JP3581506B2

JP3581506B2 - Shape processing method

Info

Publication number: JP3581506B2
Application number: JP31726096A
Authority: JP
Inventors: 透金子; 丈二中山
Original assignee: Nippon Telegraph and Telephone Corp
Current assignee: Nippon Telegraph and Telephone Corp
Priority date: 1996-11-28
Filing date: 1996-11-28
Publication date: 2004-10-27
Anticipated expiration: 2016-11-28
Also published as: JPH10162153A

Description

【０００１】
【発明の属する技術分野】
本発明は、レーザ距離センサやテレビカメラ等から得られた画像データを用いて物体の形状認識を行う形状処理技術に属する。本発明は、例えば、視覚センサ付きのロボットを使って２枚の板の溶接を行う時などに、屈曲点等の特徴点を求めることにより、この屈曲点に沿って溶接が行える等の利用分野がある。又、３次元物体認識においても、屈曲点を求めることにより、３次元物体の面を構成する線分が求められる等の応用が可能である。
【０００２】
【従来の技術】
物体の形状を表わすプロフィールデータを用いて形状認識を行う場合に、そのプロフィールデータから屈曲点などの特徴点を抽出することが基本的な処理として必要である。そこで従来から屈曲点の抽出の方法として、形状プロフィールデータの各点での微分を行ってその点での接線の傾きを求め、その傾きの変化する度合いを調べて屈曲点を求める方法が採られていた。
【０００３】
【発明が解決しようとする課題】
しかしながら、微分処理による上記従来の屈曲点の抽出方法では、形状プロフィールのデータが計測や撮影の際のノイズの影響により局所的に不安定な位置にばらついているため微分処理が多くの誤差を含み、屈曲点の検出の精度や信頼性が低いという問題点があった。
【０００４】
本発明は、上記問題点を解決するためになされたものであり、ばらついたデータからでも信頼性高く屈曲点位置を検出する形状処理方法を提供することを目的とする。
【０００５】
【課題を解決するための手段】
上記目的を達成するため、本発明による第１の発明の形状処理方法は、２次元座標の点列経路で表わされる形状プロフィールデータを処理する形状処理装置における形状処理方法であって、第１の最小２乗法実施部において、プロフィールデータ蓄積部に蓄積された前記点列経路のある着目点からみた前記点列経路に沿った一つの順方向およびその逆方向について、それぞれ最小２乗法により一定の数の点列データに対する近似直線を求める操作を、該点列データ中の各点について順次実行し、減算部において、前記各点においてそれぞれ求められた両方向の近似直線の勾配の差を該点の屈曲度として出力し、閾値処理部が、前記減算部において出力された各点の屈曲度について、その大きさが予め決められた値より大きい点列区間を求め、極値検出部が、得られた該区間のそれぞれについて屈曲度の極値を与える点を屈曲点として、その座標を出力する、ことを特徴とする。
【０００６】
また、本発明による第２の発明の形状処理方法は、２次元座標の点列経路で表わされる形状プロフィールデータを処理する形状処理装置における形状処理方法であって、第１の最小２乗法実施部において、プロフィールデータ蓄積部に蓄積された前記点列経路のある着目点からみた前記点列経路に沿った一つの順方向およびその逆方向について、それぞれ最小２乗法により一定の数の点列データに対する近似直線を求める操作を、該点列データ中の各点について順次実行し、角度算出部において、近似曲線の勾配から得られる傾斜角度を求め、減算部において、両方向の近似曲線の傾斜角度の差を該点の屈曲度として出力し、閾値処理部が、前記減算部において出力された各点の屈曲度について、その大きさが予め決められた値より大きい点列区間を求め、極値検出部が、得られた該区間のそれぞれについて屈曲度の極値を与える点を屈曲点として、その座標を出力する、ことを特徴とする。
【０００７】
また、本発明による第３の発明の形状処理方法は、端点検出部が、前記プロフィールデータ蓄積部に蓄積されたプロフィールデータから形状の端点を検出して、該端点を前記極値検出部で得られた屈曲点とともに特徴点データ蓄積部に蓄積し、第２の最小２乗法実施部が、前記特徴点データ蓄積部に蓄積された屈曲点の１つに着目し、着目した屈曲点と隣り合う特徴点を抽出し、当該着目した屈曲点と当該抽出された特徴点との間に含まれる点列データの数量が規定の値よりも多い場合に、当該点列データを前記プロフィールデータ蓄積部から抽出して、最小２乗法により直線または曲線近似を行い、図形延長処理部が、前記第２の最小２乗法実施部にて得られた近似直線または近似曲線を線が延びる両方向に延長する操作を、当該近似直線または近似曲線に沿って与えられるある一定の幅の帯状の範囲内に点列データが連続して含まれる間実行し、当該操作が終了した点の座標を延長操作前の屈曲点座標に代えて新たな屈曲点座標として出力する、ことを特徴とする。
【０００８】
また、本発明による第４の発明の形状処理方法は、端点検出部が、前記プロフィールデータ蓄積部に蓄積されたプロフィールデータから形状の端点を検出して、該端点を前記極値検出部で得られた屈曲点とともに特徴点として特徴点データ蓄積部に蓄積し、第３の最小２乗法実施部が、前記特徴点データ蓄積部に蓄積された屈曲点の１つに着目し、着目した屈曲点と隣り合う特徴点を抽出し、当該着目した屈曲点と当該抽出された特徴点との間に含まれる点列データを前記プロフィールデータ蓄積部から抽出して、最小２乗法により直線または曲線近似を行い、交点検出部が、前記着目した屈曲点に対して、前記第３の最小２乗法実施部にて得られた２つの近似直線または近似曲線どうしの交点を求めて、該交点を屈曲点として、その座標を出力する、ことを特徴とする。
【０００９】
また、本発明による第５の発明の形状処理方法は、前記図形延長処理部により得られた新たな屈曲点座標を、延長操作前の屈曲点座標に代えて前記特徴点データ蓄積部に蓄積し、第３の最小２乗法実施部が、前記特徴点データ蓄積部に蓄積された屈曲点の１つに着目し、着目した屈曲点と隣り合う特徴点を抽出し、当該着目した屈曲点と当該抽出された特徴点との間に含まれる点列データを前記プロフィールデータ蓄積部から抽出して、最小２乗法により直線または曲線近似を行い、交点検出部が、前記着目した屈曲点に対して、前記第３の最小２乗法実施部にて得られた２つの近似直線または近似曲線どうしの交点を求めて、該交点を屈曲点として、その座標を出力する、ことを特徴とする。
【００１０】
また、本発明による第６の発明の形状処理方法は、前記交点検出部にて得られた屈曲点座標を出力するに際し、前記第３の最小２乗法実施部が、当該屈曲点を求めるのに用いた近似直線または近似曲線の分散値をそれぞれ求め、第１種信頼度計算部が、前記分散値の小さいものには高い信頼度を、前記分散値の小さいものには低い信頼度を、前記屈曲点の位置に関する第１の信頼度として出力すること、を特徴とする。
【００１１】
また、本発明による第７の発明の形状処理方法は、前記交点検出部にて得られた屈曲点座標を出力するに際し、第２種信頼度計算部が、当該屈曲点の近傍の予め定められた距離で限定された範囲に存在する形状プロフィールデータ中の点の個数を調べ、該個数が多いものには高い値を、個数の少ないものには低い値を、当該屈曲点の位置に関する第２の信頼度として出力する、ことを特徴とする。
【００１２】
さらに、本発明による第８の発明の形状処理方法は、前記交点検出部にて得られた屈曲点座標を出力するに際し、前記第３の最小２乗法実施部が、当該屈曲点を求めるのに用いた近似直線または近似曲線の分散値をそれぞれ求め、第１種信頼度計算部が、前記分散値の小さいものには高い信頼度を、前記分散値の大きいものには低い信頼度を、前記屈曲点の位置に関する第１の信頼度として出力し、第２種信頼度計算部が、当該屈曲点の近傍の予め定められた距離で限定された範囲に存在する形状プロフィールデータ中の点の個数を調べ、該個数が多いものには高い値を、該個数の少ないものには低い値を、当該屈曲点の位置に関する第２の信頼度として出力し、総合信頼度計算部が、前記第１の信頼度と前記第２の信頼度とを入力して総合的な信頼度を計算し、当該屈曲点の位置に関する総合的な信頼度として出力する、ことを特徴とする。
【００１３】
本発明では、２次元座標の点列経路で表した形状プロフィールデータの各点について２つの近似直線のなす角度で定義される屈曲度を求め、さらに得られた屈曲度から屈曲点を抽出する。屈曲点は最初、点列データの各点のどれであるかは明らかでない。このため、すべての点において、該点の前後複数個の点列群により２つの近似直線を作成する。周辺が直線であるような点で屈曲度を算出した場合は当然ほぼ０度となる。しかしながら、その点が屈曲点である場合は、２本の近似直線のなす角度は最も大きくなる。従って、この各点間の屈曲度の変化を例えば図に描いたりした場合に極点となる部分が屈曲点である。本願発明では、上記において、近似直線を求める場合に、平均２乗誤差を最小にする最小２乗法を用いることにより、ばらついた形状プロフィールデータからでも信頼性高く屈曲点位置を検出可能とする。
【００１４】
【発明の実施の形態】
以下、本発明の実施形態例を、図を参照して詳細に説明する。
【００１５】
《屈曲度算出の原理》
図１は本発明における屈曲度算出の原理を示したものである。ここでは形状プロフィールデータが横軸ｘ、縦軸ｙの２次元平面に描かれているものとして説明し、該データを構成する点列のｎ番目の点をＰ（ｎ）と表わすこととする。
【００１６】
まず、点Ｐ（ｎ）についてその進行方向のｍ個の点列を加えた点列｛Ｐ（ｎ），Ｐ（ｎ＋１），…，Ｐ（ｎ＋ｍ）｝に対して、最小２乗法により近似直線ｙ＝ａ（ｎ）ｘ＋ｂ（ｎ）を求める。同じく点Ｐ（ｎ）についてその逆方向のｍ個の点列を加えた点列｛Ｐ（ｎ），Ｐ（ｎ−１），…，Ｐ（ｎ−ｍ）｝に対して、最小２乗法により近似直線ｙ＝ｃ（ｎ）ｘ＋ｄ（ｎ）を求める。このとき、係数ａが進行方向の傾き、係数ｃが逆方向の傾きを反映した値となり、しかもこの値は近傍の点列に対して最小２乗誤差をとるように直線で近似しているので、従来の微分処理に比べてデータのばらつきによる誤差の少ない信頼性の高い値となっている。ここでθａ（ｎ）＝ａｒｃｔａｎ（ａ（ｎ）），θｃ（ｎ）＝ａｒｃｔａｎ（ｃ（ｎ））とすると、θａ（ｎ）、θｃ（ｎ）はそれぞれ傾きの角度を表わし、ｄθ（ｎ）＝θａ（ｎ）−θｃ（ｎ）は点Ｐ（ｎ）で形状が屈曲している場合の角度を表わすことになる。
【００１７】
図１の（ａ）は点Ｐ（ｎ）が形状プロフィールの直線部にある場合で、ｄθ（ｎ）は理想的には０となる。しかし、実際には形状計測のノイズのためにわずかながら０ではない値をとることが一般的である。図１の（ｂ）は点Ｐ（ｎ）が滑らかに湾曲している部分に位置する場合で、この場合はｄθ（ｎ）の絶対値はやや大きめの値をとる。なおｄθ（ｎ）の符号は、湾曲が下に凸の場合は正、上に凸の場合は負となる。図１の（ｃ）は点Ｐ（ｎ）が屈曲している角の近傍に位置しているときで、その位置する近接度が増す毎にｄθ（ｎ）の絶対値が大きくなってくる。図１の（ｄ）は点Ｐ（ｎ）が屈曲している角に丁度位置したときで、ｄθ（ｎ）が最も大きくなり、その値は形状計測のノイズのために生ずる誤差の範囲内で該角のなす角度そのものを表わすことになる。
【００１８】
以上の説明からわかるように、最小２乗法による直線近似で得られたｄθ（ｎ）を用いる本発明は、形状プロフィールの屈曲度として安定した指標を与えることになる。なおここで、傾きの絶対値の大きな点列部分が存在しない形状プロフィールデータに対しては、傾きａ（ｎ）、ｃ（ｎ）そのものの差を角度の代わりに用いて、ｄθ（ｎ）＝ａ（ｎ）−ｃ（ｎ）とすることにより、ａ（ｎ）およびｃ（ｎ）の逆三角関数を求める計算を各点毎に実行するために費やす時間を削減することができる。
【００１９】
図２は図１の原理で得られる屈曲度の変化を例示して、屈曲点を求める原理を示したものである。図では２直線が上に凸な角で交わっている形状部分を示しており、その屈曲度は形状の直線部分ではほぼ０、角に近づくに従って負の大きな値となる。従ってノイズによって生ずる屈曲度の変動値よりも大きな適当な値ｔを用いて、ｄθ（ｎ）＜−ｔとなる部分点列区間をまず切り出し、その区間での極値（この場合は極小値）を与える点を求めて、これを屈曲点と定めればよいことになる。
【００２０】
図３は図２で求めた屈曲点位置をさらに精度よく求める原理を示したものである。既に説明しているように、形状プロフィールデータは形状計測の際のノイズによるばらつきを含んでおり、最小２乗近似により屈曲度を安定に求めたとしても、屈曲点としてもとのデータに含まれている点の座標そのものを用いたのでは、その点自身のばらつきによる位置誤差が屈曲点位置に含まれてしまう。そこでここでは、図２の原理により求めた屈曲点について、隣り合う屈曲点どうしの間は形状プロフィールが平坦な部分である、ということを利用し、まずその平坦部の形状を精度よく求めることを行う。そして求められた平坦部の交点を求めることにより、屈曲点位置を精度よく求めることとする。ここで平坦部分形状をばらついた点列データから精度よく求めることは、再度最小２乗法を適用することによって実現され、平坦部が直線と予めわかっている場合は最小２乗法による直線近似を、平坦部が曲線であれば曲線近似を行う。なお形状や計測の方法や視野範囲の制限によって生ずる形状プロフィールデータの端点については、端点とそれに一番近い屈曲点との間の点列データに対して、最小２乗法による形状近似を行う。以後、屈曲点とこのような端点を一括して扱う場合には、これらを特徴点とよぶことにする。図では、図２の原理で求められた屈曲点ＰａとＰｂとの間の点列を最小２乗法で直線近似し、また該屈曲点Ｐｂと形状の端点であるＰｃとの間の点列を最小２乗法で直線近似し、両近似直線の交点を求めて、これを屈曲点Ｐｂの代わりに改めて屈曲点Ｐ０と定めた場合を示している。
【００２１】
図４は、形状プロフィールデータ中に長さの短い平坦部が含まれる場合の屈曲点検出処理の原理を示したものである。図１で説明した屈曲度を求める際の点列の点の数ｍは、最小２乗法による点列データのばらつきの吸収効果を期待するためには、あまり小さな数を設定することができない。しかし一方、形状によってはこの数ｍで与えられる形状に沿った長さよりも短い平坦部を有していることが考えられる。このような場合には、屈曲度は形状に沿って鈍りを生じ、図２で示したような屈曲度の極値をもって定めた屈曲点は、正しい屈曲点の位置に対して、短い平坦部から遠ざかる方向にずれてしまう。そこでこのような現象に対処するために、形状プロフィールの端点および図２の原理に基づいて得られた屈曲点から構成される特徴点について、隣り合う特徴点の間に存在する形状プロフィールの点列データの中、ある一定の数以上の点を有する点列区間について、該点列データを最小２乗法により直線または曲線近似し、得られた近似直線または近似曲線を線が延びる両方向に延長する操作を、該近似直線または近似曲線に沿って与えられるある一定の幅の帯状の範囲内に点列データが連続して含まれる間実行する。そしてこの操作が終了したときの点の座標を延長操作前の屈曲点に代えて新たな屈曲点とすることにする。図４は、図２の原理で求めた屈曲点Ｐ１，Ｐ２に本処理を加えて、新たな屈曲点Ｐ３，Ｐ４が得られた様子を示している。さらに、このようにして求めた新たな屈曲点に対して、図３に示した原理を適用すれば、より精度のよい屈曲点位置が求められることは既に示した通りである。
【００２２】
以上、本発明の屈曲点位置を精度よく求める手法の原理を説明したが、求めた屈曲点の位置の精度がどの程度良いのかを示す指標を信頼度情報として付加することができる。具体的には、近似直線または近似曲線の交点として屈曲点を求めた場合の、それぞれの直線または曲線の最小２乗近似の分散値を求め、分散値の小さいものには高い値を、分散値の大きいものには低い値を、該屈曲点の位置に関する信頼度として与えることができる。また、該屈曲点の近傍の予め定められた距離で限定された範囲に存在する形状プロフィールデータ中の点の個数を調べ、該個数が多いものには高い値を、個数の少ないものには低い値を、該屈曲点の位置に関する信頼度として与えることもできる。そしてこれら２種類の信頼度を組み合わせて、分散値の小さいものには大きな重みを、また分散値の大きいものには小さな重みを与え、さらに前記個数が多いものには大きな重みを、また該個数の少ないものには小さな重みを与え、上記２種類の重みを総合して、該総合値の大きいものには高い値を、小さいものには低い値を、該屈曲点の位置に関する信頼度として与えることができる。
【００２３】
《実施形態例１》
図５は、本発明の第１の実施形態例である第１の発明の一実施形態例を実現する構成例を示す図である。図５の構成例において、１はプロフィールデータ蓄積部、２−１および２−２は最小２乗法実施部、３−１および３−２は角度算出部、４は減算部である。
【００２４】
この動作を説明すると、プロフィールデータ蓄積部１に蓄積された２次元の座標点列の経路で表わされる形状プロフィールデータに対して、データ点列中の１点に着目し、最小２乗法実施部２−１および２−２が、それぞれ点列経路に沿った一つの方向およびその逆方向について一定の数の点列データに対する最小２乗法に基づいた近似直線の係数を算出する。算出された該係数を用いて、角度算出部３−１および３−２は逆三角関数の計算を実施して、着目点における両方向の２つの近似直線の傾斜角度を算出する。そして最後に、減算部４が算出された両角度の差を計算して、これを屈曲度として出力する。以上の一連の動作をデータ中の各点について順次実行することにより、形状プロフィールデータの各点ごとの屈曲度が出力される。
【００２５】
《実施形態例２》
図６は、発明の第２の実施形態例である第１の発明の別の一実施形態例を実現する構成例を示す図である。図６の構成例は、図５の構成例に比べて角度算出部３−１および３−２を省略した構成となっている。
【００２６】
この動作を説明すると、プロフィールデータ蓄積部１に蓄積された２次元の座標点列の経路で表わされる形状プロフィールデータに対して、データ点列中の１点に着目し、最小２乗法実施部２−１および２−２が、それぞれ点列経路に沿った一つの方向およびその逆方向について一定の数の点列データに対する最小２乗法に基づいた近似直線の係数を算出する。算出されたそれぞれの係数のうち勾配を表わす係数に対して減算部４がその差を計算して、これを屈曲度として出力する。以上の一連の動作をデータ中の各点について順次実行することにより、形状プロフィールデータの各点ごとの屈曲度が出力される。
【００２７】
《実施形態例３》
図７は、本発明の第３の実施形態例である第２の発明の一実施形態例を実現する構成例を示す図である。図７の構成例において、５は屈曲度データ蓄積部、６は閾値処理部、７は極値検出部である。
【００２８】
この動作を説明すると、屈曲度データ蓄積部５に蓄積された図５あるいは図６による実施形態例により得られた屈曲度データに対し、閾値処理部６が予め決められた値より大きい屈曲度を有する点列区間を抽出する。極値検出部７は、抽出された該区間について屈曲度の極値を与える点を検出し、その点の座標を屈曲点の座標として出力する。極値検出部７はこの動作を抽出された区間の数だけ繰り返す。
【００２９】
《実施形態例４》
図８は、本発明の第４の実施形態例である第３の発明の一実施形態例を実現する構成例を示す図である。図８の構成例において、１はプロフィールデータ蓄積部、１１は屈曲点データ蓄積部、１２は端点検出部、１３は特徴点データ蓄積部、１４−１および１４−２は最小２乗法実施部、１５は交点検出部である。
【００３０】
この動作を説明すると、プロフィールデータ蓄積部１に蓄積されたプロフィールデータから、端点検出部１２が形状の端点を検出する。端点検出部１２で検出された端点データは、図７による第３の実施形態例で得られた屈曲点データを蓄積している屈曲点データ蓄積部１１の屈曲点データと共に、特徴点データ蓄積部１３に蓄積される。最小２乗法実施部１４−１および１４−２は、特徴点データ蓄積部１３に蓄積された特徴点のうちの１つの屈曲点に着目し、それぞれ該着目屈曲点と隣り合う特徴点を抽出し、着目屈曲点と該抽出特徴点の間に含まれる点列データをそれぞれプロフィールデータ蓄積部１から抽出して、最小２乗法に基づいて直線近似または曲線近似を行う。得られた２つの直線または曲線の係数を用いて、交点検出部１５は両直線または曲線の交点を求め、これを新たな屈曲点の座標として出力する。以上の動作を、着目したい屈曲点の数だけ繰り返す。
【００３１】
《実施形態例５》
図９は、本発明の第５の実施形態例である第４の発明の一実施形態例を実現する構成例を示す図である。図９の構成例において、１はプロフィールデータ蓄積部、１１は屈曲点データ蓄積部、１２は端点検出部１２、１３は特徴点データ蓄積部、１４は最小２乗法実施部、１６は図形延長処理部である。
【００３２】
この動作を説明すると、プロフィールデータ蓄積部１に蓄積されたプロフィールデータから、端点検出部１２が形状の端点を検出する。端点検出部１２で検出された端点データは、図７による第３の実施形態例で得られた屈曲点データを蓄積している屈曲点データ蓄積部１１の屈曲点データと共に、特徴点データ蓄積部１３に蓄積される。最小２乗法実施部１４は、特徴点データ蓄積部１３に蓄積された特徴点のうちの１つの屈曲点に着目し、該着目屈曲点と隣り合う特徴点を抽出する。そして着目屈曲点と該抽出特徴点の間に含まれる点列データの点の数が一定の数より多い場合に、該点列データをプロフィールデータ蓄積部１から抽出して、最小２乗法に基づいて直線近似または曲線近似を行う。図形延長処理部１６は、得られた近似直線または近似曲線を図形の両方向に延長する操作を、該近似直線または近似曲線に沿って与えられるある一定の幅の帯状の範囲内に点列データが連続して含まれる間実行する。そして該操作が終了した点の座標を延長操作前の屈曲点座標に代えた新たな屈曲点座標として出力する。以上の動作を、着目する屈曲点の数だけ繰り返す。
【００３３】
《実施形態例６》
図１０は、本発明の第６の実施形態例である第５の発明の一実施形態例を実現する構成例を示す図である。図１０の構成例は、図８の構成例と同様である。動作における唯一の違いは、屈曲点データ蓄積部１１には、図９による実施形態例で得られた屈曲点データを蓄積している点であり、従って動作の説明を省略する。
【００３４】
《実施形態例７》
図１１は、本発明の第７の実施形態例である第６の発明の一実施形態例を実現する構成例を示す図である。図１１の構成例は、図８の構成例に第１種信頼度計算部１７を付加した構成となっている。
【００３５】
この動作を説明するにあたり、基本的な動作は図８による第４の実施形態例と同様であるので説明を省略し、新たな構成要素である第１種信頼度計算部１７の動作を説明する。第１種信頼度計算部１７は、最小２乗法実施部１４−１および１４−２で付随的な計算結果として得られる分散値をそれぞれ入力に用いる。そして、例えば分散値が０の場合は１、分散値が非常に大きいときには０となるような変換を行い、得られた２つの変換値の積を求めて信頼度として出力する。
【００３６】
なお、本発明による第６の発明を第５の発明に適用した一実施形態例は、図１１の構成例の構成要素のうちの屈曲点データ蓄積部１１に図９による実施形態例で得られた屈曲点データを蓄積している点が異なるのみであるので、説明を省略する。
【００３７】
《実施形態例８》
図１２は、本発明の第８の実施形態例である第７の発明の一実施形態例を実現する構成例を示す図である。図１２の構成例は図８の構成例に第２種信頼度計算部１８を付加した構成となっている。
【００３８】
この動作を説明するにあたり、基本的な動作は図８による第４の実施形態例と同様であるので説明を省略し、新たな構成要素である第２種信頼度計算部１８の動作を説明する。第２種信頼度計算部１８は、交点検出部１５で得られた屈曲点座標の近傍の予め定められた距離で限定された範囲について、プロフィールデータ蓄積部１に蓄積されたプロフィールデータ中の点がいくつ存在するか個数を調べる。そして例えば、ある定められた個数より多い場合は１を与え、またその個数より少なくなるにしたがって小さな値を与え、１つも無い場合は０を与えるような変換を行い、その結果を信頼度として出力する。
【００３９】
なお、本発明による第７の発明を第５の発明に適用した一実施形態例は、図１２の構成例の構成要素のうちの屈曲点データ蓄積部１１に図９による実施形態例で得られた屈曲点データを蓄積している点が異なるのみであるので、説明を省略する。
【００４０】
《実施形態例９》
図１３は、本発明の第９の実施形態例である第８の発明の一実施形態例を実現する構成例を示す図である。図１３の構成例は図８の構成例に、第１種信頼度計算部１７、第２種信頼度計算部１８、および総合信頼度計算部１９を付加した構成となっている。
【００４１】
この動作を説明するにあたり、基本的な動作は図８による第４の実施形態例と同様であり、また第１種信頼度計算部１７、第２種信頼度計算部１８の動作もそれぞれ図１１および図１２の動作と同様であるので説明を省略し、新たな構成要素である総合信頼度計算部１９の動作を説明する。総合信頼度計算部１９は、第１種信頼度計算部１７および第２種信頼度計算部１８で算出されたそれぞれの信頼度を入力に用い、例えば両信頼度の積を求めてこれを総合的な信頼度として出力する。このとき単純な積ではなく、それぞれの信頼度を非線形に変換してから積を求めることにより、各信頼度の寄与の度合いを操作することも可能である。
【００４２】
なお、本発明による第８の発明を第５の発明に適用した一実施形態例は、図１３の構成例の構成要素のうちの屈曲点データ蓄積部１１に図９による実施形態例で得られた屈曲点データを蓄積している点が異なるのみであるので、説明を省略する。
【００４３】
以上、図面を用いて本発明の実施形態例を示したが、これら一連の動作は、その手順を実行するコンピュータ上のプログラムによって実現することができる。
【００４４】
【発明の効果】
以上説明したように、本発明によれば、形状プロフィールデータから屈曲点の位置を精度良く求めることができるので、後の画像認識処理の認識率や精度の向上を図ることができる。また本手法をもってしても屈曲点の位置の精度が低くなるようなほどにもともとの計測データにかなりのばらつきがある場合には、屈曲点位置の精度が低いということを示す信頼度指標を同時に出力することができるので、画像認識処理の結果の採用・不採用の目安をシステムに通知することが可能となる。
【図面の簡単な説明】
【図１】（ａ），（ｂ），（ｃ），（ｄ）は本発明における屈曲度算出の原理を説明する図である。
【図２】本発明における屈曲点検出の原理を説明する図である。
【図３】本発明における屈曲点検出の原理を説明する図である。
【図４】本発明における屈曲点検出の原理を説明する図である。
【図５】本発明による第１の発明の一実施形態例である本発明の第１の実施形態例を実現する構成例を示す図である。
【図６】本発明による第１の発明の一実施形態例である本発明の第２の実施形態例を実現する構成例を示す図である。
【図７】本発明による第２の発明の一実施形態例である本発明の第３の実施形態例を実現する構成例を示す図である。
【図８】本発明による第３の発明の一実施形態例である本発明の第４の実施形態例を実現する構成例を示す図である。
【図９】本発明による第４の発明の一実施形態例である本発明の第５の実施形態例を実現する構成例を示す図である。
【図１０】本発明による第５の発明の一実施形態例である本発明の第６の実施形態例を実現する構成例を示す図である。
【図１１】本発明による第６の発明の一実施形態例である本発明の第７の実施形態例を実現する構成例を示す図である。
【図１２】本発明による第７の発明の一実施形態例である本発明の第８の実施形態例を実現する構成例を示す図である。
【図１３】本発明による第８の発明の一実施形態例である本発明の第９の実施形態例を実現する構成例を示す図である。
【符号の説明】
Ｐ（ｎ）…形状プロフィールデータ中のｎ番目の点
θａ（ｎ），θｃ（ｎ）…傾き角度
ｄθ（ｎ）…屈曲度
ｔ…屈曲度閾値
Ｐａ，Ｐｂ，Ｐｃ，Ｐ０，Ｐ１，Ｐ２，Ｐ３，Ｐ４…屈曲点
１…プロフィールデータ蓄積部
２−１，２−２…最小２乗法実施部
３−１，３−２…角度算出部
４…減算部
５…屈曲度データ蓄積部
６…閾値処理部
７…極値検出部
１１…屈曲点データ蓄積部
１２…端点検出部
１３…特徴点データ蓄積部
１４，１４−１，１４−２…最小２乗法実施部
１５…交点検出部
１６…図形延長処理部
１７…第１種信頼度計算部
１８…第２種信頼度計算部
１９…総合信頼度計算部[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention belongs to a shape processing technology for performing shape recognition of an object using image data obtained from a laser distance sensor, a television camera, or the like. The present invention relates to a field of use in which, for example, when welding two plates using a robot equipped with a visual sensor, a characteristic point such as a bending point is obtained, so that welding can be performed along the bending point. There is. Also in the three-dimensional object recognition, applications such as obtaining a line segment constituting a surface of the three-dimensional object by obtaining a bending point are possible.
[0002]
[Prior art]
When performing shape recognition using profile data representing the shape of an object, it is necessary as a basic process to extract characteristic points such as bending points from the profile data. Therefore, conventionally, as a method of extracting a bending point, a method of differentiating each point of the shape profile data to obtain a slope of a tangent at that point, and examining a degree of change in the slope to obtain a bending point has been adopted. I was
[0003]
[Problems to be solved by the invention]
However, in the above-described conventional method of extracting a bending point by differential processing, the differential processing includes many errors because the data of the shape profile is locally scattered at an unstable position due to the influence of noise at the time of measurement or imaging. However, there has been a problem that the accuracy and reliability of detecting the bending point are low.
[0004]
The present invention has been made to solve the above problems, and has as its object to provide a shape processing method for detecting a bending point position with high reliability even from scattered data.
[0005]
[Means for Solving the Problems]
In order to achieve the above object, a shape processing method according to a first invention of the present invention is a shape processing method in a shape processing device that processes shape profile data represented by a point sequence path of two-dimensional coordinates, First In the least-squares method implementation unit, a fixed number of one forward direction and one backward direction along the point sequence path viewed from a point of interest of the point sequence path stored in the profile data storage unit by the least square method, respectively. An operation of obtaining an approximate straight line for the point sequence data is sequentially executed for each point in the point sequence data, and in a subtraction unit, The difference between the gradients of the approximate straight lines in both directions obtained at the respective points is output as the degree of bending of the point, and the threshold value processing unit calculates the degree of bending of each point output by the subtraction unit, and the magnitude is determined in advance. Obtain a point sequence section larger than the determined value, the extreme value detection unit, as a bending point a point that gives the extreme value of the degree of bending for each of the obtained section, and output the coordinates thereof, It is characterized by the following.
[0006]
Further, the shape processing method of the second invention according to the present invention comprises: A shape processing method in a shape processing device for processing shape profile data represented by a point sequence path represented by two-dimensional coordinates, wherein a first least squares method execution unit stores the point sequence path stored in a profile data storage unit. An operation of obtaining an approximate straight line for a certain number of point sequence data by the least squares method in one forward direction and the reverse direction along the point sequence path viewed from a certain point of interest is performed for each point in the point sequence data. Are sequentially executed, the angle calculation unit obtains an inclination angle obtained from the gradient of the approximate curve, and the subtraction unit outputs the difference between the inclination angles of the approximate curves in both directions as the degree of curvature of the point. For the degree of curvature of each point output by the subtraction unit, a point sequence section whose magnitude is larger than a predetermined value is determined, and the extreme value detection unit calculates the point sequence section. As bending point a point which gives an extreme value of tortuosity for, respectively, and outputs the coordinates, It is characterized by the following.
[0007]
Further, the shape processing method of the third invention according to the present invention, The end point detection unit detects an end point of the shape from the profile data stored in the profile data storage unit, and stores the end point in the feature point data storage unit together with the inflection point obtained by the extreme value detection unit. Two The least-squares method implementation unit Focusing on one of the inflection points stored in the feature point data storage unit, extracting a feature point adjacent to the inflection point of interest, and including the feature point between the inflection point of interest and the extracted feature point When the quantity of the point sequence data is larger than a prescribed value, the point sequence data is extracted from the profile data storage unit, and a straight line or curve approximation is performed by a least square method, The figure extension processing unit In the second least squares method implementation section An operation of extending the obtained approximate line or approximate curve in both directions in which the line extends, The It is executed while the point sequence data is continuously included in a band-like range having a certain width given along the approximate straight line or the approximate curve, The The coordinates of the point where the operation has been completed are output as new bending point coordinates instead of the bending point coordinates before the extension operation.
[0008]
The shape processing method according to a fourth aspect of the present invention includes: An end point detection unit detects an end point of the shape from the profile data stored in the profile data storage unit, and stores the end point together with the inflection point obtained by the extreme value detection unit as a feature point in the feature point data storage unit. And the third The least-squares method implementation unit Focusing on one of the inflection points stored in the feature point data storage unit, extracting a feature point adjacent to the inflection point of interest, and including the feature point between the inflection point of interest and the extracted feature point Point sequence data is extracted from the profile data storage unit, and a straight line or curve approximation is performed by the least squares method. The intersection detector is For the inflection point of interest, find the intersection of the two approximate straight lines or approximate curves obtained by the third least squares method execution unit, and output the coordinates of the intersection as the inflection point. It is characterized by the following.
[0009]
Further, the shape processing method of the fifth invention according to the present invention is characterized in that: The new inflection point coordinates obtained by the graphic extension processing unit are stored in the feature point data storage unit instead of the inflection point coordinates before the extension operation, and a third least squares method execution unit stores the feature point data Paying attention to one of the inflection points stored in the storage unit, extracting a feature point adjacent to the inflection point of interest, and extracting point sequence data included between the inflection point of interest and the extracted feature point. A straight line or a curve approximation is extracted from the profile data storage unit by the least squares method, and the intersection detection unit obtains the 2nd obtained by the third least squares method execution unit for the noted bending point. Find the intersection of two approximate straight lines or approximation curves and output the coordinates of the intersection as a bending point. It is characterized by the following.
[0010]
The shape processing method according to a sixth aspect of the present invention provides At the intersection detector In outputting the obtained inflection point coordinates, The third least squares performing unit is The variance value of the approximate straight line or the approximate curve used to determine the inflection point is determined, and the first type reliability calculating unit calculates Said For those with small variance High reliability , For those with small variance Low reliability, About the position of the inflection point First It is output as reliability.
[0011]
Further, the shape processing method of the seventh invention according to the present invention, Upon outputting the inflection point coordinates obtained by the intersection detection unit, The second type reliability calculation unit is: The Examine the number of points in the shape profile data present in a range limited by a predetermined distance near the inflection point, a high value for the large number, a low value for a small number, The About the position of the inflection point Second It is output as reliability.
[0012]
Further, the shape processing method of the eighth invention according to the present invention is characterized in that: At the intersection detector In outputting the obtained inflection point coordinates, The third least squares performing unit is Calculate the variance of the approximate straight line or approximate curve used to determine the inflection point, A first type reliability calculating unit that outputs a high reliability to the small variance value and a low reliability to the large variance value as the first reliability related to the position of the inflection point; The second type reliability calculation unit checks the number of points in the shape profile data existing in a range limited by a predetermined distance in the vicinity of the inflection point, and sets a high value to a large number of points. A low value is output as the second reliability regarding the position of the inflection point for the small number, The total reliability calculation unit Inputting the first reliability and the second reliability, calculating a total reliability, and outputting the total reliability as a total reliability regarding the position of the bending point; It is characterized by the following.
[0013]
According to the present invention, a bending degree defined by an angle between two approximate straight lines is obtained for each point of the shape profile data represented by a point sequence path of two-dimensional coordinates, and a bending point is extracted from the obtained bending degree. It is not clear which inflection point is each point of the point sequence data at first. For this reason, at each point, two approximate straight lines are created by a plurality of point sequence groups before and after the point. When the degree of bending is calculated at a point where the periphery is a straight line, it naturally becomes almost 0 degree. However, when that point is a bending point, the angle formed by the two approximate straight lines is the largest. Therefore, when the change in the degree of bending between these points is drawn, for example, in a diagram, the part that becomes the extreme point is the bending point. According to the present invention, when the approximate straight line is obtained, the least square method that minimizes the mean square error is used, so that the position of the inflection point can be detected with high reliability even from scattered shape profile data.
[0014]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[0015]
《Principle of bending degree calculation》
FIG. 1 shows the principle of calculating the degree of bending in the present invention. Here, it is assumed that the shape profile data is drawn on a two-dimensional plane having a horizontal axis x and a vertical axis y, and the n-th point in a sequence of points constituting the data is represented by P (n).
[0016]
First, a point sequence {P (n), P (n + 1),..., P (n + m)} obtained by adding m point sequences in the traveling direction of the point P (n) is approximated by the least square method. Find y = a (n) x + b (n). Similarly, for a point sequence {P (n), P (n−1),..., P (nm)} obtained by adding m point sequences in the opposite direction for the point P (n), the least square method To obtain an approximate straight line y = c (n) x + d (n). At this time, the coefficient a is a value reflecting the gradient in the traveling direction, and the coefficient c is a value reflecting the gradient in the reverse direction. Further, since this value is approximated by a straight line so as to take a least square error with respect to a nearby point sequence. This is a highly reliable value with less error due to data variation as compared with the conventional differentiation processing. Here, assuming that θa (n) = arctan (a (n)) and θc (n) = arctan (c (n)), θa (n) and θc (n) represent angles of inclination, respectively, and dθ (n ) = Θa (n) −θc (n) represents the angle when the shape is bent at the point P (n).
[0017]
FIG. 1A shows a case where the point P (n) is located in the linear portion of the shape profile, and dθ (n) is ideally 0. In practice, however, it is common to take a slightly non-zero value due to noise in shape measurement. FIG. 1B shows a case where the point P (n) is located in a smoothly curved portion. In this case, the absolute value of dθ (n) takes a slightly larger value. The sign of dθ (n) is positive when the curve is convex downward, and negative when the curve is convex upward. FIG. 1C shows a case in which the point P (n) is located near the bent corner, and the absolute value of dθ (n) increases as the degree of proximity increases. FIG. 1D shows a case where the point P (n) is just located at the bent corner, and dθ (n) is the largest, and its value is within the range of an error caused by the noise of the shape measurement. The angle itself is represented by the angle.
[0018]
As can be seen from the above description, the present invention using dθ (n) obtained by the linear approximation by the least-squares method gives a stable index as the degree of curvature of the shape profile. Here, for shape profile data in which there is no point sequence part having a large absolute value of the slope, the difference between the slopes a (n) and c (n) itself is used instead of the angle, and dθ (n) = By setting a (n) −c (n), it is possible to reduce the time spent executing the calculation for obtaining the inverse trigonometric function of a (n) and c (n) for each point.
[0019]
FIG. 2 illustrates the principle of obtaining a bending point, exemplifying a change in the degree of bending obtained by the principle of FIG. The figure shows a shape portion where two straight lines intersect at an upwardly convex corner, and the degree of bending is substantially zero in the straight portion of the shape, and becomes a large negative value as the angle approaches. Therefore, using a suitable value t larger than the fluctuation value of the degree of bending caused by noise, a partial point sequence section where dθ (n) <− t is first cut out, and an extreme value (a minimal value in this case) in that section Is determined, and this point is determined as the inflection point.
[0020]
FIG. 3 shows the principle of obtaining the bending point position obtained in FIG. 2 with higher accuracy. As already described, the shape profile data includes the variation due to noise at the time of shape measurement, and even if the bending degree is stably obtained by the least square approximation, it is included in the original data as a bending point. If the coordinates of the point itself are used, a position error due to the variation of the point itself is included in the bending point position. Therefore, here, regarding the bending points obtained based on the principle of FIG. 2, the fact that the shape profile is a flat portion between adjacent bending points is utilized, and first, the shape of the flat portion is accurately obtained. Do. Then, by determining the intersection of the determined flat portions, the bending point position is determined with high accuracy. Accurately obtaining the flat portion shape from the point sequence data having variations is realized by applying the least-squares method again. If the flat portion is known in advance as a straight line, a straight-line approximation by the least-squares method is performed. If the part is a curve, curve approximation is performed. Regarding the end point of the shape profile data generated by the shape, the measuring method, and the restriction of the visual field range, the shape approximation by the least squares method is performed on the point sequence data between the end point and the nearest bending point. Hereinafter, when the inflection points and such end points are collectively handled, these will be referred to as feature points. In the figure, the point sequence between the inflection points Pa and Pb obtained by the principle of FIG. 2 is linearly approximated by the least square method, and the point sequence between the inflection point Pb and the end point Pc of the shape is expressed as A case is shown in which a straight line is approximated by the least-squares method, an intersection of the two approximated straight lines is obtained, and this is set again as a bending point P0 instead of the bending point Pb.
[0021]
FIG. 4 shows the principle of the bending point detection processing when the shape profile data includes a flat portion having a short length. The number m of the points in the point sequence at the time of obtaining the degree of bending described with reference to FIG. 1 cannot be set to a very small number in order to expect the effect of absorbing the dispersion of the point sequence data by the least squares method. However, on the other hand, it is conceivable that some shapes have a flat portion shorter than the length along the shape given by this several meters. In such a case, the bending degree becomes blunt along the shape, and the bending point defined by the extreme value of the bending degree as shown in FIG. It shifts in the direction to go away. Therefore, in order to cope with such a phenomenon, regarding a feature point composed of an end point of the shape profile and a bending point obtained based on the principle of FIG. 2, a point sequence of a shape profile existing between adjacent feature points. In a point sequence section having a certain number of points or more in the data, an operation of approximating the point sequence data with a straight line or a curve by the least squares method and extending the obtained approximate line or approximate curve in both directions in which the line extends. Is performed while the point sequence data is continuously included in a band-like range having a certain width provided along the approximate straight line or the approximate curve. Then, the coordinates of the point at the time when this operation is completed are set as new bending points instead of the bending points before the extension operation. FIG. 4 shows a state in which the present process is added to the bending points P1 and P2 obtained by the principle of FIG. 2 to obtain new bending points P3 and P4. Further, if the principle shown in FIG. 3 is applied to the new bending point obtained in this manner, a more accurate bending point position can be obtained as described above.
[0022]
As described above, the principle of the method for accurately determining the position of the bending point according to the present invention has been described. However, an index indicating how good the accuracy of the obtained position of the bending point is can be added as reliability information. Specifically, when an inflection point is determined as an intersection of an approximate straight line or an approximate curve, a variance value of the least square approximation of each straight line or curve is determined, and a higher variance value is assigned to a smaller variance value. The lower value can be given to the one having a larger value as the reliability regarding the position of the bending point. Further, the number of points in the shape profile data existing in a range limited by a predetermined distance in the vicinity of the bending point is checked, and a higher value is set for a larger number and a lower value is set for a smaller number. The value may be given as a reliability regarding the position of the inflection point. Then, by combining these two kinds of reliability, a large weight is given to a small variance value, a small weight is given to a large variance value, and a large weight is given to a large variance value. A small weight is given to an object having a small number, and the above two types of weights are combined, and a high value is given to a thing having a large total value, and a low value is given to a thing having a small total value as reliability regarding the position of the inflection point. be able to.
[0023]
<< Embodiment 1 >>
FIG. 5 is a diagram showing a configuration example for realizing an embodiment of the first invention which is the first embodiment of the present invention. In the configuration example of FIG. 5, 1 is a profile data storage unit, 2-1 and 2-2 are least square method execution units, 3-1 and 3-2 are angle calculation units, and 4 is a subtraction unit.
[0024]
This operation will be described. With respect to the shape profile data represented by the path of the two-dimensional coordinate point sequence stored in the profile data storage unit 1, attention is focused on one point in the data point sequence, and the least squares method execution unit 2 -1 and 2-2 calculate the coefficients of the approximate straight line based on the least squares method for a fixed number of point sequence data in one direction along the point sequence path and the opposite direction. Using the calculated coefficients, the angle calculation units 3-1 and 3-2 calculate the inverse trigonometric function to calculate the inclination angles of two approximate straight lines in both directions at the point of interest. Finally, the subtraction unit 4 calculates the difference between the calculated angles, and outputs this as the degree of bending. By sequentially executing the above series of operations for each point in the data, the degree of bending for each point of the shape profile data is output.
[0025]
<< Embodiment 2 >>
FIG. 6 is a diagram showing a configuration example for realizing another embodiment of the first invention which is the second embodiment of the invention. The configuration example in FIG. 6 is different from the configuration example in FIG. 5 in that the angle calculation units 3-1 and 3-2 are omitted.
[0026]
This operation will be described. With respect to the shape profile data represented by the path of the two-dimensional coordinate point sequence stored in the profile data storage unit 1, attention is focused on one point in the data point sequence, and the least squares method execution unit 2 -1 and 2-2 calculate the coefficients of the approximate straight line based on the least squares method for a fixed number of point sequence data in one direction along the point sequence path and the opposite direction. The subtraction unit 4 calculates the difference of the coefficient representing the gradient among the calculated coefficients, and outputs the difference as the degree of bending. By sequentially executing the above series of operations for each point in the data, the degree of bending for each point of the shape profile data is output.
[0027]
<< Embodiment 3 >>
FIG. 7 is a diagram showing a configuration example for realizing one embodiment of the second invention which is the third embodiment of the present invention. In the configuration example of FIG. 7, reference numeral 5 denotes a bending degree data accumulation unit, 6 denotes a threshold value processing unit, and 7 denotes an extreme value detection unit.
[0028]
This operation will be described. The threshold value processing unit 6 determines that the degree of bending is larger than a predetermined value with respect to the degree of bending data obtained by the embodiment shown in FIG. 5 or FIG. The point sequence section which has the same is extracted. The extremum detection unit 7 detects a point that gives the extremum of the degree of flexion for the extracted section, and outputs the coordinates of the point as the coordinates of the flexion point. The extreme value detecting section 7 repeats this operation for the number of extracted sections.
[0029]
<< Embodiment 4 >>
FIG. 8 is a diagram showing a configuration example for realizing an embodiment of the third invention which is a fourth embodiment of the present invention. 8, 1 is a profile data storage unit, 11 is a bend point data storage unit, 12 is an end point detection unit, 13 is a feature point data storage unit, 14-1 and 14-2 are least square method execution units, Reference numeral 15 denotes an intersection detection unit.
[0030]
To explain this operation, the end point detection unit 12 detects the end point of the shape from the profile data stored in the profile data storage unit 1. The end point data detected by the end point detecting section 12 is stored in the characteristic point data storage section together with the bending point data of the bending point data storage section 11 storing the bending point data obtained in the third embodiment example shown in FIG. 13 is stored. The least-squares method implementation units 14-1 and 14-2 focus on one inflection point among the feature points stored in the feature point data storage unit 13, and extract a feature point adjacent to the noted inflection point, respectively. Then, the point sequence data included between the inflection point of interest and the extracted feature point is extracted from the profile data storage unit 1 and linear approximation or curve approximation is performed based on the least square method. Using the obtained coefficients of the two straight lines or curves, the intersection detection unit 15 obtains the intersection of both straight lines or curves, and outputs this as the coordinates of the new bending point. The above operation is repeated by the number of bending points to be focused on.
[0031]
<< Embodiment 5 >>
FIG. 9 is a diagram showing a configuration example for realizing an embodiment of the fourth invention which is a fifth embodiment of the present invention. In the configuration example of FIG. 9, 1 is a profile data storage unit, 11 is a bending point data storage unit, 12 is an end point detection unit 12, 13 is a feature point data storage unit, 14 is a least squares method execution unit, and 16 is a figure extension process. Department.
[0032]
To explain this operation, the end point detection unit 12 detects the end point of the shape from the profile data stored in the profile data storage unit 1. The end point data detected by the end point detecting section 12 is stored in the characteristic point data storage section together with the bending point data of the bending point data storage section 11 storing the bending point data obtained in the third embodiment example shown in FIG. 13 is stored. The least-squares method implementation unit 14 focuses on one inflection point among the feature points stored in the feature point data storage unit 13 and extracts a feature point adjacent to the attention inflection point. When the number of points in the point sequence data included between the inflection point of interest and the extracted feature point is larger than a certain number, the point sequence data is extracted from the profile data storage unit 1 and based on the least squares method. To perform linear approximation or curve approximation. The graphic extension processing unit 16 performs an operation of extending the obtained approximate straight line or approximate curve in both directions of the graphic, by converting the point sequence data into a band-like range having a certain width given along the approximate straight line or approximate curve. Execute while included in succession. Then, the coordinates of the point where the operation has been completed are output as new bending point coordinates in place of the bending point coordinates before the extension operation. The above operation is repeated by the number of the bending points of interest.
[0033]
<< Embodiment 6 >>
FIG. 10 is a diagram showing a configuration example for realizing one embodiment of the fifth invention which is the sixth embodiment of the present invention. The configuration example of FIG. 10 is the same as the configuration example of FIG. The only difference in the operation is that the inflection point data storage unit 11 stores the inflection point data obtained in the example of the embodiment shown in FIG. 9, and therefore, the description of the operation is omitted.
[0034]
<< Embodiment 7 >>
FIG. 11 is a diagram showing a configuration example for realizing one embodiment of the sixth invention which is the seventh embodiment of the present invention. The configuration example of FIG. 11 has a configuration obtained by adding the first type reliability calculating unit 17 to the configuration example of FIG.
[0035]
In describing this operation, the basic operation is the same as that of the fourth embodiment shown in FIG. 8, and therefore the description is omitted, and the operation of the first type reliability calculating unit 17, which is a new component, will be described. . The first-type reliability calculating unit 17 uses, as inputs, variance values obtained as ancillary calculation results by the least squares method executing units 14-1 and 14-2. Then, for example, the conversion is performed such that the variance becomes 0 when the variance is 0, and becomes 0 when the variance is extremely large, and the product of the obtained two converted values is obtained and output as reliability.
[0036]
An embodiment in which the sixth invention according to the present invention is applied to the fifth invention is obtained in the embodiment shown in FIG. 9 in the inflection point data storage unit 11 among the components of the configuration example shown in FIG. The only difference is that the inflection point data is stored, and a description thereof will be omitted.
[0037]
<< Embodiment 8 >>
FIG. 12 is a diagram showing a configuration example for realizing one embodiment of the seventh invention which is the eighth embodiment of the present invention. The configuration example of FIG. 12 has a configuration in which the second type reliability calculating unit 18 is added to the configuration example of FIG.
[0038]
In describing this operation, the basic operation is the same as that of the fourth embodiment shown in FIG. 8, and thus the description is omitted, and the operation of the second type reliability calculating unit 18 as a new component will be described. . The second-class reliability calculator 18 calculates a point in the profile data stored in the profile data storage 1 for a range limited by a predetermined distance in the vicinity of the inflection point coordinates obtained by the intersection detector 15. Check how many exist. For example, when the number is larger than a predetermined number, 1 is given, and when the number is smaller than the given number, a smaller value is given. When there is no number, 0 is given, and the result is output as reliability. I do.
[0039]
An embodiment in which the seventh invention according to the present invention is applied to the fifth invention is obtained in the embodiment shown in FIG. 9 in the inflection point data storage unit 11 among the components of the configuration example shown in FIG. The only difference is that the inflection point data is stored, and a description thereof will be omitted.
[0040]
<< Embodiment 9 >>
FIG. 13 is a diagram showing a configuration example for realizing an embodiment of the eighth invention which is the ninth embodiment of the present invention. The configuration example of FIG. 13 has a configuration in which a type 1 reliability calculation unit 17, a type 2 reliability calculation unit 18, and an overall reliability calculation unit 19 are added to the configuration example of FIG.
[0041]
In describing this operation, the basic operation is the same as that of the fourth embodiment shown in FIG. 8, and the operations of the first type reliability calculating unit 17 and the second type reliability calculating unit 18 are also shown in FIG. Since the operation is the same as that of FIG. 12, the description is omitted, and the operation of the integrated reliability calculating unit 19, which is a new component, will be described. The overall reliability calculation unit 19 uses the respective reliability calculated by the first type reliability calculation unit 17 and the second type reliability calculation unit 18 as an input, for example, obtains a product of both reliability, and integrates them. Is output as realistic reliability. At this time, instead of a simple product, the degree of contribution of each reliability can be manipulated by calculating the product after each reliability is nonlinearly converted.
[0042]
An embodiment in which the eighth invention according to the present invention is applied to the fifth invention is obtained in the embodiment shown in FIG. 9 in the inflection point data storage unit 11 among the components of the configuration example shown in FIG. The only difference is that the inflection point data is stored, and a description thereof will be omitted.
[0043]
As described above, the embodiment of the present invention has been described with reference to the drawings. These series of operations can be realized by a program on a computer that executes the procedure.
[0044]
【The invention's effect】
As described above, according to the present invention, the position of the inflection point can be obtained with high accuracy from the shape profile data, so that the recognition rate and accuracy of the subsequent image recognition processing can be improved. If there is considerable variation in the original measurement data such that the accuracy of the position of the inflection point becomes low even with this method, a reliability index indicating that the accuracy of the position of the inflection point is low is also simultaneously displayed. Since the data can be output, it is possible to notify the system of the standard of adoption / non-employment of the result of the image recognition processing.
[Brief description of the drawings]
FIGS. 1A, 1B, 1C, and 1D are diagrams for explaining the principle of calculating the degree of bending in the present invention.
FIG. 2 is a diagram for explaining the principle of bending point detection in the present invention.
FIG. 3 is a diagram for explaining the principle of bending point detection in the present invention.
FIG. 4 is a diagram for explaining the principle of bending point detection in the present invention.
FIG. 5 is a diagram showing a configuration example for realizing the first embodiment of the present invention, which is one embodiment of the first invention according to the present invention;
FIG. 6 is a diagram showing a configuration example for realizing a second embodiment of the present invention, which is an embodiment of the first invention according to the present invention.
FIG. 7 is a diagram showing a configuration example for realizing a third embodiment of the present invention which is an embodiment of the second invention according to the present invention.
FIG. 8 is a diagram showing a configuration example for realizing a fourth embodiment of the present invention, which is an embodiment of the third invention according to the present invention.
FIG. 9 is a diagram showing a configuration example for realizing a fifth embodiment of the present invention, which is an embodiment of the fourth invention according to the present invention.
FIG. 10 is a diagram showing a configuration example for realizing a sixth embodiment of the present invention, which is one embodiment of the fifth invention according to the present invention.
FIG. 11 is a diagram showing a configuration example for realizing a seventh embodiment of the present invention, which is an embodiment of the sixth invention according to the present invention.
FIG. 12 is a diagram showing a configuration example for realizing an eighth embodiment of the present invention, which is an embodiment of the seventh invention according to the present invention.
FIG. 13 is a diagram showing a configuration example for realizing a ninth embodiment of the present invention, which is an embodiment of the eighth invention according to the present invention.
[Explanation of symbols]
P (n): n-th point in shape profile data
θa (n), θc (n) ... Tilt angle
dθ (n): degree of bending
t: Flexibility threshold
Pa, Pb, Pc, P0, P1, P2, P3, P4 ... Bending point
1. Profile data storage unit
2-1, 2-2 ... Least squares method implementation unit
3-1, 3-2 ... angle calculation unit
4: Subtraction unit
5. Bending degree data storage unit
6: threshold processing unit
7 Extreme value detection section
11 ... Bend point data storage unit
12 ... End point detection unit
13. Feature point data storage unit
14, 14-1, 14-2 ... Least-squares method implementation unit
15 ... Intersection detection unit
16 Figure extension processing unit
17 Class 1 reliability calculator
18 Type 2 reliability calculator
19: Total reliability calculation unit

Claims

A shape processing method in a shape processing device that processes shape profile data represented by a two-dimensional coordinate point sequence path,
In a first least-squares method implementation unit, one forward direction and the opposite direction along the point sequence path viewed from a point of interest of the point sequence path stored in the profile data storage unit are each calculated by the least square method. An operation of obtaining an approximate straight line for a certain number of point sequence data is sequentially executed for each point in the point sequence data,
In the subtraction unit, the difference between the gradients of the approximate straight lines in both directions obtained at each point is output as the degree of curvature of the point,
A threshold processing unit obtains a point sequence section whose magnitude is larger than a predetermined value for the degree of bending of each point output in the subtraction unit,
The extreme value detection unit outputs the coordinates of a point that gives the extreme value of the degree of bending for each of the obtained sections as a bending point.
A shape processing method characterized in that:

A shape processing method in a shape processing device that processes shape profile data represented by a two-dimensional coordinate point sequence path,
In a first least-squares method implementation unit, one forward direction and the opposite direction along the point sequence path viewed from a point of interest of the point sequence path stored in the profile data storage unit are each calculated by the least square method. An operation of obtaining an approximate straight line for a certain number of point sequence data is sequentially executed for each point in the point sequence data,
In the angle calculation unit, determine the inclination angle obtained from the gradient of the approximate curve,
In the subtraction unit, the difference between the inclination angles of the approximate curves in both directions is output as the degree of curvature of the point,
A threshold processing unit obtains a point sequence section whose magnitude is larger than a predetermined value for the degree of bending of each point output in the subtraction unit,
The extreme value detection unit outputs the coordinates of a point that gives the extreme value of the degree of bending for each of the obtained sections as a bending point.
A shape processing method characterized in that:

The end point detection unit detects an end point of the shape from the profile data stored in the profile data storage unit, and stores the end point in the feature point data storage unit together with the inflection point obtained by the extreme value detection unit,
A second least-squares method implementation unit paying attention to one of the inflection points accumulated in the feature point data accumulation unit, extracting a feature point adjacent to the inflection point in question, and extracting the inflection point and the extracted inflection point If the number of point sequence data included between the extracted feature points is larger than a prescribed value, the point sequence data is extracted from the profile data storage unit, and a straight line or curve approximation is performed by the least square method. ,
Graphic extension processing unit, manipulating, for certain given along the approximate line or approximate curve extending in both directions of a line an approximate straight line or approximation curve obtained extending in the second least squares implementation unit run between the point sequence data in a strip-shaped range of the width is included in succession, and outputs it as a new inflection point coordinates instead the coordinates of the point the operation is completed the bending point coordinates before the extension operation,
The shape processing method according to claim 1 or 2 , wherein:

An end point detection unit detects an end point of the shape from the profile data stored in the profile data storage unit, and stores the end point together with the inflection point obtained by the extreme value detection unit as a feature point in the feature point data storage unit. And
A third least-squares method implementation unit paying attention to one of the inflection points accumulated in the feature point data accumulation unit, extracting a feature point adjacent to the inflection point of interest, and The point sequence data included between the extracted feature points is extracted from the profile data storage unit, and a straight line or curve approximation is performed by the least squares method.
The intersection detection unit obtains the intersection of the two approximate straight lines or approximate curves obtained by the third least squares method execution unit with respect to the noted bending point, and sets the intersection as a bending point. Output coordinates,
The shape processing method according to claim 1 or 2 , wherein:

The new bending point coordinates obtained by the figure extension processing unit are stored in the feature point data storage unit instead of the bending point coordinates before the extension operation,
A third least-squares method implementation unit paying attention to one of the inflection points accumulated in the feature point data accumulation unit, extracting a feature point adjacent to the inflection point of interest, and The point sequence data included between the extracted feature points is extracted from the profile data storage unit, and a straight line or curve approximation is performed by the least squares method.
The intersection detection unit obtains the intersection of the two approximate straight lines or approximate curves obtained by the third least squares method execution unit with respect to the noted bending point, and sets the intersection as a bending point. Output coordinates,
The shape processing method according to claim 4 , wherein:

Upon outputting the inflection point coordinates obtained by the intersection detection unit ,
The third least-squares method implementation unit obtains a variance value of an approximate straight line or an approximate curve used to obtain the inflection point,
It first kind reliability calculating unit, a high degree of confidence in the smaller of said dispersion value, a low confidence in the smaller of said variance value, and outputs a first reliability regarding the position of the bending point ,
The shape processing method according to claim 4 or 5 , wherein:

Upon outputting the inflection point coordinates obtained by the intersection detection unit,
Type 2 reliability calculation unit checks the number of points in the shape profile data present in limited extent with a predetermined distance in the vicinity of the bending point, a high value to those many number of the coefficients, the low value in with less number, and outputs a second confidence regarding the position of the bending point,
The shape processing method according to claim 4 or 5 , wherein:

Upon outputting the inflection point coordinates obtained by the intersection detection unit ,
The third least-squares method implementation unit obtains a variance value of an approximate straight line or an approximate curve used to obtain the inflection point,
A first type reliability calculating unit that outputs a high reliability to the small variance value and a low reliability to the large variance value as the first reliability related to the position of the inflection point;
The second type reliability calculation unit checks the number of points in the shape profile data existing in a range limited by a predetermined distance in the vicinity of the inflection point, and sets a high value to a large number of points. A low value is output as the second reliability regarding the position of the inflection point for the small number,
A total reliability calculating unit that inputs the first reliability and the second reliability, calculates a total reliability, and outputs the total reliability as a total reliability regarding the position of the bending point;
The shape processing method according to claim 4 or 5 , wherein: