JP4359939B2 - Image measuring device - Google Patents

Description

式の加算において、Ｒ（ａ，ｂ）の値が過去の残差の最小値を越えたら加算を打ち切り、次の（ａ，ｂ）に移るよう計算処理を行うことにより、処理の高速化をはかることができる。
【００５９】
３．Ｒ（ａ，ｂ）最小かつ隣接するマーク位置の間隔等の条件により、投影特徴パターンのマーク位置をマーク数分決定し、位置座標とする。
【００６０】
（詳細位置決定）
１．探索領域決定
前述の（概略位置検出）によって決まった点に対し、そこを中心とした探索領域を設定する。
【００６１】
２．マーク領域の決定
探索領域について、濃度値がしきい値以上の領域を一つのマークとする（図８参照）。
【００６２】
しきい値は、ある適当な値を予め決めておく。画像は、マーク以外は差分処理を施しているため殆ど０となっている。
【００６３】
３．重心位置算出
モーメント法を行い、重心位置座標を算出する。
【００６４】
［モーメント法］
図８に示されている通り、しきい値Ｔ以上の点について（マークＫ）、以下の式を施す。
【００６５】
【数２】

【００６６】
【数３】

この数式２及び数式３によって、サブピクセル位置まで重心位置が算出可能となる。
【００６７】
４．点数分だけ１〜３を行う。
【００６８】
以上で、図５あるいは図６のような特徴パターン投影によるマーク像の位置座標を算出することができる。
【００６９】
概略位置検出を行わずに、最初から詳細位置検出を行っても良く、また、他のアルゴリズムによって位置検出しても構わない。
【００７０】
いずれにしても，画像は特徴パターン情報のみなので、高速かつ、高精度に位置を算出することが可能である。
【００７１】
ｃ．標定
次に，特徴パターン位置検出部５５で求めた座標値を姿勢算出部６に送り、標定計算を行う。
【００７２】
この計算により、左右それぞれのカメラの位置等が求めるられる。
【００７３】
次の共面条件式により、それらのパラメータを求める。
【００７４】
【数４】

モデル座標系の原点を左側の投影中心にとり、右側の投影中心を結ぶ線をＸ軸にとるようにする。縮尺は、基線長を単位長さにとる。このとき求めるパラメータは、左側のカメラのＺ軸の回転角κ１、Ｙ軸の回転角φ１、右側のカメラのＺ軸の回転角κ２、Ｙ軸の回転角φ２、Ｘ軸の回転角ω２の５つの回転角となる。この場合、左側のカメラのＸ軸の回転角ω１は０なので、考慮する必要ない。
【００７５】
このような条件にすると、数式４の共面条件式は数式５のようになり、この式を解けば各パラメータが求まる。
【００７６】
【数５】

ここで、モデル座標系ＸＹＺとカメラ座標系ｘｙｚの間には、次に示すような座標変換の関係式が成り立つ。
【００７７】
【数６】

【００７８】
【数７】

これらの数式を用いて、次の手順により、未知パラメータを求める。
１．初期近似値は通常０とする。
２．共面条件式５を近似値のまわりにテーラー展開し、線形化したときの微分係数の値を数式６、７により求め、観測方程式をたてる。
３．最小二乗法をあてはめ、近似値に対する補正量を求める。
４．近似値を補正する。
５．補正された近似値を用いて、上記２．〜５．までの操作を収束するまで繰り返す。
【００７９】
仮に、どこかのマークが対象物の形状等で、うまく位置検出されていない場合には、収束しない場合がありうる。その場合は、１点１点削除して、上記１〜５までを行い、収束したもの、あるいは、一番良いもののパラメータを使用する。
【００８０】
ｄ．ステレオモデル形成
次に、標定により求められたパラメータにより、立体視可能なステレオモデル座標系へ画像を変換し、ステレオモデルを形成する。
【００８１】
モデル座標への変換式は以下のようになる。
【００８２】
【数８】

【００８３】
【数９】

【００８４】
【数１０】

このようにして、三次元計測が可能な状態となる。
【００８５】
ｅ．三次元計測
（１）ポイント計測
図６のような６点以上の特徴パターン２０を使用し、複数のマークを位置検出（ｂ）と同様な処理をすることにより、その三次元座標を自動で高精度に求められる。
（２）面計測
図６のような投影特徴パターン２０を使用することにより、ステレオマッチングの初期値とすることが可能となる。
【００８６】
面計測の場合の初期値の決めかたについて説明する。
【００８７】
図９と図１０は説明図であり、図６のような投影特徴パターンのうちの一部について抜き出したものである。図９は基準とする基準画像、図１０は探索を行う捜索画像である。これら基準・捜索画像は、左右どちらの画像でもよい。例えば、左画像を基準画像、右画像を捜索画像と決める。
【００８８】
投影特徴パターンは基準画像Ｓ’１、Ｓ’２、Ｓ’５、Ｓ’６に対し捜索画像上ではＳ１、Ｓ２、Ｓ５、Ｓ６が対応している。
【００８９】
対応点探索は、基準画像中の基準データブロックをテンプレート画像として、捜索画像中の捜索領域にステレオマッチングを施すことにより行う。ステレオマッチングには画像相関処理等を使用する。
【００９０】
［捜索領域の決定］
投影特徴パターンをもとに、捜索領域を決定する方法について説明する。
【００９１】
投影特徴パターンＳ１〜Ｓ２（水平方向）を幅Ａ、Ｓ１〜Ｓ５（垂直方向）を幅Ｂとし、Ｓ１，Ｓ２、Ｓ５，Ｓ６で囲まれた部分を捜索領域Ｒ１とする。以下同様にＳ２，Ｓ３、Ｓ６，Ｓ７で囲まれた部分を捜索領域Ｒ２、…と順次それぞれの投影特徴パターンで囲まれた部分を捜索領域として決定する。
【００９２】
実際の投影特徴パターンによるマーク像は、それぞれ図１１に示されるように、対象物の形状やＣＣＤの向きによって同一ライン上にあるとは限らない。そこで、各投影特徴パターンの４点から最大のＡ、Ｂの幅が得られるような四角形を作成し、捜索領域とする。例えば図１１では、Ｓ１，Ｓ２，Ｓ５，Ｓ６でとりうる最大の幅は水平方向Ａ１，垂直方向Ｂ１となり、これで作成した四角形の領域を捜索領域Ｒ１、同様に、Ｓ２，Ｓ３，Ｓ６，Ｓ７でとりうる最大幅Ａ２，Ｂ２で作成される四角形を捜索領域Ｒ２とする。このようにすると、多少オーバーラップする領域ができるが、Ｒ１、Ｒ２領域の境界も確実に探索が可能となる。
【００９３】
これは、既に求めた捜索画像中のマーク像の位置に基づいて、捜索画像に捜索領域の範囲を決定するものである。
【００９４】
図１２はこの捜索領域決定の変形例である。
【００９５】
例えば、それぞれの投影特徴パターンは、左右画像上ですでに対応点として求められているので、捜索領域の捜索開始位置と終了位置を投影点近隣の領域とすると効率がよい。すなわち、図１１の例だと、垂直方向にＳ１からＳ５まで捜索幅Ａ１づつ探索を進めていくと、Ｓ５のライン周辺に近づくにつれ、無駄な探索領域（明らかに対応点が存在しない領域）がでる。
【００９６】
そこで、図１２のＳ１〜Ｓ５間（垂直方向）の１／２のＤ１までは、Ｓ１と同一の水平方向の点を捜索領域の開始点とし、Ｄ１〜Ｓ５間ではＤ１とＳ５を結んだ線上を水平方向の捜索開始点にとる。
【００９７】
このような処理を各投影点につき捜索領域として設定し、捜索領域を多少オーバーラップさせながらステレオマッチングを行えば、ある程度効率のよい探索が可能となる。
【００９８】
図１３、図１４は、捜索領域中の捜索データブロックを設定することにより、さらに捜索領域の探索を効率化したものである。
【００９９】
たとえば、捜索画像中のマークＳ１〜Ｓ２のＡ１の区間において、図９で示される基準画像上の基準データブロックＴ１、Ｔ２、Ｔ３…、それぞれに対応する捜索領域上の位置を求めるとき、それぞれの捜索データブロックを図１３で示されるように、Ｕ１、Ｕ２、Ｕ３、…、と何ブロックかにわけて設定する。このようにすれば、捜索領域探索における時間を短かくし効率的に探索できる。この場合、捜索データブロックの範囲は、基準データブロックをいくつにとるかで決定できる。例えば基準画像Ｓ’１〜Ｓ’２のＡ’間でｎ個基準データブロックを設定するとすれば、Ａ１／ｎ、あるいはＡ１／（ｎ−１）等とすれば良い。（ｎ−１）としたのは、多少捜索データブロックをオーバーラップさせて探索を行う場合である。
【０１００】
これは、既に求めた捜索画像中のマーク像の位置及び捜索領域の位置に基づいて、捜索画像に捜索領域の範囲を設定するものである。見方を変えると基準画像中のマーク像の位置及び基準データブロック位置との位置関係に基づいて、捜索画像に捜索領域の範囲を設定するものともいえる。
【０１０１】
図１３の方式の変形例が図１４に示されている。これは、捜索データブロックのサイズを可変としたものである。すなわち、マークＳ１、Ｓ２の近隣領域は、基準データブロックに対する対応点が近くにあることがわかっているので、捜索データブロックのサイズを小さく、離れるにしたがって対応点位置が不確かとなるため大きくするものである。従って、基準データブロックがＡ’／２の位置で最大の捜索データブロックサイズとなる。
【０１０２】
これら捜索データブロックのサイズは、Ｓ１〜Ｓ２の捜索領域Ａ１より決定する。
【０１０３】
例えば、ｎをＡ’間の基準データブロック数とすれば、
・Ｓ１〜Ｓ１＋Ａ１／２の区間の捜索データブロックサイズ：
（１＋ｔ×ｉ／ｎ）×Ａ１／ｎ、
・Ｓ１＋Ａ１／２〜Ｓ２の区間の捜索データブロックサイズ：
（１＋ｔ×（ｎ−ｉ）／ｎ）×Ａ１／ｎ、
但し、ｉはそれぞれの基準データブロックに対応する捜索データブロックの位置、すなわちｉ＝１〜ｎとする。また、ｔは倍率の定数で所定の値に設定する。例えばＳ１＋Ａ１／２の位置でＳ１の位置（Ｕ１）の捜索データブロックサイズの倍にしたければ２を選ぶ。
【０１０４】
これら捜索データブロックサイズは、基準画像中のマークＳ’１、Ｓ’２及び基準データブロックＴ１，Ｔ２、…、に基づいて決定してもよい。その際は、上述のＡ１をＡ’とし、倍率Ａ１／Ａ’を加味した項を掛け合わせる。
【０１０５】
このように、捜索領域及び捜索データブロックを設定することにより、対応点探索を効率よく、すなわち高速かつ信頼性を高めながら行うことが可能となる。
【０１０６】
この方式では、捜索画像中のマーク像の位置及び捜索領域の位置との位置関係に基づいて、見方を変えれば基準画像中のマーク像の位置及び基準データブロック位置との位置関係に基づいて、捜索画像に捜索領域の範囲の大きさを設定している。
【０１０７】
［基準データブロックの決定］
基準となる基準データブロックは、基準画像上のＳ’１、Ｓ’２、Ｓ’５、Ｓ’６、あるいは捜索画像上のＳ１、Ｓ２、Ｓ５、Ｓ６から決定する。例えば基準画像（図９）Ｓ’１〜Ｓ’２をＡ’、Ｓ’１〜Ｓ’５をＢ’とすれば、水平方向の基準データブロック幅はＡ’／ｎ、垂直方向はＢ’／ｍのように決定できる。
【０１０８】
あるいは、左右画像の比率に比例した大きさにしてもよい。例えば、水平方向はＡ’／Ａ＊ｎ、垂直方向は、Ｂ’／Ｂ＊ｍのように設定する。
【０１０９】
ｎ、ｍは計測したい対象物の大きさと画素数の関係によって適切な大きさを求めることができるが、Ａ’、Ｂ’の値によって適宜定数としても良い。
【０１１０】
図１５は基準データブロックサイズを１種類でなく３種類として相関積をとりながら、上述の処理と同様にステレオマッチングしていく方法である。この場合も、この３種類のサイズをＡ’，Ｂ’の情報をもとに決定できる。
【０１１１】
図１６は、本発明の基準データブロック決定における更なる変形例である。Ｓ’１及びＳ’２近隣の領域は、基準・捜索画像上で比較的対応が取れているために基準データブロックは小さくてよいが、離れるにしたがって、対応位置が不確かとなるため基準データブロックサイズを動的に変化させる。たとえば、Ｔ１位置、Ｔ２位置、Ｔ３位置において図１４のように基準データブロックを拡大していく。そして、基準データブロック位置がＡ’／２地点で最大の大きさとする。Ａ’／２からＳ’２に向かっては、基準データブロックサイズを逆に順次小さくしていく。
【０１１２】
このようにすることにより、ステレオマッチングの信頼性を高めることができる。
【０１１３】
これら基準データブロックサイズは、Ｓ’１〜Ｓ’２の水平方向の幅Ａ’、Ｓ’１〜Ｓ’５の垂直方向の幅Ｂ’より決定する。
【０１１４】
例えば、ｎ，ｍをＡ’、Ｂ’間の基準データブロック数とすれば、
・Ｓ’１〜Ｓ’１＋Ａ’／２の区間の基準データブロックサイズ：
水平方向：（１＋ｔ×ｉ／ｎ）×Ａ’／ｎ、
垂直方向：（１＋ｔ×ｌ／ｍ）×Ｂ’／ｍ
・Ｓ’１＋Ａ’／２〜Ｓ’２の区間の基準データブロックサイズ
水平方向：（１＋ｔ×（ｎ−ｉ）／ｎ）×Ａ’／ｎ、
垂直方向：（１＋ｔ×（ｍ−ｉ）／ｍ）×Ｂ’／ｍ
但し、ｉ，ｌはそれぞれの基準データブロック位置、すなわちｉ＝１〜ｎ、ｌ＝１〜ｍとする。
【０１１５】
ここで、ｔは倍率の定数であり、所望の値に設定する。例えばＳ’１＋Ａ’／２の位置でＳ’１の位置（Ｔ１）の基準データブロックサイズの倍にしたければ２を選ぶ。
【０１１６】
このようにすれば、基準データブロックを可変にできる。
【０１１７】
また、図１５のような３種類の基準データブロックを上述の例のように可変とし、相関積をとることにより、ステレオマッチングしていけば、更に信頼性の高い対応点探索が可能となる。
【０１１８】
これら基準データブロックの決定は、捜索画像におけるマークＳ１，Ｓ２によって同様に行ってもよい。その際は、Ａ’をＡ１、Ｂ’をＢ１として更に基準画像と捜索画像の倍率Ａ’／Ａ１を加味した項を掛け合わせる。
【０１１９】
以上のようにすることによって、各初期値が自動で求められる。
【０１２０】
次に、実際に行う計測（ステレオマッチング）の手順について説明する。
【０１２１】
例として図６の投影特徴パターンＳ１〜Ｓ８の場合について、図９、１１により説明する。
【０１２２】
垂直方向は標定処理が終了し、縦視差が除去されている（位置合わせしてある）ので、基準画像、捜索画像の同一ライン上を探索するだけで良い。また捜索データブロックを設定する際は、各捜索領域において上述のように適宜設定して行う。
【０１２３】
１．捜索画像のＳ１〜Ｓ２までの捜索領域Ｒ１の水平方向のラインＬ１の探索幅Ａ１に対し、
基準画像中のＴ１、Ｔ２、Ｔ３、…の基準データブロックで順次対応点探索を行う。
【０１２４】
これら初期値は、先に述べたような方法で決定しておく。
【０１２５】
捜索領域Ｒ１のラインＬ１上のＡ１が終了したら、
２．Ｓ２〜Ｓ３の捜索領域Ｒ２に対し、この領域で決められた基準データブロック位置からラインＬ１のＡ２上の対応点探索を順次繰り返す。
【０１２６】
３．次に、Ｓ３〜Ｓ４の捜索領域Ｒ３のラインＬ１のＡ３を対応点探索する。捜索領域Ｒ３に対しては、この領域で決められたテンプレートサイズ、位置で順次行う。
【０１２７】
ラインＬ１の対応点探索が終了したら、次のラインＬ２に移動して、また、捜索領域Ｒ１の探索幅Ａ１から、対応点探索を１〜３と同様に繰り返す。
【０１２８】
これを必要なライン数分繰り返す。
【０１２９】
以上のようにすれば、投影特徴パターンの位置から、ステレオマッチングする際の初期値、すなわち、捜索領域、捜索データブロック、および基準データブロックを自動で決定し、自動計測が可能となる。
【０１３０】
さらに、基準・捜索画像の捜索領域と基準データブロックをＡ’、Ｂ’以内と限定できるため、通常のステレオマッチングよりはるかに高速な処理が可能となる。すなわち、通常は、各探索位置において１水平ライン分（図６、line）ステレオマッチングを行うのであるが、その処理時間が投影点の数により数分の１となる。
【０１３１】
また、基準・捜索画像上の対応領域があらかじめ限定され求められているため、ミスマッチングが大幅に減少できる。すなわち、ステレオマッチングの信頼性を大幅に向上させることができる。
【０１３２】
これに加えて、基準データブロックや捜索領域、捜索データブロックを投影点の位置情報により適切に決定できるので、更にステレオマッチングの信頼性が高められる。
【０１３３】
また、基準データブロックや捜索データブロックをその探索位置により動的に可変可能となるので、信頼性がそれ以上に高められる。
【０１３４】
そして、これらステレオマッチングは、投影特徴パターンのない画像上で行えるので特徴パターンによる誤検出が生じない。また、特徴パターンがない画像は画像データベースとしての価値が生じ，計測と同時に原画像の蓄積が行える。
【０１３５】
さらに、前述の実施例によれば、特徴パターンを投影した画像と投影しない画像を使用するために、対象物に計測用のマークを貼るという作業が必要なくなり、マークの貼れないような対象物においても計測が可能になる。
【０１３６】
このように、特徴パターンがある画像とない画像を差分することにより、特徴パターン情報のみの画像が作成できることから、特徴パターン位置検出〜標定〜ステレオモデル形成〜三次元計測までを、人手によらず自動で精度良く行うことが可能となる。すなわち、従来熟練を要し煩雑であった人手による標定作業と三次元計測作業をなくすことができ、すべてを自動で行って信頼性を向上することが可能となる。
【０１３７】
さらに，カメラをステレオ運台上に強固に固定する必要はなく、現場で２台のカメラをラフに設置して撮影するだけで、高精度な三次元計測が可能になるので、現場や対象物によらず簡便に計測可能であるという卓越した効果がある。
【０１３８】
また，特徴パターンの点を増やすことにより，ステレオマッチングの初期値とすること、すなわち自動で探索幅、テンプレート画像の決定を行いステレオマッチングすることが可能となり、更にステレオマッチング時間の短縮と信頼性を大幅に向上させながら，対象物の表面形状を自動測定できるという卓越した効果がある。
【０１３９】
【発明の効果】
本発明によれば、測定対象物を異なる方向から撮影した一対の画像からそれぞれの画像中での測定対象物の位置関係を概略求め、一対の画像中、一方の画像を基準画像、他方の画像を捜索画像と定め、前記概略位置測定部で求めた位置関係に基づいて基準画像に基準データブロックを設定し、捜索画像に捜索領域を設けその捜索領域中に捜索データブロックを設定し、その捜索領域中に設定される捜索データブロックと基準データブロックとの対応関係を求めるに当たって、概略の位置関係に基づいて、基準画像への基準データブロックの設定、捜索画像への捜索領域の設定及び捜索データブロックの設定の少なくとも一つを行うように構成することにより、対応関係を正確に求めることができる。
【図面の簡単な説明】
【図１】本発明の形状測定装置における計測の流れを示すブロック図。
【図２】本発明の形状測定装置を概念的に示すブロック図。
【図３】形状測定装置の左右画像撮影部を示す概念図。
【図４】形状測定装置のパターン抽出部を示すブロック図。
【図５】投影パターンの一例を示す平面図
【図６】投影パターンの別の例を示す平面図
【図７】ＳＳＤＡ法によるテンプレートマッチングを示す説明図。
【図８】モーメント法を示す説明図。
【図９】基準画像を示す説明図。
【図１０】捜索画像を示す説明図。
【図１１】捜索画像を示す説明図。
【図１２】捜索画像を示す説明図。
【図１３】捜索画像を示す説明図。
【図１４】捜索画像を示す説明図。
【図１５】基準画像を示す説明図。
【図１６】基準画像を示す説明図。
【図１７】本発明による画像測定装置の概要を示すブロック図。
【符号の説明】
０ 測定対象物
１，２ 左右画像撮影部
３ パターン投影部
４ コントローラ
５ パターン抽出部
６ 姿勢算出部
７ ステレオモデル形成部
８ 形状測定部
９ テンプレートマッチング画像
１０ 形状測定装置
１１ 光学系
１２ ＣＣＤ
１３ ＣＣＤドライバ
１４ オペアンプ
１５ Ａ／Ｄ変換器
２０ 投影パターン
３０ 画像測定装置
３１ 概略位置測定部
３２ データ設定部
３３ 対応判別部
５１ パターン投影用画像メモリ
５２ パターンなし用画像メモリ
５３ 画像差分演算部（器）
５４ パターン画像メモリ
５５ パターン位置検出部[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a technique for automating three-dimensional measurement by automating a pre-stage orientation work when performing three-dimensional measurement from different camera positions and automatically acquiring an initial value of stereo matching.
[0002]
[Prior art]
When performing three-dimensional measurement in ground photogrammetry, measurement is performed according to the flow shown in FIG. That is, it is necessary to perform stereo imaging (a) → position detection (b) → location (c) → stereo model formation (d) → three-dimensional measurement (e) of an object. These processes are mainly performed by a computer. Among them, position detection (b) and three-dimensional measurement (e) are conventionally performed manually. The position detection (b) is a pre-processing for orientation that obtains the position and tilt of the camera to be photographed.
[0003]
By obtaining the positional relationship between the camera and the measurement object by the orientation (c), a stereo model that can be stereoscopically viewed can be formed, and three-dimensional measurement can be performed. The process of position detection (b) for performing orientation (c) is an operation for obtaining coordinate positions on each of the corresponding six or more points captured in different cameras.
[0004]
There are two types of three-dimensional measurement (e): point measurement and surface measurement.
[0005]
In the case of point measurement, the measurement points on the object are manually measured through human hands, but in order to achieve automation and improve accuracy, work is performed to mark the object.
[0006]
In the case of surface measurement, it is automatically performed using a technique called stereo matching by image processing. At that time, initial settings such as determination of a template image and setting of a search width have to be performed manually.
[0007]
[Problems to be solved by the invention]
The position detection (b) is performed by selecting six or more measurement points on the object in the left and right images taken by the operator, and linking the measurement points on the object and coordinate positions while observing each image. The detection is performed. However, these operations need to be performed while stereoscopically viewing, and thus require skill, and are complicated, difficult, and difficult.
[0008]
In particular, the task of determining measurement points, associating left and right images, and detecting detailed position coordinates is likely to cause individual differences, and it is difficult to obtain sufficient accuracy for each person. In addition, some people could not measure. In order to avoid such a problem, processing such as attaching a mark to an object may be performed.
[0009]
However, the increase in the work of attaching a mark to the object itself is a minus, and since there are some objects that cannot be easily attached with a mark, this method is not so popular.
[0010]
On the other hand, there is also a technique in which two cameras are firmly fixed on a stereo carriage and three-dimensional measurement is performed without performing an orientation work.
[0011]
However, in that case, the positional relationship between the two cameras on the carriage must never come, and the measurement environment and measurement object are greatly limited. At the same time, such a device is large, heavy, difficult to carry, and expensive, so that the three-dimensional measurement method by this method is not very popular.
[0012]
Further, in the point measurement of the three-dimensional measurement (e), when a mark or the like is not pasted on the measurement object, it is necessary for the operator to instruct and measure the measurement point while observing the photographed image. . For this reason, when there were many measurement points, it took time and effort. In addition, when trying to measure with high accuracy, individual differences are inevitably caused and there is a situation where measurement is impossible.
[0013]
The above situation can be avoided if the mark is attached to the object and measured, but in that case, as described above, a new time and effort is required to attach the mark, and it may be difficult to attach the mark depending on the object. Sometimes it was impossible to measure.
[0014]
In the case of surface measurement, it is necessary to manually determine a template image when performing automatic measurement by stereo matching as a preprocessing, and determine an optimal search width. Furthermore, if there is a mismatching point or the like, it is necessary to correct it manually, which makes it difficult to automate after all.
[0015]
SUMMARY OF THE INVENTION In view of the problems of the prior art, the present invention provides an image measuring apparatus having a simple configuration that can be carried out efficiently and with high accuracy from the orientation work to the measurement, can be carried regardless of the measurement object. The purpose is that.
[0016]
[Means for solving problems]
The present invention includes an approximate position measurement unit that roughly obtains a positional relationship of a measurement object in each image from a pair of images obtained by photographing the measurement object from different directions, and one image of the pair of images as a reference image, The other image is determined as a search image, a reference data block is set in the reference image based on the positional relationship obtained by the approximate position measurement unit, a search area is set in the search image, and a search data block is set in the search area A data setting unit, and a correspondence determination unit that obtains a correspondence relationship between the search data block set in the search region and the reference data block, and the data setting unit has the positional relationship obtained by the approximate position measurement unit. Based on the above, the reference data block is set to the reference image, the search area is set to the search image, and the search data block is set. The image measuring apparatus which is the subject matter.
[0017]
DETAILED DESCRIPTION OF THE INVENTION
The image measurement apparatus of the present invention includes a rough position measurement unit that roughly obtains a positional relationship of a measurement object in each image from a pair of images obtained by photographing the measurement object from different directions, and one image in the pair of images. Is set as a reference image, the other image is set as a search image, a reference data block is set in the reference image based on the positional relationship obtained by the approximate position measuring unit, a search area is provided in the search image, and search data is set in the search area. A data setting unit for setting a block; and a correspondence determining unit for obtaining a correspondence relationship between a search data block set in the search region and a reference data block, and the data setting unit is obtained by the approximate position measurement unit. Configured to perform at least one of the setting of the reference data block to the reference image, the setting of the search area to the search image and the setting of the search data block based on the determined positional relationship It is characterized in that was.
[0018]
The pair of images includes a feature pattern image whose positional relationship is known in advance, and the approximate position measurement unit includes the position of the feature pattern image in the reference image and the feature pattern image in the search image. Based on the positional relationship, the positional relationship of the measurement object in each of the pair of images can be roughly determined.
[0019]
The data setting unit can be configured to set a search area range in the search image based on the position of the feature pattern image in the search image.
[0020]
The data setting unit includes a search area in the search image based on a positional relationship between the position of the feature pattern image and the position of the search area in the search image, or the position of the feature pattern image in the reference image and the position of the reference data block. It is possible to configure so as to set the range.
[0021]
The data setting unit includes a search area in the search image based on a positional relationship between the position of the feature pattern image and the position of the search area in the search image, or the position of the feature pattern image in the reference image and the position of the reference data block. Can be configured to set the size of the range.
[0022]
The data setting unit is configured to increase the width or area of the search area range of the search image as the feature pattern image and search area in the search image are separated from the feature pattern image and reference data block in the reference image. Can be configured.
[0023]
The data setting unit may be configured to set a plurality of search areas having different sizes.
[0024]
The data setting unit can be configured to set a reference data block in the reference image based on the position of the feature pattern image in the reference image.
[0025]
The data setting unit can be configured to set the size of the reference data block in the reference image based on the position of the feature pattern image in the reference image and the reference data block position.
[0026]
The data setting unit can be configured to increase the height or width of the reference data block in the reference image as the position of the feature pattern image in the reference image and the reference data block position are separated.
[0027]
The data setting unit can be configured to set a plurality of reference data blocks having different sizes based on a feature pattern image in the reference image.
[0028]
【Example】
First, the outline of the image measuring apparatus according to the present invention will be described with reference to FIG.
[0029]
The image measuring device 30 includes an approximate position measuring unit 31, a data setting unit 32, and a correspondence determining unit 33.
[0030]
The approximate position measurement unit 31 roughly obtains the positional relationship of the measurement object in each image from a pair of images obtained by photographing the measurement object from different directions, using a feature pattern image whose positional relationship is known in advance. It is configured.
[0031]
The data setting unit 32 determines one of the pair of images as a reference image and the other as a search image, sets a reference data block in the reference image based on the positional relationship obtained by the approximate position measurement unit 31, and searches the search image. It is configured to set the area.
[0032]
The correspondence determination unit 33 is configured to obtain a correspondence relationship between the search data block set in the search area and the reference data block.
[0033]
By using this image measuring device 30, it is possible to automate the three-dimensional measurement described in detail below and improve the measurement accuracy.
[0034]
Hereinafter, a shape measuring apparatus using the image measuring apparatus of the present invention will be described in detail.
[0035]
FIG. 1 shows a flow of measurement in the shape measuring apparatus, and FIG. 2 is a block diagram conceptually showing the configuration.
[0036]
a. Stereo shooting of an object
First, a feature pattern is applied to the measurement object 0 from the feature pattern projection unit 3 according to an instruction from the controller 4. Then, the image of the measurement object 0 is taken by the left and right image taking units 1 and 2, and the image data is transferred to the feature pattern extracting unit 5.
[0037]
As shown in FIG. 3, the left and right image capturing units 1 and 2 include an optical system 11, a CCD 12, a CCD driver 13, an operational amplifier 14, an A / D converter 15, and the like.
[0038]
5 and 6 show examples of feature patterns. These feature patterns 20 are formed in a circular shape, but may have any shape other than a circle as long as the position of the mark image by the feature pattern projection is required.
[0039]
The feature pattern projection unit 3 may be anything that can project the feature pattern 20, such as a slide projector or a laser pointer.
[0040]
FIG. 4 is a detailed block diagram of the feature pattern extraction unit 5.
[0041]
Image data transferred by the left and right image capturing units 1 and 2 is converted into digital data by the A / D converter 15 of FIG. 3 and transferred to the feature pattern projection image memory 51 of the feature pattern extraction unit 5.
[0042]
Next, in response to an instruction from the controller 4, the feature pattern projection is stopped, the left and right image capturing units 1 and 2 capture images without the feature pattern, and the digital image is stored in the feature pattern-less image memory 52 of the feature pattern extraction unit 5. Transfer data.
[0043]
When the image transfer to the feature pattern projection image memory 51 and the feature pattern absence image memory 52 is completed, the two images are differentiated through the image difference calculator 53 in accordance with an instruction from the controller 4. Then, the difference image is taken into the feature pattern image memory 54.
[0044]
As a result, the data in the feature pattern image memory 54 is only information obtained by deleting the image information of the measurement object 0, that is, information on the feature pattern 20 (mark image data).
[0045]
b. Position detection
In order to perform the orientation process, the position of the mark image is detected by the feature pattern projection.
[0046]
The feature pattern position detection unit 55 detects the feature pattern coordinate position in the feature pattern image memory 54.
[0047]
The feature pattern 20 may have any shape as long as its position is clearly known, but here, the feature pattern 20 as shown in FIGS. 5 and 6 is assumed. Further, any number of feature patterns may be used as long as there are 6 or more feature patterns as shown in FIGS.
[0048]
Since the feature pattern image memory 54 contains no information other than the mark image obtained by the feature pattern projection, erroneous detection can be eliminated. Furthermore, since it becomes easy to perform position detection automatically, stable position coordinate detection can always be performed with high accuracy without individual differences.
[0049]
Here, a case will be described in which the template matching method is used for the approximate position detection of points and the moment method is used for detailed position detection.
[0050]
As the template matching method, a residual sequential test method (SSDA method), which is a kind of correlation method, will be described, but a normalized correlation method or the like may be used. For detailed position detection, the LOG filter method or the like may be used instead of the moment method.
[0051]
Hereinafter, the position detection process will be described.
[0052]
(Rough position detection)
1. Register the template image.
[0053]
As the template image, a simulation image similar to one mark of the projection feature pattern 20 of FIGS. 5 and 6 may be created, or an actual image may be selected and used.
[0054]
2. A point where S> R (a, b) is searched in the entire image (see Equation 1).
[0055]
The closer R (a, b) is to 0, the higher the degree of similarity. S is determined in advance with an appropriate value. In this case, since image information other than the feature pattern has been erased, it can be easily determined. A normalized correlation method or the like may be used for template matching, but the processing can be further speeded up by using a residual sequential test method.
[0056]
[Residual sequential test (SSDA method)]
The principle diagram of the SSDA method is shown in FIG.
[0057]
The point where the residual R (a, b) is minimized is the mark position to be obtained.
[0058]
[Expression 1]

In the addition of the expression, if the value of R (a, b) exceeds the minimum value of the past residuals, the addition is aborted, and the calculation process is performed to move to the next (a, b), thereby speeding up the processing. Can measure.
[0059]
3. R (a, b) The mark position of the projection feature pattern is determined by the number of marks according to conditions such as the minimum and the interval between adjacent mark positions, and set as position coordinates.
[0060]
(Detailed position determination)
1. Search area determination
For the point determined by the above (approximate position detection), a search area centered there is set.
[0061]
2. Determination of mark area
Regarding the search area, an area having a density value equal to or higher than a threshold is defined as one mark (see FIG. 8).
[0062]
As the threshold value, an appropriate value is determined in advance. The image is almost 0 because the difference processing is applied except for the mark.
[0063]
3. Center of gravity calculation
The moment method is performed, and the barycentric position coordinates are calculated.
[0064]
[Moment method]
As shown in FIG. 8, the following equation is applied to points that are equal to or higher than the threshold value T (mark K).
[0065]
[Expression 2]

[0066]
[Equation 3]

The center of gravity position can be calculated up to the sub-pixel position by Expression 2 and Expression 3.
[0067]
4). Perform steps 1-3 for the number of points.
[0068]
Thus, the position coordinates of the mark image by the feature pattern projection as shown in FIG. 5 or FIG. 6 can be calculated.
[0069]
Detailed position detection may be performed from the beginning without performing approximate position detection, or position detection may be performed by another algorithm.
[0070]
In any case, since the image has only feature pattern information, the position can be calculated at high speed and with high accuracy.
[0071]
c. orientation
Next, the coordinate value obtained by the feature pattern position detection unit 55 is sent to the posture calculation unit 6 to perform orientation calculation.
[0072]
By this calculation, the positions of the left and right cameras are obtained.
[0073]
These parameters are obtained by the following coplanar conditional expression.
[0074]
[Expression 4]

The origin of the model coordinate system is taken as the left projection center, and the line connecting the right projection centers is taken as the X axis. For the scale, the base line length is taken as the unit length. The parameters to be obtained at this time are 5 of the left camera Z-axis rotation angle κ1, the Y-axis rotation angle φ1, the right-hand camera Z-axis rotation angle κ2, the Y-axis rotation angle φ2, and the X-axis rotation angle ω2. One rotation angle. In this case, the X-axis rotation angle ω1 of the left camera is 0, so it is not necessary to consider.
[0075]
Under such conditions, the coplanar conditional expression of Expression 4 becomes Expression 5, and each parameter can be obtained by solving this expression.
[0076]
[Equation 5]

Here, the following relational expression for coordinate transformation is established between the model coordinate system XYZ and the camera coordinate system xyz.
[0077]
[Formula 6]

[0078]
[Expression 7]

Using these formulas, the unknown parameters are obtained by the following procedure.
1. The initial approximate value is usually 0.
2. The coplanar conditional expression 5 is Taylor-expanded around the approximate value, and the value of the differential coefficient when linearized is obtained by Expressions 6 and 7, and an observation equation is established.
3. The least square method is applied to obtain a correction amount for the approximate value.
4). Correct approximate values.
5). Using the corrected approximate value, 2. ~ 5. Repeat until the operation is converged.
[0079]
If some mark is in the shape of the object and the position is not detected well, it may not converge. In that case, one point is deleted and the above steps 1 to 5 are performed, and the converged parameter or the best parameter is used.
[0080]
d. Stereo model formation
Next, an image is converted into a stereo model coordinate system that can be viewed stereoscopically according to parameters obtained by orientation, and a stereo model is formed.
[0081]
The conversion formula to model coordinates is as follows.
[0082]
[Equation 8]

[0083]
[Equation 9]

[0084]
[Expression 10]

In this way, three-dimensional measurement is possible.
[0085]
e. 3D measurement
(1) Point measurement
By using the feature pattern 20 having six or more points as shown in FIG. 6 and processing the plurality of marks in the same manner as the position detection (b), the three-dimensional coordinates can be automatically obtained with high accuracy.
(2) Surface measurement
By using the projection feature pattern 20 as shown in FIG. 6, it is possible to set the initial value of stereo matching.
[0086]
How to determine the initial value in the case of surface measurement will be described.
[0087]
FIG. 9 and FIG. 10 are explanatory diagrams, and a part of the projection feature pattern as shown in FIG. 6 is extracted. FIG. 9 shows a reference image as a reference, and FIG. 10 shows a search image for searching. These reference / search images may be either left or right images. For example, the left image is determined as the reference image, and the right image is determined as the search image.
[0088]
The projected feature patterns correspond to the reference images S'1, S'2, S'5, and S'6 on the search image with S1, S2, S5, and S6.
[0089]
The corresponding point search is performed by performing stereo matching on the search area in the search image using the reference data block in the reference image as a template image. Image correlation processing or the like is used for stereo matching.
[0090]
[Determination of search area]
A method for determining the search area based on the projection feature pattern will be described.
[0091]
The projection feature patterns S1 to S2 (horizontal direction) are defined as a width A, S1 to S5 (vertical direction) as a width B, and a portion surrounded by S1, S2, S5, and S6 is defined as a search region R1. Similarly, the portion surrounded by S2, S3, S6, and S7 is determined as the search region R2,.
[0092]
As shown in FIG. 11, mark images based on actual projection feature patterns are not always on the same line depending on the shape of the object and the orientation of the CCD. Therefore, a quadrilateral that can obtain the maximum A and B widths from the four points of each projection feature pattern is created and used as a search area. For example, in FIG. 11, the maximum widths that can be taken by S1, S2, S5, and S6 are the horizontal direction A1 and the vertical direction B1, and the rectangular area created by this is the search area R1, and similarly S2, S3, S6, and S7. The search area R2 is a quadrilateral created with the maximum widths A2 and B2 that can be taken. In this way, there are some overlapping areas, but the boundary between the R1 and R2 areas can be reliably searched.
[0093]
This is to determine the range of the search area in the search image based on the position of the mark image in the search image that has already been obtained.
[0094]
FIG. 12 shows a modification of the search area determination.
[0095]
For example, since each projection feature pattern has already been obtained as a corresponding point on the left and right images, it is efficient to set the search start position and end position of the search area as areas near the projection point. That is, in the example of FIG. 11, if the search is advanced by S1 to S5 in the vertical direction by the search width A1, as the vicinity of the line of S5 is approached, useless search areas (areas where no corresponding points clearly exist) are obtained. Out.
[0096]
Therefore, up to half of D1 between S1 and S5 in FIG. 12 (vertical direction), the same horizontal point as S1 is set as the start point of the search area, and D1 and S5 are connected on the line connecting D1 and S5. To the horizontal search start point.
[0097]
If such a process is set as a search area for each projection point, and stereo matching is performed while the search areas are somewhat overlapped, a somewhat efficient search can be performed.
[0098]
FIG. 13 and FIG. 14 further improve the efficiency of search of the search area by setting the search data block in the search area.
[0099]
For example, when the positions on the search area corresponding to the reference data blocks T1, T2, T3,... On the reference image shown in FIG. As shown in FIG. 13, the search data block is divided into several blocks such as U1, U2, U3,. In this way, it is possible to shorten the search area search time and efficiently search. In this case, the range of the search data block can be determined by how many reference data blocks are taken. For example, if n reference data blocks are set between A ′ of the reference images S′1 to S′2, A1 / n or A1 / (n−1) may be set. (N-1) is the case where the search is performed by slightly overlapping the search data blocks.
[0100]
This is to set the range of the search area in the search image based on the position of the mark image and the position of the search area already obtained. In other words, it can be said that the range of the search area is set in the search image based on the position of the mark image in the reference image and the positional relationship with the reference data block position.
[0101]
A modification of the method of FIG. 13 is shown in FIG. This is a variable search data block size. That is, since it is known that the corresponding points with respect to the reference data block are close to each other in the vicinity of the marks S1 and S2, the size of the search data block is reduced and the corresponding point position becomes uncertain as the distance increases. It is. Accordingly, the reference data block has the maximum search data block size at the position A ′ / 2.
[0102]
The size of these search data blocks is determined from the search area A1 of S1 to S2.
[0103]
For example, if n is the number of reference data blocks between A ′,
・ Search data block size in the section of S1-S1 + A1 / 2:
(1 + t × i / n) × A1 / n,
-Search data block size in the section of S1 + A1 / 2 to S2:
(1 + t × (n−i) / n) × A1 / n,
However, i is the position of the search data block corresponding to each reference data block, i.e., i = 1 to n. T is a constant of magnification and is set to a predetermined value. For example, if you want to double the search data block size at the S1 position (U1) at the S1 + A1 / 2 position, select 2.
[0104]
These search data block sizes may be determined based on the marks S′1, S′2 and the reference data blocks T1, T2,... In the reference image. At that time, the above-mentioned A1 is set to A ', and the terms taking the magnification A1 / A' into consideration are multiplied.
[0105]
Thus, by setting the search area and the search data block, it is possible to perform the corresponding point search efficiently, that is, at high speed and with high reliability.
[0106]
In this method, based on the positional relationship between the position of the mark image in the search image and the position of the search area, based on the positional relationship between the position of the mark image in the reference image and the position of the reference data block, if the view is changed, The size of the search area is set in the search image.
[0107]
[Determination of reference data block]
The reference data block serving as a reference is determined from S′1, S′2, S′5, S′6 on the reference image, or S1, S2, S5, S6 on the search image. For example, if the reference image (FIG. 9) S′1 to S′2 is A ′ and S′1 to S′5 is B ′, the horizontal reference data block width is A ′ / n, and the vertical direction is B ′. / M can be determined.
[0108]
Alternatively, the size may be proportional to the ratio of the left and right images. For example, the horizontal direction is set to A '/ A * n, and the vertical direction is set to B' / B * m.
[0109]
n and m can be determined as appropriate sizes depending on the relationship between the size of the object to be measured and the number of pixels, but may be set as appropriate constants depending on the values of A ′ and B ′.
[0110]
FIG. 15 shows a method of performing stereo matching in the same manner as in the above process while obtaining correlation products with three reference data block sizes instead of one. Also in this case, these three sizes can be determined based on the information of A ′ and B ′.
[0111]
FIG. 16 is a further modification in the determination of the reference data block of the present invention. The regions near S′1 and S′2 may be relatively small because the correspondence is relatively obtained on the reference / search image, but the correspondence position becomes uncertain as the distance increases, so the reference data block Change size dynamically. For example, the reference data block is expanded as shown in FIG. 14 at the T1, T2, and T3 positions. The reference data block position is set to the maximum size at the A '/ 2 point. On the contrary, the reference data block size is gradually reduced from A ′ / 2 to S′2.
[0112]
By doing in this way, the reliability of stereo matching can be improved.
[0113]
These reference data block sizes are determined from the horizontal width A ′ of S ′ 1 to S ′ 2 and the vertical width B ′ of S ′ 1 to S ′ 5.
[0114]
For example, if n and m are the reference data block numbers between A 'and B',
Reference data block size in the section of S′1 to S′1 + A ′ / 2:
Horizontal direction: (1 + t × i / n) × A ′ / n,
Vertical direction: (1 + t × l / m) × B ′ / m
Reference data block size in the interval S'1 + A '/ 2 to S'2
Horizontal direction: (1 + t × (n−i) / n) × A ′ / n,
Vertical direction: (1 + t × (m−i) / m) × B ′ / m
However, i and l are the respective reference data block positions, i.e., i = 1 to n and l = 1 to m.
[0115]
Here, t is a constant of magnification and is set to a desired value. For example, if it is desired to double the reference data block size at the position S′1 (T1) at the position S′1 + A ′ / 2, 2 is selected.
[0116]
In this way, the reference data block can be made variable.
[0117]
Further, if the three types of reference data blocks as shown in FIG. 15 are made variable as in the above-described example and the correlation product is taken, the corresponding points can be searched with higher reliability by performing stereo matching.
[0118]
The determination of these reference data blocks may be similarly performed by using the marks S1 and S2 in the search image. In this case, A ′ is A1 and B ′ is B1, and the reference image is multiplied by a term that takes into account the magnification A ′ / A1 of the search image.
[0119]
By doing as described above, each initial value is automatically obtained.
[0120]
Next, the actual measurement (stereo matching) procedure will be described.
[0121]
As an example, the case of the projection feature patterns S1 to S8 in FIG. 6 will be described with reference to FIGS.
[0122]
In the vertical direction, since the orientation process has been completed and the vertical parallax has been removed (aligned), it is only necessary to search on the same line of the reference image and the search image. Further, when setting the search data block, the search data block is appropriately set as described above in each search area.
[0123]
1. For the search width A1 of the horizontal line L1 of the search area R1 from S1 to S2 of the search image,
The corresponding point search is sequentially performed on the reference data blocks T1, T2, T3,... In the reference image.
[0124]
These initial values are determined by the method described above.
[0125]
When A1 on the line L1 in the search area R1 ends,
2. For the search region R2 of S2 to S3, the corresponding point search on A2 of the line L1 is sequentially repeated from the reference data block position determined in this region.
[0126]
3. Next, a corresponding point search is performed on A3 of the line L1 in the search region R3 of S3 to S4. The search area R3 is sequentially performed with the template size and position determined in this area.
[0127]
When the corresponding point search of the line L1 is completed, the process moves to the next line L2, and the corresponding point search is repeated in the same manner as 1 to 3 from the search width A1 of the search region R1.
[0128]
Repeat this for the required number of lines.
[0129]
In this way, the initial values for stereo matching, that is, the search area, the search data block, and the reference data block are automatically determined from the position of the projection feature pattern, and automatic measurement can be performed.
[0130]
Furthermore, since the search area of the reference / search image and the reference data block can be limited to within A ′ and B ′, processing much faster than normal stereo matching is possible. That is, normally, stereo matching is performed for one horizontal line (FIG. 6, line) at each search position, but the processing time is a fraction of the number of projection points.
[0131]
In addition, since the corresponding area on the reference / search image is limited and required in advance, mismatching can be greatly reduced. That is, the reliability of stereo matching can be greatly improved.
[0132]
In addition, since the reference data block, the search area, and the search data block can be appropriately determined based on the position information of the projection point, the reliability of stereo matching can be further improved.
[0133]
Further, since the reference data block and the search data block can be dynamically changed according to the search position, the reliability is further increased.
[0134]
And since these stereo matching can be performed on the image without a projection feature pattern, the false detection by a feature pattern does not arise. In addition, an image without a feature pattern has value as an image database, and an original image can be stored simultaneously with measurement.
[0135]
Furthermore, according to the above-described embodiment, in order to use an image in which the feature pattern is projected and an image in which the feature pattern is not projected, an operation of attaching a measurement mark to the object is not necessary, and in an object in which the mark cannot be applied. Can also be measured.
[0136]
In this way, since an image with only feature pattern information can be created by subtracting an image with and without a feature pattern, feature pattern position detection, orientation, stereo model formation, and three-dimensional measurement can be performed without human intervention. It is possible to perform automatically and accurately. That is, it is possible to eliminate the manual orientation work and the three-dimensional measurement work, which have conventionally required skill and are complicated, and it is possible to improve the reliability by performing everything automatically.
[0137]
Furthermore, it is not necessary to fix the camera firmly on the stereo platform, and it is possible to perform high-precision 3D measurement simply by installing two cameras roughly at the site and taking pictures. Regardless of this, there is an excellent effect that it can be measured easily.
[0138]
In addition, by increasing the number of feature pattern points, it becomes possible to set the initial value of stereo matching, that is, to automatically determine the search width and template image, and to perform stereo matching, further reducing the stereo matching time and reliability. It has the outstanding effect of being able to automatically measure the surface shape of the object while greatly improving it.
[0139]
【The invention's effect】
According to the present invention, the positional relationship of the measurement object in each image is roughly determined from a pair of images obtained by photographing the measurement object from different directions, and one image is the reference image and the other image is the pair of images. Is determined as a search image, a reference data block is set in the reference image based on the positional relationship obtained by the approximate position measurement unit, a search area is set in the search image, a search data block is set in the search area, and the search is performed. In determining the correspondence between the search data block set in the area and the reference data block, the reference data block is set in the reference image, the search area is set in the search image, and the search data based on the approximate positional relationship. By configuring so that at least one of the block settings is performed, the correspondence can be accurately obtained.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a measurement flow in a shape measuring apparatus of the present invention.
FIG. 2 is a block diagram conceptually showing the shape measuring apparatus of the present invention.
FIG. 3 is a conceptual diagram illustrating a left and right image capturing unit of the shape measuring apparatus.
FIG. 4 is a block diagram showing a pattern extraction unit of the shape measuring apparatus.
FIG. 5 is a plan view showing an example of a projection pattern
FIG. 6 is a plan view showing another example of a projection pattern
FIG. 7 is an explanatory diagram showing template matching by the SSDA method.
FIG. 8 is an explanatory diagram showing a moment method.
FIG. 9 is an explanatory diagram showing a reference image.
FIG. 10 is an explanatory diagram showing a search image.
FIG. 11 is an explanatory diagram showing a search image.
FIG. 12 is an explanatory diagram showing a search image.
FIG. 13 is an explanatory diagram showing a search image.
FIG. 14 is an explanatory diagram showing a search image.
FIG. 15 is an explanatory diagram showing a reference image.
FIG. 16 is an explanatory diagram showing a reference image.
FIG. 17 is a block diagram showing an outline of an image measuring apparatus according to the present invention.
[Explanation of symbols]
0 Measurement object
1, 2 Left and right image capturing unit
3 Pattern projection unit
4 Controller
5 Pattern extraction unit
6 Attitude calculation unit
7 Stereo model formation part
8 Shape measuring unit
9 Template matching image
10 Shape measuring device
11 Optical system
12 CCD
13 CCD driver
14 operational amplifier
15 A / D converter
20 Projection pattern
30 Image measuring device
31 Approximate position measurement unit
32 Data setting part
33 Correspondence discrimination part
51 Image memory for pattern projection
52 Image memory without pattern
53 Image difference calculation unit
54 pattern image memory
55 Pattern position detector

Claims (10)

A measurement object on which a feature pattern with a known positional relationship is formed is photographed from different directions, and the positional relationship of the characteristic points of the measurement object in a pair of captured images is detected by detecting the position of the mark image. An approximate position measuring unit for obtaining an outline;
During a pair of images, one image to the reference image, defines the other image and search image, sets the reference data block in the standards image, the search area in the search image, sets a search data block in the search area A data setting unit to
A correspondence determination unit for obtaining a correspondence relationship between the search data block set in the search region and the reference data block;
The data setting unit, the characterized in that on the basis of the positional relationship obtained by the approximate position measuring portion, and configured to perform the setting of settings and the search data block settings and the search region of the reference data block Image measuring device.   The image measurement apparatus according to claim 1, wherein the data setting unit is configured to set a range of a search area in the search image based on a position of a feature pattern image in the search image. measuring device. The image measurement apparatus according to claim 1, wherein the data setting unit includes a position of the feature pattern image and the position of the search area in the search image, or a position of the feature pattern image and the reference data block position in the reference image. An image measuring apparatus configured to set a search area range in a search image based on the relationship.   The image measurement apparatus according to claim 1, wherein the data setting unit includes a position of the feature pattern image and the position of the search area in the search image, or a position of the feature pattern image and the reference data block position in the reference image. An image measuring apparatus configured to set the size of a search area range in a search image based on the relationship.   5. The image measurement apparatus according to claim 4, wherein the data setting unit includes a search area of the search image as the feature pattern image and the search area in the search image are separated from the feature pattern image and the reference data block in the reference image. An image measuring apparatus configured to set a large width or area of a range.   6. The image measurement device according to claim 2, wherein the data setting unit is configured to set a plurality of search areas having different sizes.   2. The image measuring apparatus according to claim 1, wherein the data setting unit is configured to set a reference data block in the reference image based on a position of a feature pattern image in the reference image. .   The image measurement apparatus according to claim 1, wherein the data setting unit is configured to set the size of the reference data block in the reference image based on the position of the feature pattern image in the reference image and the reference data block position. An image measuring apparatus characterized by that.   The image measurement apparatus according to claim 1, wherein the data setting unit sets the height or width of the reference data block to be larger in the reference image as the position of the feature pattern image in the reference image and the reference data block position are separated from each other. An image measuring apparatus configured as described above.   10. The image measuring apparatus according to claim 7, wherein the data setting unit is configured to set a plurality of reference data blocks having different sizes based on a feature pattern image in the reference image. An image measuring apparatus characterized by the above.

Images