (the diameter is estimated from the standard data) obtained using standard data. The reason why the guide lines b and c are not vertical and have a trapezoidal oblique angle is that the standing position where the person to be recognized can be captured by the camera 101 differs depending on the height of the recognition target person (a person with a high height is placed behind. Standing, the face is in a high position, a short person stands in front, and the face is in a low position), to indicate that the size of the circular pattern d changes relatively depending on the height of the face. It is.
[0039]
In this case, the guide lines b and c are displayed or, as shown in FIG. 6, a display area (shaded portion) other than the guide lines b and c is not displayed and the area is masked (the area is white or gray). Is also effective. In this case, it is also effective in the following point not to display a portion slightly outside the guide lines b and c. That is, it is assumed that the recognition target person strictly specifies the face size because it is assumed that the usability is deteriorated and that the recognition target cannot be tolerated as a problem of accuracy, and a slight variation in the face size is allowed.
[0040]
Also, since the slopes of the guide lines b and c (defined by the upper width and the lower width) are composed of average data, in addition to the standard values as devices, the actually registered data May be set based on
[0041]
A circular pattern d as a face input guide for keeping the size of the face constant is drawn with the center position and diameter of the face calculated from the positions of the eyes and nose actually detected by the face detection unit 107. You. This display control is performed by the display control unit 111.
[0042]
Hereinafter, the display control of the guide lines b and c and the circular pattern d will be described in detail with reference to the flowchart shown in FIG.
First, the apparatus always displays the guide lines b and c as shown in FIG. 5 on the display unit 104 (step S11).
On the other hand, the person to be recognized 100 stands in front of the apparatus (camera 101) and visually checks the screen of the display unit 104 (step S21) to check whether or not the circle pattern d is displayed on the display unit 104. If the circle pattern d is not displayed (step S22), the user moves the face or the standing position to display the circle pattern d (step S23).
[0043]
On the other hand, on the present apparatus side, the face detection unit 107 detects a face area from the image captured by the camera 101, and calculates the center coordinates and the size (size of the face) (step S12). Next, it is checked whether or not the detected face size is ± 10% of the predetermined Y coordinate (step S13). If not, the process returns to step S12 and repeats the same operation as described above.
[0044]
As a result of the check in step S13, if the detected face size is ± 10% of the predetermined Y coordinate, as shown in FIG. 5, inside the guide lines b and c displayed on the display unit 104, A circle pattern d having a diameter corresponding to the detected face size is displayed at a height corresponding to the height of the recognition target person 100 (that is, the height at which the eyes and nose are detected) (step S14). That is, if the face size does not reach a certain ratio (about ± 10%) with respect to the standard guide lines b and c, control is performed so that the display of the circular pattern d and the collation processing described below are not performed. is there.
[0045]
In this case, it is also possible to instruct the recognition target person 100 to change the standing position to a position where the circle pattern d is in contact with the guide lines b and c by text or voice.
[0046]
Next, as described above, the recognizing unit 110 collates the feature amount obtained by the feature amount extracting unit 108 with the recognition data stored in the registration information holding unit 109, thereby obtaining the similarity between the two. Step S15). Next, it is checked whether or not the obtained similarity is equal to or greater than a preset threshold value (step S16). If the similarity is equal to or greater than the threshold value, it is determined that the person to be recognized 100 is himself, and the recognition process ends.
[0047]
As a result of the check in step S16, if the obtained similarity is not equal to or greater than the threshold, it is checked whether or not a specified time has elapsed since the start of the recognition processing (step S17). Returning to step S12, the same operation as above is repeated, and if the specified time has elapsed, the recognition processing is ended here.
[0048]
In the above description, the display of the guide lines b and c and the circular pattern d has been described by taking face recognition as an example. However, the guide lines b and c and the circular pattern d are also displayed at the time of face registration. Since the display control is the same as that for the face recognition, the description is omitted.
[0049]
In this way, by displaying the guide lines b and c and the circular pattern d on the display unit 104 during face recognition and face registration, an unspecified number of persons to be recognized 100 having different heights and face sizes can be recognized. The face can be presented at an appropriate and reproducible position. In addition, since the input image can be fixed, the matching score (similarity) can be increased.
[0050]
Next, a second embodiment will be described.
The second embodiment is an example in which the person recognition device described in the first embodiment is applied to a traffic control device that recognizes a passer's face image and controls the traffic of the passer.
[0051]
FIG. 8 schematically illustrates a configuration of a traffic control device according to the second embodiment. This traffic control device performs, for example, entry / exit management to an important facility (a room with an emphasis on security), recognizes a face image of a user (passer), and enters an important facility based on the recognition result. The opening / closing control of the exit door is performed, and the opening / closing control of the entrance / exit door 202 of the important facility 201 is performed according to the recognition result of the camera 101, the illumination unit 102, the display unit 104, the processing unit 105, and the recognition unit 110. And a door control unit 113 as a traffic control means for performing the operation.
Note that, in FIG. 8, since the configuration other than the door control unit 113 is the same as that of the person recognition device in FIG.
[0052]
For example, when the obtained similarity is larger than the threshold in step ST6 of FIG. 4, or when the obtained similarity is larger than the threshold in step ST11, the recognizing unit 110 notifies the door control unit 113 of “ A signal of “door open” is output, and if the calculated similarity is smaller than the threshold, a signal of “door closed” is output to the door control unit 113.
[0053]
Upon receiving the signal of “door open” from the recognizing unit 110, the door control unit 113 controls the entrance / exit door 202 to an open state to permit the recognition target person (passer in this case) 100 to enter the room. When the signal of “door closed” is received, the entrance / exit door 202 is kept in a closed state, and the entrance of the passerby 100 is rejected.
As described above, according to the second embodiment, it is possible to control the passage of a pedestrian using the person recognition device described in the first embodiment.
[0054]
As described above, according to the above-described embodiment, at the time of face registration and face recognition, guide lines b and c in which optimum positions that change according to the height of the recognition target person are displayed in advance, and the actually detected face position By providing an interface for circumscribing the circle pattern d indicating the size and size, a reproducible standing position and the resulting reproducible image can be input and collated. As a result, the matching score (similarity) can be kept high.
[0055]
In addition, when performing matching with respect to a plurality of recognition targets in one-to-N matching, information on the recognition targets cannot be specified, so that the size and position of the face at the time of face registration cannot be displayed accurately. According to the present embodiment, the recommended face size when the height is determined can be determined, and the size of the circular pattern d for that can be specified. Therefore, since a reproducible image can be input at the time of face recognition, and no extreme posture is forced, a user interface that is easy to use for the person to be recognized can be provided.
[0056]
In the above-described embodiment, the case where the guide lines b and c are displayed on the display unit 104 has been described. However, the guide lines need not necessarily be displayed on the display unit 104, and may be displayed on another display unit. For example, a transparent acrylic plate (or a glass plate) in which guide lines b and c are drawn in an arbitrary color on the surface in advance may be provided on the screen of the display unit 104.
Further, the annular pattern d displayed inside the guide lines b and c of the display unit 104 is not limited to the circular pattern, but may be another polygon pattern.
[0057]
【The invention's effect】
As described above in detail, according to the present invention, it is possible to input and collate a reproducible standing position and a reproducible image as a result, and as a result, a person who can obtain a high recognition rate A recognition device, a person recognition method, and a traffic control device can be provided.
[0058]
Further, according to the present invention, it is possible to provide a person recognizing device, a person recognizing method, and a traffic control device that can input a reproducible image at the time of recognition, do not require an excessive posture, and are easy for a user to use.
[Brief description of the drawings]
FIG. 1 is a block diagram schematically showing a configuration of a person recognition device according to a first embodiment of the present invention.
FIG. 2 is a view for explaining processing of a face detection unit.
FIG. 3 is a view for explaining processing of a feature amount extraction unit;
FIG. 4 is a flowchart illustrating the flow of a recognition process performed by a recognition unit.
FIG. 5 is a diagram showing a display example of a guide line and a circle pattern on a display unit.
FIG. 6 is a diagram showing another display example of a guide line and a circle pattern on a display unit.
FIG. 7 is a flowchart illustrating display control of a guide line and a circle pattern.
FIG. 8 is a block diagram schematically showing a configuration of a traffic control device according to a second embodiment of the present invention.
[Explanation of symbols]
100: person to be recognized (person), 101: camera (image input means), 102: illumination unit, 104: display unit (first display unit, second display unit), 105: processing unit, 106: image input Unit (image input unit), 107: face detection unit (face detection unit), 108: feature amount extraction unit (feature amount extraction unit), 109: registration information holding unit (registration information holding unit), 110: recognition unit (recognition) Means), 111: display control unit (first display means, second display means), 113: door control unit (passage control means), 201: important facility, 202: door for entrance and exit, b, c: guide Line, d: circular pattern (annular pattern).

Claims (13)

Image input means for inputting at least a face image of a person to be recognized;
Face detection means for detecting the face area of the person from the image input by the image input means,
Feature value extracting means for extracting a feature value of the face of the person based on the detection result of the face detecting means;
Registration information holding means for holding the feature amount extracted by the feature amount extraction means as registration information,
Recognition means for recognizing the person by comparing the feature quantity extracted by the feature quantity extraction means with the registration information held in the registration information holding means,
First display means for displaying a guide line that guides an optimal standing position that changes according to the height of the person with respect to the image input means,
Second display means for displaying an annular pattern indicating the position and size of the face obtained from the detection result of the face detection means inside a guide line displayed by the first display means;
A person recognition device characterized by comprising: Image input means for inputting at least a face image of a person to be recognized;
Face detection means for detecting the face area of the person from the image input by the image input means,
Feature value extracting means for extracting a feature value of the face of the person based on the detection result of the face detecting means;
Registration information holding means for holding the feature amount extracted by the feature amount extraction means as registration information,
Recognition means for recognizing the person by comparing the feature quantity extracted by the feature quantity extraction means with the registration information held in the registration information holding means,
First display means for displaying a guide line that guides an optimal standing position that changes according to the height of the person with respect to the image input means,
Second display means for displaying an annular pattern indicating the position and size of the face obtained from the detection result of the face detection means inside a guide line displayed by the first display means;
A person recognition device characterized by comprising: The person recognition device according to claim 1, wherein the guide line includes two straight lines arranged in a trapezoidal shape and is always displayed. 3. The person recognition apparatus according to claim 1, wherein the annular pattern is a circular pattern drawn with a line having a predetermined thickness, and is displayed in a state of being in contact with the inside of the guide line. . When displaying the annular pattern by the second display means, a guide instruction means for instructing the person by text or voice to change the standing position to a position where the annular pattern is in contact with the inside of the guide line is provided. 3. The person recognition device according to claim 1, further comprising: An image input step of inputting at least a face image of a person to be recognized;
A face detection step of detecting a face area of the person from the image input by the image input step;
A feature value extraction step of extracting a feature value of the face of the person based on the detection result of the face detection step;
A registration information holding step of holding the feature amount extracted by the feature amount extraction step as registration information in a registration information holding unit;
A recognition step of recognizing the person by comparing the feature amount extracted in the feature amount extraction step with registration information held in the registration information holding means,
A first display step of displaying a guide line that guides an optimal standing position that changes according to the height of the person with respect to the image input step;
A second display step of displaying an annular pattern indicating the position and size of the face obtained from the detection result of the face detection step inside a guide line displayed by the first display step;
A person recognition method characterized by comprising: An image input step of inputting at least a face image of a person to be recognized;
A face detection step of detecting a face area of the person from the image input by the image input step;
A feature value extraction step of extracting a feature value of the face of the person based on the detection result of the face detection step;
A registration information holding step of holding the feature amount extracted by the feature amount extraction step as registration information in a registration information holding unit;
A recognition step of recognizing the person by comparing the feature amount extracted in the feature amount extraction step with registration information held in the registration information holding means,
A first display step of displaying a guide line that guides an optimal standing position that changes according to the height of the person with respect to the image input step;
A second display step of displaying an annular pattern indicating the position and size of the face obtained from the detection result of the face detection step inside a guide line displayed by the first display step;
A person recognition method characterized by comprising: When displaying the annular pattern by the second display step, a guidance instruction step of giving a guidance instruction by text or voice to the person to change the standing position to a position where the annular pattern is in contact with the inside of the guide line is provided. The method according to claim 6 or 7, further comprising: In a traffic control device that recognizes a pedestrian and controls the traffic of the pedestrian,
Image input means for inputting at least a face image of the passer-by,
Face detection means for detecting the passer's face area from the image input by the image input means,
Feature value extracting means for extracting a feature value of the passer-by's face based on the detection result of the face detecting means;
Registration information holding means for holding the feature amount extracted by the feature amount extraction means as registration information,
Recognition means for recognizing the passer-by by comparing the feature quantity extracted by the feature quantity extraction means with the registration information held in the registration information holding means,
A traffic control means for controlling the traffic of the passer in accordance with the recognition result of the recognition means,
First display means for displaying a guide line that guides an optimal standing position for the image input means that changes according to the height of the passer-by;
Second display means for displaying an annular pattern indicating the position and size of the face obtained from the detection result of the face detection means inside a guide line displayed by the first display means;
A traffic control device comprising: In a traffic control device that recognizes a pedestrian and controls the traffic of the pedestrian,
Image input means for inputting at least a face image of the passer-by,
Face detection means for detecting the passer's face area from the image input by the image input means,
Feature value extracting means for extracting a feature value of the passer-by's face based on the detection result of the face detecting means;
Registration information holding means for holding the feature amount extracted by the feature amount extraction means as registration information,
Recognition means for recognizing the passer-by by comparing the feature quantity extracted by the feature quantity extraction means with the registration information held in the registration information holding means,
A traffic control means for controlling the traffic of the passer in accordance with the recognition result of the recognition means,
First display means for displaying a guide line that guides an optimal standing position for the image input means that changes according to the height of the passer-by;
Second display means for displaying an annular pattern indicating the position and size of the face obtained from the detection result of the face detection means inside a guide line displayed by the first display means;
A traffic control device comprising: The traffic control device according to claim 9, wherein the guide line is formed of two straight lines arranged in a trapezoidal shape and is always displayed. The traffic control device according to claim 9, wherein the annular pattern is a circular pattern drawn by a line having a predetermined thickness, and is displayed in a state of being in contact with the inside of the guide line. . Guidance instructing means for instructing the passerby by text or voice so as to change the standing position to a position where the annular pattern contacts the inside of the guide line when the annular pattern is displayed by the second display means. The traffic control device according to claim 9 or 10, further comprising:

Images