JP4161766B2 - Pattern recognition device - Google Patents

Description

【００５９】
【表２】

【００６０】
平滑化・エッジ検出後のデータを検知ライン１１１、１１２中央で二分割して、前半・後半各３番目までの最大値を取る点を外周候補点１１３とする。この際、ライン中央からの距離に比例した補正値を加えた上で３番目までの最大値を求めることにより、硬貨２３の外周が写されている可能性の高い画像の外側、すなわちライン中央から離れた点を優先して選択でき、照明手段１５の照度にムラが発生し硬貨２３の外周が暗く写された場合などでも、硬貨の位置を精度良く検知できる。外周候補点１１３の一部を、図１３に白及び黒点で示す。ライン毎に前半の候補点と後半の候補点から１点ずつ選んだ２点の中点を合計９通り求める。６ライン分、合計５４通りの中点の度数分布から最も度数の高い点を仮中心座標とし、これを通過方向と平行及び垂直方向について求めて仮中心１１４とする。この仮中心１１４と各外周候補点１１３（全７２点）との距離を計算し、これらの距離の度数分布から最も度数の高い距離を求める。そして、求めた距離に対して一定の許容範囲内にある外周候補点１１３のうちで、同一検知ラインの前半と後半に１点ずつのみ属するような２点の組を選び出し、図１３に黒点で示す外周点１１５とする。例えば外周候補点１１６は、仮中心１１４との距離は許容範囲内であるが、同一の検知ライン１１２上に距離が許容範囲内の外周候補点１１３がないため、外周点１１５としない。最後に通過方向に平行な検知ライン１１１上の外周点１１５の通過方向に平行な平均座標と、通過方向に垂直な検知ライン１１２上の外周点１１５の通過方向に垂直な平均座標とを求めて中心とする。さらに、この中心と外周点１１５との平均の距離を求め、２倍して外径とする。本実施の形態では、外周に相当する複数の候補点を選択し、各候補点と仮の中心との距離で検証を行って、外周に相当すると判定した候補点から中心を求めることにより、確実に硬貨２３の位置を検知することができる。検知ライン１１１、１１２を複数本設定すると共に、平滑化・エッジ検出を行って各検知ライン１１１、１１２上の外周候補点１１３を、画像の外側を優先するよう補正をしながら複数選択することで、硬貨２３・通路１３の汚れや照明手段１５のムラの影響を受けにくい確実な中心・外径検知が可能になる。エッジ検出ではなくエッジ強調でも同様の効果が得られる。平滑化とエッジ検出または強調を（表１）と（表２）の配列データで同時に行うことで、高速に中心・外径を検知できる。なお、識別対象が円形以外の場合、例えば長方形の紙幣等に対しては中心や外周と共に長辺や短辺寸法を求めればよいが、これらに限らず特徴となる孔等の形状や模様などを基準点として検出し識別対象の位置を検知することもできる。
【００６１】
なお、複数の検知ラインで外周点を検知する技術は、例えば特開平９−１７８２１号や特開平１０−６３８５２号に開示されているが、本実施の形態では外周の候補点を複数選択し、仮中心との距離で検証を行って外周点を求めているため、図１３に示すように底面２２が写り込んだりノイズのある場合でも外径点を誤って選択することがないので、正確に中心と外径を検出できる。
【００６２】
金種仮判定（Ｓ４３）では、中心・外径検出手段８５から出力された外径から、投入された硬貨２３がどの金種に近いかを仮判定する。あらかじめ定められている各金種毎の外径基準値と記憶している撮像倍率とから、検知した外径に近い２金種を選び出して出力する。但し、あらかじめ定めた許容範囲内で該当する金種が１金種以下の場合は、その金種を選び出す。外径に基づけば、容易に金種を仮判定できる。１金種に絞らず２金種とすることで、硬貨２３が通路１３の中で撮像手段１６の近くを通過するか遠くを通過するかといった通過位置の影響で金種を誤判別するのを防止して識別精度を向上させることができる。例えば日本の百円硬貨（２２．６ｍｍ）と十円硬貨（２３．５ｍｍ）では約４％の外径差しかなく、特に広角レンズを使用した場合は通過位置の影響が大きく、例えば±０．５ｍｍの通過位置の変化で画像データ上の外径は±２％変化するため誤判別の恐れがある。また３金種以上に仮判定すると、以降のデータ処理、特に後述の模様照合工程（Ｓ２６）の回転角度検出（Ｓ５０）に多大な時間を要するため、最大２金種としてデータ処理時間が一定の上限時間を超えないようにしているが３金種以上としてもよい。
【００６３】
なお金種を１金種に確定して、その金種に応じた処理を行う技術は、例えば特開平１０−６３８５２号に開示されているが、本実施の形態では複数の金種に仮判定することで、硬貨２３の通過位置等の影響による金種の誤判別を防止できる。
【００６４】
続く凹凸計算（Ｓ４４）では、凹凸入力工程（Ｓ２３）で入力した第１及び第２の画像データから、凹凸検知手段８３で硬貨２３の凹凸度合いを検知する。各凹凸検知ライン１０８〜１１０について、第１の画像データと第２の画像データとの画素毎の差の絶対値を、その凹凸検知ライン上の全画素にわたって加算する。画素毎の差を求める際には、前述の硬貨２３移動の影響を打ち消すため、先に入力を行う第１の画像データの画素に対して、第２の画像データの１画素後ろ（出口１８側）の画素のデータとの差を求めている。このように、３本の凹凸検知ライン１０８〜１１０について計算を行い、３本のうちの最大値を出力する。凹凸検知ライン１０８〜１１０を複数本設定することで、硬貨２３や通路１３の汚れの影響を受けにくく、凹凸模様の少ない硬貨２３でも確実に凹凸度合いを検知できる。
【００６５】
さらに、模様照合工程（Ｓ２６）を行う。模様照合工程（Ｓ２６）では、多値デジタル値の画像データをエッジ検出（Ｓ４５）してから、撮像倍率の補正（Ｓ４６）を行いながら直交座標から極座標に座標変換（Ｓ４７）し、二値化（Ｓ４９）した上で、基準画像に対する投入された硬貨２３の画像データの回転角度を検出（Ｓ５０）して、検出した回転角度を考慮して基準画像との照合（Ｓ５１）を行う。これらの工程を、金種仮判定（Ｓ４３）で仮判定した２金種分の各表裏２面に対して行うが、例えばユーロ硬貨のように片面の模様が共通で、他面が１５ヵ国で異なる場合には、合計１６面分の処理を行えばよい。
【００６６】
模様照合工程（Ｓ２６）の各動作を、エッジ検出（Ｓ４５）から詳細に説明する。エッジ検出手段８４では、第３の画像データの各画素データに対して中心・外径検出（Ｓ４２）と同様に平滑化を兼ねたエッジ検出を行い、エッジ検出後の画像データを第４の画像データとして出力する。具体的には注目画素の周囲の８画素を含む９画素のデータについて、（表１）の配列データと配列要素毎に積算した９要素の和の絶対値と、（表２）の配列データと配列要素毎に積算した９要素の和の絶対値との和を求め、注目画素のエッジ検出後のデータとする。なおエッジ検出（Ｓ４５）は、中心・外径検出（Ｓ４２）で検出した中心を基準に、識別対象とする最大外径硬貨の外径範囲内に位置する画素に関して行えば十分である。画像データに対してエッジ検出（Ｓ４５）を行うことにより、反射光の明るさ（輝度）そのものではなく隣接部分との明るさの差で判定することができるので、画像データの明るさの全体的な変化や局所的なムラの影響を受けにくく、正確な識別が可能になる。エッジ検出ではなくエッジ強調を行うことでも、同様の効果を得られる。ここに画像データの明るさの全体的な変化や局所的なムラは硬貨２３の材質や色、通路１３又は硬貨２３の全体的な或いは一部の局所的な汚れなどによって発生するものであり、これらの影響を低減することができる。
【００６７】
次に座標変換（Ｓ４７）について、倍率補正（Ｓ４６）より先に述べる。座標変換手段８８では、第４の画像データを直交座標から極座標に変換し、座標変換後の画像データを第５の画像データとして出力する。本実施の形態では日本硬貨を識別対象として、最大外径の五百円硬貨がＮ×Ｎ画素（例えば１００×１００画素）となる光学設計を行っている。また、硬貨の通過位置や組み立て寸法及びレンズ２８等の部品特性のバラツキにより、撮像倍率を０．９から１．１まで想定している。すなわち五百円硬貨は製品毎また硬貨投入毎に、最小０．９Ｎ×０．９Ｎから最大１．１Ｎ×１．１Ｎ画素（例えば９０×９０から１１０×１１０画素）に撮像される場合が発生するが、簡略化のため最大７×７画素として座標変換（Ｓ４７）と倍率補正（Ｓ４６）の説明を行う。図１４（ａ）は第４の画像データを模式的に示しており、中心・外径検出（Ｓ４２）で検出した中心１２１のＸ座標１２２（硬貨２３の通過方向に平行な方向）、Ｙ座標１２３（硬貨２３の通過方向に垂直な方向）を（０，０）として、直交座標（−３，−３）から（３，３）までの４９画素分を示す。図１４（ｂ）は座標変換後の第５の画像データを模式的に示しており、極座標の半径方向（Ｒ座標１２４）に非等間隔に４分割して０〜３、円周方向（Ｓ座標１２５）に等間隔に８分割して０〜７とした場合の極座標（０，０）から（３，７）までを例示している。この座標変換は、（表３）及び（表４）の変換用テーブルを用いて行う。
【００６８】
【表３】

【００６９】
【表４】

【００７０】
（表３）は直交座標のＸ座標とＹ座標に対応した極座標のＳ座標を表しており、（表４）は直交座標のＸ座標とＹ座標に対応した極座標のＲ座標を表している。例えば、直交座標（０，０）の画素は極座標（０，０）に、直交座標（−１，−３）の画素は極座標（３，７）に変換される。このように直交座標の全画素を変換し、同一の極座標に複数の直交座標が該当する場合はこれらの画素の平均値を用い、該当する直交座標が１画素もない極座標は隣接する極座標のデータを複写するなど隣接データに基づいて設定する。極座標に座標変換することで、後述の回転角度検出（Ｓ５０）及び照合（Ｓ５１）が容易になると共に、データ量が削減され処理速度を高速化できる。なお極座標変換用の半径方向テーブル（表４）において、半径方向座標のＲ座標が最大値の３より大きい４となっている座標があるが、これは硬貨２３の外径範囲外であり以降使用しない画素であることを示すと共に、座標変換後の画像データを記憶するための配列のＲ座標を４まで用意しておくことで、座標変換時にＲ座標が３以下かどうか判定しながら変換する必要がなく、全画素を共通的に扱って高速に変換することができる。また、エッジ検出（Ｓ４５）した後に座標変換（Ｓ４７）を行うことで、硬貨２３の模様に忠実にエッジを検出し識別精度を向上することができる。なぜならば、極座標変換によって隣接する周囲の画素（座標）との距離が変化するために、座標変換（Ｓ４７）後にエッジ検出（Ｓ４５）を行うと、注目画素のエッジ検出または強調に用いる周囲の隣接画素との距離が注目画素の座標により、また隣接画素の座標により異なって一定にならないためである。
【００７１】
倍率補正（Ｓ４６）は、硬貨２３の通路１３中での通過位置や撮像手段１６の組み立て誤差等に起因する撮像倍率の補正を行う。この補正は、中心・外径検出（Ｓ４２）で検出した外径と金種仮判定（Ｓ４３）で仮判定した金種とに基づいて、極座標変換用の半径方向テーブル（表４）の値を更新することによって行う。半径方向テーブルは、（表５）のような各画素と中心との距離を示す距離テーブルと、（表６）のような各半径方向座標に対する距離の上限を示す上限距離テーブルとから求める。
【００７２】
【表５】

【００７３】
【表６】

【００７４】
例えば、Ｘ座標をｘ、Ｙ座標をｙとし距離テーブル値をＲ’（ｘ，ｙ）で表すと、Ｒ’（０，０）は０．００で１．７５未満であるから半径方向テーブル値Ｒ（０，０）を０とし、Ｒ’（３，１）は３．１６で３．０３以上３．５０未満であるからＲ（３，１）を３とする。なおＸ座標＜０及びＹ座標＜０の部分の距離テーブルは、データ記憶容量及び処理時間の削減のため対象性を利用して省略し、ｘ＜０かつｙ≧０の場合はＲ（ｘ，ｙ）＝Ｒ（−ｘ，ｙ）、ｘ≧０かつｙ＜０の場合はＲ（ｘ，ｙ）＝Ｒ（ｘ，−ｙ）、ｘ＜０かつｙ＜０の場合はＲ（ｘ，ｙ）＝Ｒ（−ｘ，−ｙ）の計算式で半径方向座標Ｒを求めている。
【００７５】
倍率補正（Ｓ４６）では、例えば金種仮判定（Ｓ４３）で仮判定した金種の１つが五百円で、中心・外径検出（Ｓ４２）で検出した外径（例えば９０）が五百円の標準外径（例えば１００）の０．９倍である場合、撮像倍率は０．９倍であるので、（表７）のように上限距離テーブルの上限距離の値を標準の０．９倍の値に更新する。
【００７６】
【表７】

【００７７】
そして、この上限距離テーブルと距離テーブル（表５）とから、極座標変換用の半径方向テーブルの値を更新すると（表８）となる。
【００７８】
【表８】

【００７９】
この変換用テーブルに基づいて座標変換することで、画像データを半径方向に１．１１倍に拡大しながら同時に座標変換が実現でき、容易に撮像倍率が補正される。倍率補正を行うことで、硬貨２３の通過位置や組み立て寸法・部品特性のバラツキに影響されずに正確な識別を行うことができる。また、製品毎に異なる基準画像を記憶させることなく、基準画像を共通化して製品毎の識別性能のバラツキを抑えることが可能であると共に、組み立てや部品特性の許容範囲の拡大も図れる。
【００８０】
本実施の形態では、上限距離テーブルの上限距離の値は半径が小さいほど間隔が広く半径が大きいほど間隔を狭く、非等間隔に設定している。（表９）に示したように等間隔に設定してもよい。
【００８１】
【表９】

【００８２】
しかし各極座標の面積に大きな差が生じ、座標変換時に同一の極座標に変換される直交座標の画素数が不均一になることにより、照合時に各極座標の重みに偏りが発生する。特に各極座標が等しい面積を持つように上限距離を設定すれば各極座標の重みが等しくなり、半径方向座標を非等間隔に設定することで識別性能を向上することができる。
【００８３】
なお、基準となる対象の画像を取得して倍率誤差を求め補正して画像切出を行う技術が特開平９−２４５２１３号に開示されているが、基準となる対象が固定された金種であり、調整モードでその金種だけが投入される場合は支障ないが、複数の金種が投入される場合には適用は困難である。これに対して、本実施の形態では複数の金種に仮判定することで、複数の金種が投入される場合にも適用できる。
【００８４】
続いて、二値化（Ｓ４９）のための閾値設定（Ｓ４８）を行う。二値化手段９０では、多値デジタル値である座標変換後の第５の画像データの各画素（座標）に対して、閾値以上であるかどうかによって１か０の二値デジタル値に変換し、二値化後の画像データを第６の画像データとして出力する。二値化（Ｓ４９）を行うことで、画像データにおけるノイズの影響を低減できると共に、画像データ量を削減し処理を高速化できる。また、エッジ検出（Ｓ４５）またはエッジ強調及び座標変換（Ｓ４７）を二値化（Ｓ４９）より前に行うことにより、多値データを用いた正確なエッジ検出または強調と座標変換が可能であり識別精度を向上できる。
【００８５】
この二値化（Ｓ４９）のための閾値を設定するのが閾値設定手段８９であり、第５の画像データと金種仮判定手段８６で仮判定した金種に基づいて設定した閾値を出力する。具体的には、二値化後の第６の画像データにおいて、値が１である画素数の全画素数に対する割合が金種と面（表面か裏面か）毎に定められた割合になるように、すなわち二値化後の度数分布が一定になるように閾値を設定する。これにより、二値化後の第６の画像データの画質を均一化すると共に、後述の模様照合（Ｓ５１）に用いる基準画像も同様の割合で作成することにより第６の画像と基準画像との均質化が可能であり、識別性能を向上できる。この割合を本実施の形態では全金種・面共通に５０％に設定しているが、金種・面毎に異ならせて各金種・面の模様に合った最適な二値化も可能であり、実験に基づくと３０〜７０％の割合において模様がくっきりと読みとれ高い識別性能が達成できた。閾値の求め方は、例えば第５の画像データからデータ値と画素数との度数分布（ヒストグラム）を作成し、データの大きい方からの累積度数が５０％となる値を求めればよい。また、閾値設定に用いる第５の画像データは、硬貨部分すなわち外径の内部に当然限るべきであり、倍率補正及び極座標変換を行ったことにより半径方向座標があらかじめ定められたその金種の半径以下である部分のみを使用すれば容易に可能である。さらに本実施の形態では、硬貨２３の状態や発行年度により硬貨２３毎に異なる外周・孔の縁、硬貨２３ではなく背景である孔の内部、及び同一の硬貨２３でも回転角度や位置により投入毎に異なる潜像・細線模様の部分も、閾値設定への使用から除外している。図１５は新五百円硬貨の基準画像（後述）に灰色で除外部分を上書きし直交座標に逆変換した模式図であり、外周の縁部１３１と潜像模様部１３２、及び細線模様部１３３を示す。外周・孔の縁部は摩耗やバリが発生しやすく、孔の内部は背景が見え、潜像・細線模様部は画像センサ２６の画素配列方向と硬貨の模様との角度により斑紋（モアレ）が現れ反射率が著しく変化し、閾値設定に使用するとこれら部分的な画像に影響されて全体の正確な二値化に支障をきたす。同様の理由で、硬貨毎に異なる模様である年号部分も除いてもよい。第５の画像データでどの部分を閾値設定に使用するかは、各金種・面毎に極座標形式で表した二値デジタル値のテーブルをあらかじめ記憶しておけば、各金種・面毎の模様に適した安定した二値化を行うことができる。また、この閾値設定に使用する部分が硬貨の中心に対して回転対称でない場合、硬貨の回転角度毎に閾値が変わるため、後述の回転角度検出（Ｓ５０）において画像データを回転させる度に閾値を求め直す多大な処理を要する。本実施の形態では閾値設定に使用する部分を、除外部分を含まずかつ硬貨の中心に対して回転対称すなわち同心円状に設定しているので、回転角度検出（Ｓ５０）において画像データを回転させる度に閾値を求め直す必要がなく、閾値設定に使用する部分を記憶するテーブルも半径方向の座標に対する１次元のデータ（以降、閾値設定用領域データという）で済む。このように、第６の画像データにおいて値が１である画素数の全画素数に対する割合が定められた割合になるように、すなわち二値化後の度数分布が一定になるように閾値を設定することで、硬貨２３の材質や、縁部や表面等の状態、及び回転角度、また照明手段１５や組み立て寸法のバラツキにおける照度の差に左右されることなく、安定した識別が可能になる。
【００８６】
なお、特定の閾値算出領域で二値化の閾値を設定する技術は、例えば特開２００２−１０９５９６号に開示されているが、二値化に支障をきたす硬貨内の特定部分（硬貨２３の状態や発行年度により硬貨２３毎に異なる外周・孔の縁、及び同一の硬貨２３でも回転角度や位置により投入毎に異なる潜像・細線模様）を除く技術については触れられていない。本実施の形態では、これらの領域を除く領域で閾値を設定することで、安定した閾値設定及び二値化が可能である。
【００８７】
回転角度検出（Ｓ５０）では、回転角度検出手段９１で第６の画像データと記憶手段９２に金種・面毎に記憶された基準画像とに基づいて、基準画像に対する投入された硬貨２３すなわち第６の画像データの回転角度を検出して出力する。逆に、第６の画像データに対する基準画像の回転角度を検出するようにしてもよい。なお基準画像データは、第６の画像データ同様に、極座標形式で表した二値デジタル値のテーブルとしてあらかじめ記憶されている。第６の画像データを回転させながら基準画像データと比較していき、最も一致する角度を検出する。第６の画像データは極座標で表されているので、回転させるには円周方向にずらす、すなわち円周方向座標を変えるだけで容易に回転させることができる。第６の画像データをＢ（ｒ，ｓ）（ｒは極座標の半径方向、ｓは円周方向）、基準画像データをＲ（ｒ，ｓ）、ａをｂで割った剰余をｍｏｄ（ａ，ｂ）、極座標における円周方向の分割数をＮｓで表すと、Δｓを０からＮｓ−１まで２ずつ変化させてＢ（ｒ，ｍｏｄ（ｓ＋Δｓ，Ｎｓ））とＲ（ｒ，ｓ）を比較していく。一致度合いは、第６の画像データと基準画像データの値が一致している画素（座標）数で判断する。第６の画像データと基準画像データとは共に二値デジタル値であるので、これらの排他的論理和をとり、結果が０である画素数で容易に判断可能である。Δｓを２ずつ変化させているのは検出時間を短縮するためであり、２ずつ変化させて最も一致度合いの高いΔｓを求める。そしてΔｓに対して隣接する回転角度、すなわちｍｏｄ（Δｓ−１，Ｎｓ）とｍｏｄ（Δｓ＋１，Ｎｓ）での一致度合いも求めて、３つの回転角度のうち最も一致度の高い回転角度を選択することにより、回転角度を間引きしても精度良く回転角度を検出できる。２ずつ変化させる方法に限らず、一致度合いが低ければ変化量を大きく、一致度合いが高ければ変化量を小さくするなど、回転角度を間引きして一致度合いを算出することで、精度を維持したまま回転角度検出の高速化が可能である。また本実施の形態では、閾値設定（Ｓ４８）において使用から除外した硬貨の外周・孔の縁、孔の内部、潜像・細線模様の部分を、閾値設定時と同じ理由により一致度合いの算出（回転角度検出及び照合）においても除外している。具体的には、回転角度検出（Ｓ５０）及び照合（Ｓ５１）に用いる画素（領域）を示すデータ（以下、照合領域データという）を、基準画像データや第６の画像データ同様の極座標形式で表した二値デジタル値のテーブルとしてあらかじめ記憶しておく。そして、第６の画像データと基準画像データとの排他的論理和と、この照合領域データとの論理積をとり、結果が１である不一致画素数が少ないほど一致度合いが高いと判断できる。なお、一致度合いを相対的ではなく絶対的に表すため、本実施の形態では照合領域データにおける値が１である画素数を照合画素数として、照合画素数から不一致画素数を減算し、この差を照合画素数で割って一致度合いとしている。これにより、照合画素数が異なっても、一致度合いを容易に比較することができる。
【００８８】
模様照合（Ｓ５１）では、回転角度検出手段９１で検出した回転角度における第６の画像データと基準データとの照合を行う。まず第６の画像データを回転角度検出手段９１で検出した回転角度分だけ円周方向にずらし、すなわち円周方向座標を変えて回転角度の補正を行い、第７の画像データとする。そして、この第７の画像データと基準画像データとの排他的論理和をとり、さらに回転角度照合で用いた照合領域データとの論理積をとって照合結果データとする。第７の画像データと基準画像データが一致しない画素（座標）の値は排他的論理和が１となり、さらに照合領域内であれば照合領域データとの論理積が１となるので、結果が１の画素数は不一致画素数を表す。また、照合領域データにおける値が１の画素数は照合画素数を表すので、照合画素数と不一致画素数との差を照合画素数で割った商は、第７の画像データと基準画像との一致度合いを示す。
【００８９】
本実施の形態では、硬貨全体の一致度合いを求めるのではなく、硬貨領域を複数の領域に分割し、各領域毎の一致度合いが一定値以上である領域の数すなわち一致領域数を求めている。一致度合いを一定値と比較することで、容易に一致領域数を求めることができる。具体的には図１６を用い半径方向画素数は３２、円周方向画素数１６の例で説明すると、図１６（ａ）に示す画像データに対して図１６（ｂ）の領域分割を示す模式図のように、極座標における半径方向１４１は半径方向画素数の約数で分割（この例の分割数は８）し、円周方向１４２も同様に円周方向画素数の約数で分割（この例の分割数は４）して、複数（この例では３２個）の領域１４３とする。極座標に基づくことで、容易に領域分割できると共に、等間隔に分割すれば各領域１４３の重みが均等になる。そして各領域毎に、その領域内に属する画素で一致度合いを求める。さらに領域毎の一致度合いがあらかじめ定めた下限値以上である領域数と、全領域数との割合を一致領域割合として、この値を模様照合手段９３では出力する。なお、各領域の一致度の算出に用いる情報量が不足している（照合領域が狭い、すなわち照合画素数が少ないなどの）領域１４３は一致度合いを正確に判定するのが困難であるので、照合画素数が一定値未満の領域１４３は対象外として、一致領域の判定の正確化を図っている。このように硬貨全体の模様の一致度合いを求めるのではなく、硬貨領域を複数の領域１４３に分割し各領域毎の一致度合いに基づいて照合を行うことにより、硬貨２３や通路１３の一部分に汚れが局所的に存在しても影響を受けにくく、正確な識別が可能である。また、一致領域と判定する領域毎の一致度合いの下限値や比較手段９７における一致領域割合の許容範囲を変えるだけで、汚れに対する耐性と偽貨排除性能の優先度合いを容易に変更することができる。例えば硬貨の１／３の部分が汚れなどで模様の正確な読み取りが不可能であっても識別可能とするには、一致領域割合の許容範囲を２／３以上とすればよい。
【００９０】
さらに、一致領域数は領域毎の一致度合いがあらかじめ定めた下限値以上である領域数としているが、この下限値は各金種・面毎に定めると共に、固定値とせず可変とすれば、硬貨２３や通路１３の汚れの影響を低減して正確な識別が可能になる。なぜならば、硬貨２３や通路１３に汚れがある場合は、一致度合いが低下し一致領域数が減少して正貨と判定されない可能性が高まる。これは、照合領域全体の一致度合いが回転角度検出におけるどの回転角度でもほぼ一様に低下するためであり、これに対して例えば照合領域全体の一致度合いが最小となる回転角度での一致度合いを用いて、最大の一致度合いと最小の一致度合いとの差を利用する、または最小の一致度合いに一定の値を加えて下限値とすることなどで改善が可能である。この例では正確な識別が困難なほど汚れが著しい場合に、最大の一致度合いが非常に低い値になり誤識別の恐れがある。しかし、最大の一致度合いの絶対値が低いと正貨と判定しないという下限値をあらかじめ定めたり、または最小の一致度合いに一定の値を加えて算出する下限値があらかじめ定めた一定値以下であれば下限値をこの一定値とすることで、誤識別が発生するのをさけることができる。
【００９１】
模様照合（Ｓ５１）は、まず金種仮判定（Ｓ４３）で近似していると仮判定した第１の金種の表面について、その基準データ及び照合領域データに基づいて行う。次に、閾値設定用領域データは面毎に異なるので、閾値設定（Ｓ４８）から模様照合（Ｓ５１）までを、裏面について同様に繰り返す。さらに、倍率補正は金種毎に異なるので、倍率補正（Ｓ４６）からＳ５２までを、金種仮判定（Ｓ４３）で近似していると仮判定した第２の金種について行う。従って、模様照合手段９３は２金種×２面分の一致領域割合を出力する。
【００９２】
比較（Ｓ６１）では、各比較手段９４〜９７のそれぞれに入力される厚み検知手段７４、材質検知手段７８、凹凸検知手段８３、模様照合手段９３の各出力を記憶手段９２に記憶された基準と比較し、許容範囲内で一致していればその正貨の種類を示す信号を出力し、どの種類の基準値とも一致しない場合には偽貨であることを示す信号を出力する。判定（Ｓ６２）は、判定手段９９に入力される比較手段９４〜９７からの信号が全て同じ正貨の種類を示す場合に限りその正貨の種類を示す信号を出力し、それ以外の場合には偽貨を示す信号を判定信号１００として出力する。
【００９３】
厚みセンサで投入検知（Ｓ１６でＹＥＳ）した場合や材質センサで投入検知（Ｓ１７でＹＥＳ）した場合、すなわち結露検知中の場合の比較（Ｓ７１）は、撮像手段１６に関連した凹凸検知手段８３及び模様照合手段９３が接続された比較手段９６、９７を除く、各比較手段９４、９５のそれぞれに入力される厚み検知手段７４、材質検知手段７８の各出力を記憶手段９２に記憶された基準と比較し、許容範囲内で一致していればその正貨の種類を示す信号を出力し、どの種類の基準値とも一致しない場合には偽貨であることを示す信号を出力する。判定（Ｓ７２）でも同様に撮像手段１６に関連した比較手段９６、９７を除き、判定手段９９に入力される比較手段９４、９５からの信号が全て同じ正貨の種類を示す場合に限りその正貨の種類を示す信号を出力し、それ以外の場合には偽貨を示す信号を判定信号１００として出力する。
【００９４】
判定手段９９にてＳ６２とＳ７２のいずれの判定を行うかを判断するため、判定手段９９には結露検知手段８２の出力が接続されている。また、判定手段９９にはスイッチ等の切替手段９８の出力も入力されており、結露検知中の判定を撮像手段１６に関連した識別項目（凹凸及び模様）を除いた識別項目（厚みと材質）だけで行って、結露中も硬貨の受け付けを優先し自動販売機等としての販売機会を逃すことを防ぐか、それとも結露中は撮像手段１６での識別が困難であるため投入された硬貨を強制的に受け付けないで偽貨排除性能を優先するかを、図９には図示しないが切替手段９８で選択が可能である。このように切替手段９８を備えることにより、硬貨の受け付けを優先するか偽貨排除性能を優先するかを選択し、なおかつ自動販売機等の設置現場でも容易に選択を変更することができる。また、図示しないが発熱手段を用いた結露防止手段を備え、結露検知時にこの結露防止手段を動作させることで、消費電力を節約しながら常に撮像手段１６に関連した識別も行うことが可能である。発熱手段としては、専用の部品を用いることなく発熱量の大きい電子部品を兼用して通路１３の近傍に配置したり、安価なヒーターを用いることができる。
【００９５】
判定（Ｓ６２）において図９には図示しないが、撮像手段１６に関連しない識別項目（本実施の形態では厚み及び材質）では同じ種類の正貨であって、撮像手段１６に関連した識別項目（凹凸と模様）では偽貨が異なる種類の正貨であることが続いた場合、通路１３の汚れなどで画質が低下したと判断する。また全ての識別項目で同じ種類の正貨であることが続いた場合は、画質に問題ないと判断することで、画質の低下度合いを検知できる。画質の低下を検知した場合、自動販売機の制御部に対しては通信で、自動販売機の操作者に対しては表示を行って画質が低下したことを報知し、清掃を促すことが可能である。
【００９６】
さらに、切替手段９８とは別に第２の切替手段（図示せず）により、撮像手段１６での識別（以下、画像識別という）そのものを行うか行わないかを選択可能としている。第２の切替手段を備えることにより、撮像手段１６や光学透明体１４に故障や重大な傷・汚れがあるような場合に画像識別を行わない方を選択し、結露検知（Ｓ１４）と撮像手段１６での投入検知（Ｓ１５）及び凹凸及び模様検知（Ｓ２１）を行わない（図９には図示せず）ことで、誤動作による製品の動作停止や識別性能の低下を防止することができる。
【００９７】
なお、硬貨投入の検知方法や、図２と図４とを用いて説明した通路構成、及び結露検知や結露防止や画質検知は、従来構成の模様識別装置にも適用可能である。それぞれ、確実な投入検知、汚れの影響を受けにくい確実な識別、結露や汚れ等の影響を低減した正確な識別が可能である。
【００９８】
以上のように、本実施の形態によれば、硬貨２３からの正反射光２９の非入射部分で識別を行うことにより、画像にムラが発生するのを防止できる。また、撮像手段１６を通路１３に対して鉛直上方側に設けることにより、撮像手段１６への汚れの付着を低減させることができる。撮像手段１６は部分的な画像を出力可能な画像センサ２６からなるので、特定領域６６の画像データを短時間で読み出すことができる。前記撮像手段１６は行列状に配置された受光素子のアレイと、前記アレイの行に感度制御のための電圧を供給する制御回路６８と、前記アレイの列から接地へ流れる電流を処理する神経ネットワーク６９とを備える画像センサ２６からなるので、画像センサ２６内でフィルタ行列と画像の積和演算を実行できる。撮像手段１６近傍の通路１３が結露しているかどうかを検知する結露検知手段８２を備えることにより、通路１３が結露しているかどうかを検知できる。そして、撮像手段１６の制御処理を省いて行わないことで正常でない画像データによる誤動作を防止したり、結露防止手段を動作させたり、結露の発生を報知することが可能である。硬貨２３が投入されたことを撮像手段１６を用いて検知することにより、硬貨２３が投入されたことを確実に検知することができる。撮像手段１６で事前に読み取った画像に基づいて読み取り条件の調整を行うことにより、一定画質の画像データを得ることができる。ほぼ同一の波長特性を有し異なる方向から硬貨２３に光を照射する第２の照明手段３６と及び第３の照明手段３７を備え、第２の照明手段３６で照射した際の硬貨２３からの反射光と、第３の照明手段３７で照射した際の硬貨２３からの反射光との差に基づいて硬貨２３を識別することにより、硬貨２３の凹凸を検知することができる。第３の画像における硬貨２３の位置の基準となる点（中心１１４）を複数の外周候補点１１６を検証して検知することにより、確実に硬貨２３の位置を検出することができる。１種類以上の金種に仮判定して識別を行うことにより、硬貨２３の通過位置の影響で金種を誤判別するのを防止できる。撮像手段１６を通路１３の鉛直上方側に設けることにより、撮像手段１６などへの汚れの付着を低減させることができる。第３の画像のエッジ検出強調後に座標変換を行って識別を行うことにより、硬貨２３の模様に忠実にエッジを検出することができる。硬貨２３の特定部分を除く画像に基づいて閾値を設定し、この閾値で二値化を行うことにより、安定した二値化を行うことができる。硬貨２３の特定部分を除く画像に基づいて基準画像との照合を行うことにより、安定した照合を行うことができる。回転角度を間引いて回転角度の検出を行うことにより、精度を維持したまま高速に回転角度検出を行うことができる。硬貨２３の画像を複数の領域１４３に分割し、基準画像との各領域毎の一致度合いに基づいて識別を行うので、硬貨２３や通路１３の汚れの影響が低減される。
【００９９】
そのため、硬貨の模様を正確に識別することができるので、偽貨の不正使用を防止することが可能な、識別性能の高い模様識別装置を実現できる。
【０１００】
【発明の効果】
以上のように本発明によれば、通路と、前記通路を通過する識別対象に光を照射する照明手段と、前記識別対象からの反射または透過光を受光する撮像手段とを備え、前記撮像手段で取得した画像に基づいて前記識別対象の模様を識別する模様識別装置において、前記撮像手段では、少なくとも前記識別対象の撮像と、撮像した画像に基づいて前記撮像手段の入力条件の調整とを行う調整工程の後、さらに同じ撮像手段で、前記識別対象を撮像し、前記識別対象の模様を識別することを特徴とする模様識別装置としたものであるため、模様を正確に識別でき、識別性能を向上することが可能である。
【図面の簡単な説明】
【図１】本発明の一実施の形態における硬貨識別装置の概要を示した正面図
【図２】同、撮像手段近傍の断面図
【図３】同、照明手段近傍の正面図
【図４】同、通路の断面図
【図５】同、厚み・材質兼用センサ近傍の断面図
【図６】同、画像センサの制御回路の構成図
【図７】同、画像センサの制御回路の構成図
【図８】同、制御回路の構成を示すブロック図
【図９】同、動作について説明するフローチャート
【図１０】同、凹凸及び模様検知の動作を説明するフローチャート
【図１１】同、撮像手段の撮像領域の概念図
【図１２】（ａ）同、第２の照明手段のみで照明した場合の硬貨近傍の断面図
（ｂ）同、第２の照明手段のみで照明した場合の画像データの模式図
（ｃ）同、第３の照明手段のみで照明した場合の硬貨近傍の断面図
（ｄ）同、第３の照明手段のみで照明した場合の画像データの模式図
【図１３】同、中心・外径検出について説明する第３の画像データの模式図
【図１４】（ａ）座標変換と倍率補正について説明する第４の画像データの模式図
（ｂ）座標変換と倍率補正について説明する第５の画像データの模式図
【図１５】同、閾値設定について説明する、新五百円硬貨の基準画像を直交座標に逆変換した模式図
【図１６】（ａ）模様照合について説明する第６の画像データの模式図
（ｂ）模様照合の領域分割について説明する第６の画像データの模式図
【符号の説明】
１１ 硬貨識別装置本体
１２ 投入口
１３ 通路
１４ 光学透明体
１５ 照明手段（第１の照明手段）
１６ 撮像手段
１７ 厚み・材質兼用センサ
１８ 出口
２１ 側壁
２２ 底面
２３ 硬貨
２４ 側壁
２５ ＬＥＤ
２６ 画像センサ
２７ 受光面
２８ レンズ
２９ 正反射光
３１ 硬貨位置
３２ 硬貨位置
３３ 撮像時硬貨位置
３４ 正反射位置
３５ 直線
３６ 第２の照明手段
３７ 第３の照明手段
４１ 排出補助手段
５１ コア
５２ コア
５３ コイル
５４ コイル
５５ コイル
５６ コイル
６１ 画素
６２ 水平走査回路
６３ 垂直走査回路
６４ 水平走査線
６５ 垂直走査線
６６ 特定領域
６７ 出力
６８ 制御回路
６９ 神経ネットワーク
７１ 厚みセンサ
７２ 発振回路
７３ 整流回路
７４ 厚み検知手段
７５ 材質センサ
７６ 発振回路
７７ 整流回路
７８ 材質検知手段
７９ 投入検知手段
８０ Ａ／Ｄ変換手段
８１ 調整手段
８２ 結露検知手段
８３ 凹凸検知手段
８４ エッジ検出手段
８５ 中心・外径検出手段
８６ 金種仮判定手段
８７ 倍率補正手段
８８ 座標変換手段
８９ 閾値設定手段
９０ 二値化手段
９１ 回転角度検出手段
９２ 記憶手段
９３ 模様照合手段
９４ 比較手段
９５ 比較手段
９６ 比較手段
９７ 比較手段
９８ 切替手段
９９ 判定手段
１００ 判定信号
１０１ 撮像範囲
１０２ 第１の投入検知領域
１０３ 第２の投入検知領域
１０４ 第３の投入検知領域
１０５ 硬貨位置
１０６ 入力条件設定領域
１０７ 硬貨位置
１０８ 凹凸検知ライン
１０９ 凹凸検知ライン
１１０ 凹凸検知ライン
１１１ 検知ライン
１１２ 検知ライン
１１３ 外周候補点
１１４ 仮中心
１１５ 外周点
１１６ 外周候補点
１２１ 中心
１２２ Ｘ座標
１２３ Ｙ座標
１２４ Ｒ座標
１２５ Ｓ座標
１３１ 外周の縁部
１３２ 潜像模様部
１３３ 細線模様部
１４１ 半径方向
１４２ 円周方向
１４３ 領域[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a pattern identification device that reads a pattern such as a coin and performs determination or recognition of authenticity or type.
[0002]
[Prior art]
In recent years, vending machines have become widespread, and coin recognition devices used for them are required to have high identification performance.
[0003]
Conventionally, this type of coin discriminating apparatus includes a magnetic sensor composed of a coil and a ferrite core, and identifies the coin by detecting the material, thickness, and outer diameter of the coin.
[0004]
As prior art document information related to the invention of this application, for example, Patent Document 1 is known.
[0005]
[Patent Document 1]
JP-A-8-161574 [0006]
[Problems to be solved by the invention]
However, in such a conventional configuration, what is identified is the material, thickness, and outer diameter of the coin, and it is difficult to detect the pattern of the coin. For this reason, there are cases in which false coins with different patterns are illegally used. For example, it is a fake coin that is made by cutting the thickness of a similar foreign coin whose material and outer diameter are almost equal.
[0007]
An object of the present invention is to provide a pattern identification device with high identification performance that can accurately identify a pattern and prevent such illegal use of fake coins.
[0008]
[Means for Solving the Problems]
In order to achieve this object, a pattern identification apparatus according to the present invention includes a passage, illumination means for irradiating light to an identification target passing through the passage, and imaging means for receiving reflected or transmitted light from the identification target. And a pattern identification device for identifying a pattern to be identified based on an image acquired by the imaging unit, wherein the imaging unit includes at least imaging of the identification target and input conditions of the imaging unit based on the captured image. After the adjustment step of performing the adjustment, the pattern identification device is characterized in that the identification target is further imaged by the same imaging means to identify the pattern of the identification target .
[0009]
As a result, the pattern can be accurately identified, and the identification performance can be improved.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
The invention according to claim 1 of the present invention includes a passage, illumination means for irradiating light to the identification target passing through the passage, and imaging means for receiving reflected or transmitted light from the identification target, In the pattern identification apparatus for identifying the pattern to be identified based on the image acquired by the imaging unit, the imaging unit is configured to capture at least the image of the identification target and adjust input conditions of the imaging unit based on the captured image. After the adjustment process, the pattern identification device is characterized in that the identification target is further imaged by the same imaging means to identify the pattern of the identification target , and image data with a constant image quality can be obtained. The pattern can be accurately identified.
[0011]
The invention according to claim 2 of the present invention is determined in advance so that the imaging of the identification target in the adjustment step includes the coin position of the assumed minimum speed and the assumed maximum speed of the coin. 2. The pattern identification apparatus according to claim 1, wherein a predetermined partial image is used for adjusting the input condition of the image pickup means, and can be adjusted at high speed.
[0012]
According to a third aspect of the present invention, in the adjustment step, the input condition of the image pickup apparatus is adjusted by discriminating the brightness of the image obtained by imaging the identification target for each pixel. This is a pattern identification device, which makes it possible to easily adjust the input conditions of the imaging device by counting the number of bright pixels and the number of dark pixels.
[0013]
According to a fourth aspect of the present invention, in the adjustment step, the adjustment amount of the input condition of the imaging device is stored every time the identification target passes, and the identification target passes next based on the stored adjustment amount. The pattern identification apparatus according to claim 1, wherein an adjustment amount of an input condition at the time of correction is corrected, and the influence of a change with time such as component deterioration can be reduced and the identification performance can be maintained.
[0026]
(Embodiment)
Hereinafter, embodiments of the present invention will be described with reference to FIGS. Although the pattern identification device in the present embodiment is intended to identify a coin stamp pattern, the present invention is applicable to a pattern such as a printed pattern such as a banknote or a card, a stamp pattern such as a medal or token, etc. as an identification target. It is. The pattern can be applied not only to the front and back surfaces of coins and the like, but also to the side pattern. Furthermore, the object to be identified is not limited to rolling using gravity as in the present embodiment, but during natural fall, during movement by a conveying means such as a belt, and further by using a stopping means. Identification can be performed even when stopped.
[0027]
In the embodiment of the present invention, the imaging unit and the illumination unit are arranged on the same side of the passage, and identification is performed by reflected light from the identification target. However, the imaging unit and the illumination unit are arranged on the opposite side of the passage. The identification can also be performed with the transmitted light from the identification target.
[0028]
FIG. 1 is a front view showing an outline of a coin identifying device according to an embodiment of the present invention, and identifies the authenticity and type of the inserted coin. In FIG. 1, a coin insertion slot 12 is provided at the upper part of the coin identification device main body 11, and a coin passage 13 is connected downward from the insertion slot 12. A part of the side wall of the passage 13 is constituted by an optical transparent body 14, and an illumination means 15 and an imaging means 16 for detecting a coin pattern are arranged on the back of the optical transparent body 14. Further, a thickness / material sensor 17 is disposed downstream of the passage 13. Further, the passage 13 is connected to a coin outlet 18 located in the lower part of the coin identification device main body 11.
[0029]
FIG. 2 is a cross-sectional view of the vicinity of the imaging means 16. A part of one side wall 21 constituting the passage 13 forms a bottom surface 22 of the passage. The inserted coins 23 roll on the bottom surface 22. A part of the other side wall 24 of the passage 13 is composed of the optical transparent body 14, and the coin 23 is illuminated with illumination light from the illumination means 15 including a plurality of LEDs 25 and a flexible wiring board. The reflected light reflected by the coin 23 is detected by the imaging means 16 arranged on the side wall 24 side. The imaging unit 16 includes an image sensor 26, a light receiving surface 27 on the image sensor 26, and a lens 28 that forms an image of reflected light from the coin 23 on the light receiving surface 27.
[0030]
The bottom surface 22 of the passage 13 is provided on the side wall 21 opposite to the illumination unit 15 and the imaging unit 16, thereby preventing a shadow from being formed on the bottom surface 22 and lowering the identification performance. In addition, the width 22a of the bottom surface 22 is made smaller than the minimum thickness of the coin 23 to be identified so as not to prevent the illumination light from the illumination means 15 from being applied to the edge of the coin 23, that is, not to be a shadow. The dimension is set such that the coin 23 having the minimum thickness does not fall even if it passes along the side wall 24 side.
[0031]
Further, the passage 13 is inclined with respect to the vertical so that the coin 23 passes stably along one of the side walls 21 and 24. The side wall 24 opposite to the inclined side is not always necessary. In the present embodiment, the optical transparent body 14, the illumination means 15, and the imaging means 16 are arranged on the side opposite to the inclined side to reduce the adhesion of dirt to them. The optical transparent body 14 is formed of glass, plastic, or the like, so that the passage 13 side and the imaging means 16 side can be separated, and dust or coins 23 can be prevented from entering the imaging means 16 side from the passage 13 side. The optical transparent body 14 may be air, that is, nothing may be used.
[0032]
The side wall 21 and the side wall 24 are black in order to block disturbance light and reduce the reflected light of the illumination light from the illumination means 15 other than the coins 23. Not only black but also non-reflective or low-reflective material, color, and surface state, so that disturbance light other than illumination light from the illumination means 15 and reflected light other than the coins 23 of illumination light from the illumination means 15 Can be prevented from being detected by the imaging means 16, and the identification accuracy can be improved.
[0033]
The illumination means 15 is configured as shown in FIG. That is, a range including all of the coin position 31 passing through the passage 13 at the assumed minimum speed to the coin position 32 passing through the assumed maximum speed is defined as the coin position 33 at the time of imaging, and the coin position 33 at the time of imaging is completely set. It is configured to cover the shape. The shape of the illuminating means 15 is a non-circular shape having a long side and a short side, for example, an oval shape or an elliptical shape. By coinciding the long side with the passing direction of the coin 23, the coin 23 is a light receiving surface of the image sensor 26. The tolerance for the accuracy of detecting the timing at which the position corresponding to the center of 27 is detected is expanded, and it becomes possible to discriminate even during free fall or rolling where the passing speed of the coin 23 is not constant. Moreover, illumination can be made uniform by making the center of the illumination means 15 substantially coincide with the center of the coin position 33 during imaging. Further, the center of the light receiving surface 27 on the image sensor 26 is also arranged at these centers.
[0034]
In FIG. 2, the regular reflection light 29 is the regular reflection of the illumination light from the illuminating means 15 on the plane portion near the outer peripheral edge of the coin 23 having the maximum outer diameter to be identified, that is, the incident angle and the reflection angle are equal. The reflected light component is shown. When this regular reflection light 29 is incident on the image pickup means 16, a very bright unevenness is generated in the input image, which prevents accurate identification. Therefore, the passage 13, the illumination unit 15, and the imaging are arranged so that the regular reflection light 29 does not enter the imaging unit 16, in other words, in such a positional relationship that the regular reflection position 34 does not fall within the range of the coin position 33 during imaging. The means 16 is arranged to prevent the occurrence of unevenness in the image at the coin position 33 during imaging. By making the shape of the illumination means 15 a non-circular shape having a long side and a short side, the shape of the regular reflection position 34 can be a non-circular shape having a long side and a short side, and within the range of the coin position 33 at the time of imaging. It can be difficult to enter. Furthermore, by making the long side of the illumination means 15 coincide with the passing direction of the coin 23, the long side of the regular reflection position 34 in the imaging means 16 can be made substantially parallel to the passing direction of the coin 23. By making the center coincide with the center of the coin position 33 at the time of imaging in the imaging means 16, the center of the regular reflection position 34 can be substantially coincident with the center of the coin position 33 at the time of imaging. Alternatively, a missing portion is provided in the regular reflection position 34 by having a missing portion in which the LED 25 is not disposed in the portion 15a (long side portion) of the illumination means 15 having the regular reflection position 34a that falls within the range of the coin position 33 during imaging. Can do. In any case, there is an effect that the passage 13, the illumination unit 15, and the imaging unit 16 can be easily arranged so that the regular reflection light 29 from the coin 23 is incident outside the coin position 33 at the time of imaging in the imaging unit 16. The same effect can be obtained by matching the center of the illumination unit 15 with the center of the imaging unit 16. When identification is performed using transmitted light, a transmitted light component (referred to as regular transmitted light) having the same incident angle and emission angle corresponds to regular reflected light 29 when identification is performed using reflected light, and unevenness of an image. The same can be said about.
[0035]
In order to detect the marking pattern of the coin 23, the incident angle of the irradiation light is set to a low angle with respect to the coin 23, and dark field illumination is performed to make the marking more prominent. The plurality of LEDs 25 constituting the illuminating means 15 (first illuminating means) are divided into two groups by a straight line 35 passing through the center of the illuminating means 15 and perpendicular to the passage 13, that is, the second illuminating means 36 on the outlet 18 side. And the third illumination means 37 on the inlet 12 side, and the second illumination means 36 and the third illumination means 37 can emit light independently. By dividing the illumination means 15 into two groups and causing each illumination means 36 and 37 to emit light independently, the degree of unevenness of the coin 23 can be detected by a method described later.
[0036]
FIG. 4 is a sectional view of a part of the passage 13. A discharge assisting means 41 is provided on the other side wall 24 located on the opposite side of the one side wall 21 having the bottom surface 22. When a deformed coin, a coin wet with rain or the like, or a coin with a sticky substance attached thereto is inserted and stopped in the passage 13, the distance between the one side wall 21 and the other side wall 24 is increased, whereby the coin 23 is Although it is dropped downward and returned from the outlet 18, the discharge assisting means 41 ensures the discharge of the coin 23 by holding the upper part of the coin 23 on the side wall 24 side.
[0037]
FIG. 5 is a cross-sectional view of the vicinity of the thickness / material sensor 17. E-type cores 51 and 52 made of a ferrite material are attached to one side wall 21 and the other side wall 24 constituting the passage 13 so as to face each other. The E-type cores 51 and 52 have a length smaller than the outer diameter of the smallest coin 23 to be identified, and are arranged at a position where the center of the coin 23 passes near the center of the E-type cores 51 and 52. ing. The combined thickness / material sensor 17 includes an E-type core 51, two coils 53 and 54 wound inside the E-type core 51, and an E-type core 52 and two coils 55 and 56 wound therein. . The coil 53 and the coil 55 are connected in series in reverse phase so that the mutual inductance becomes negative. By arranging the two coils 53 and 55 opposite to the passage 13 and connecting them in reverse phase, the magnetic flux direction of the alternating magnetic field generated in the passage 13 is parallel to the surface of the coin as shown by the broken line in FIG. Thus, the thickness of the coin can be sensitively detected. Further, the coil 54 and the coil 56 are connected in series and in phase so that mutual inductance becomes positive. By arranging the two coils 54 and 56 facing the passage 13 and connecting them in phase, the magnetic flux direction of the alternating magnetic field generated in the passage 13 is changed to the surface of the coin as shown by a two-dot chain line in FIG. It becomes vertical and can detect the material of coins sensitively. The thickness sensor 71 and the material sensor 75 are used for the purpose of downsizing and cost reduction, but separate sensors may be used.
[0038]
FIG. 6 is a block diagram of the control circuit of the image sensor 26 forming the imaging means 16. In this embodiment, a MOS type two-dimensional (area) image sensor is used as the image sensor 26 to obtain a two-dimensional image at high speed. However, the present invention is not limited to this, and time series data of a CCD or a one-dimensional (line) sensor may be used. The image data detected by each pixel 61 composed of the light receiving element of the image sensor 26 selects the pixel 61 by a horizontal scanning line 64 and a vertical scanning line 65 controlled by a horizontal scanning circuit 62 and a vertical scanning circuit 63 which are a kind of shift register. Then, it is sequentially output 67 from the horizontal scanning circuit 62. At this time, by controlling only a part of the scanning lines by the horizontal scanning circuit 62 and the vertical scanning circuit 63, it is possible to output only the image data of the specific area 66. Output of partial image data can be realized by a MOS type image sensor, and by providing such an imaging means 16, it is possible to speed up reading by narrowing the output of image data to only a necessary portion. Is possible. In addition, each pixel 61 has a square shape, and uses the image sensor 26 in which the number of pixels in the horizontal direction is larger than the number of pixels in the vertical direction and the detection area is rectangular, and the longitudinal direction of the image sensor 26 in the passing direction of the coins 23. Are arranged so that they substantially coincide with each other to obtain rectangular image data. The reason why the longitudinal direction is made coincident with the passing direction is the tolerance for the accuracy of detecting the timing when the coin 23 comes to the position corresponding to the center of the light receiving surface 27 of the image sensor 26 by effectively using the pixels of the image sensor 26. The image sensor 26 performs scanning of the horizontal scanning circuit 62 to which the output 67 is connected in preference to scanning of the vertical scanning circuit 63 to which the output 67 is not connected (for horizontal scanning). This is because the horizontal pause period (horizontal blanking period) in the output signal is small to detect an image in a horizontally long region, and the image data can be read out in a shorter time.
[0039]
In the present embodiment, as shown in the block diagram of the control circuit in FIG. 7, the image sensor 26 in the imaging unit 16 includes a control circuit 68 that supplies a voltage for sensitivity control to the rows of the pixel matrix, and a pixel matrix. And a neural network 69 for processing the current flowing from the row to the ground. Since the details of the control circuit 68 and the neural network 69 are described in Japanese Patent Laid-Open No. 6-139361 (Example 6), the description of the configuration and operation is omitted. For example, a product-sum operation of a 3 × 3 filter matrix and an image can be executed and output. That is, the function of the edge detection means 84 and the edge enhancement function described later can be realized in the image sensor 26. Arithmetic operations based on adjacent surrounding pixel data can be executed particularly quickly. It is natural that the original image can be output as it is, instead of the edge image, by changing the filter setting. Since the image sensor 26 is provided with the control circuit 68 and the neural network 69 in this manner, the product-sum operation of the filter matrix and the image can be executed in the image sensor 26, so that the edge detection means 84 is not necessary and the edge detection is performed. Processing time can be shortened and data processing can be speeded up. The image sensor 26 with a built-in edge detection function has both the functions of the image pickup means 16 and the edge detection means 84, and the image pickup means 16 and the edge detection means 84 may be separate. In the description, the imaging unit 16 and the edge detection unit 84 will be described separately.
[0040]
FIG. 8 is a block diagram showing the configuration of the control circuit. The thickness sensor 71 is connected to the oscillation circuit 72 and input to the thickness detection means 74 via a rectifier circuit 73 that converts the oscillation waveform from a sine wave to a signal indicating the oscillation level. Has been. The material sensor 75 is connected to the oscillation circuit 76 and is input to the material detection means 78 via the rectifier circuit 77. The outputs of the rectifier circuits 73 and 77 are also input to the input detecting means 79 for detecting that the coin 23 has been input. The input detection means 79 controls the illumination means 15 and the imaging means 16, detects the reflected light irradiated from the illumination means 15 and reflected by the coin 23, and detects the captured analog image data as a multivalued digital signal. It outputs to the A / D conversion means 80 to convert. The output of the A / D conversion means 80 includes an input detection means 79, an adjustment means 81 for controlling and adjusting the imaging means 16, a dew condensation detection means 82 for detecting whether or not the passage 13 in the vicinity of the imaging means 16 is condensed, a coin 23 is input to the unevenness detecting means 83 for detecting the degree of unevenness of the image 23, the edge detecting means 84 for detecting the edge portion of the image data, and the center / outer diameter detecting means 85 for detecting the center and outer diameter of the coin portion in the image data. ing. The center / outer diameter detection means 85 is connected to the edge detection means 84, the denomination temporary determination means 86 and the magnification correction means 87, and the denomination temporary determination means 86 detects the value of the outer diameter detected by the center / outer diameter detection means 85. Tentatively determine up to two types of denominations that are considered to be applicable. The output of the edge detection means 84 enters the coordinate conversion means 88 for converting the image data from orthogonal coordinates to polar coordinates, and the magnification correction means 87 is caused by the passage position of the coin 23 in the passage 13, the assembly error of the imaging means 16, and the like. The imaging magnification is corrected and the coordinate conversion means 88 is controlled. The output of the coordinate conversion unit 88 is input to a threshold setting unit 89 that sets a binarization threshold and a binarization unit 90 that converts image data from a multi-value digital value to a binary digital value. The output of the binarization means 90 is connected to the rotation angle detection means 91, detects the rotation angle of the image data of the inserted coin 23 with respect to the reference image data stored in the storage means 92, and The image data and the reference image data are output to the pattern matching means 93 for matching. The output of the denomination temporary determination means 86 is connected to the magnification correction means 87, the threshold setting means 89, the rotation angle detection means 91, and the pattern matching means 93, and the storage means 92 includes the rotation angle detection means 91, the pattern matching means 93, And it is connected to each comparison means 94-97. Each of the comparison means 94 to 97 receives the outputs of the thickness detection means 74, the material detection means 78, the unevenness detection means 83, and the pattern matching means 93. The comparison means 94 to 97 compare with the reference stored in the storage means 92. If they match within the allowable range, a signal indicating the type of the genuine coin is output, and if it does not match any type of reference value, a signal indicating that it is a false coin is output. The determination means 99 outputs a signal indicating the type of the genuine coin only when the signals from the comparison means 94 to 97 all indicate the same genuine coin type, and determines a signal indicating the false coin otherwise. Output as signal 100. Each output of the dew condensation detection means 82 and the switching means 98 is also input to the determination means 99.
[0041]
The operation of the coin discriminating apparatus configured as described above will be described with reference to the block diagram of FIG. 8 using the flowcharts shown in FIGS. In steps S11 to S17, it is detected that the coin 23 has been inserted. Here, the insertion and detection of insertion of the coin 23 mean to include arrival of the coin 23 to the identification unit and detection of arrival. First, in order to detect coin insertion by the imaging unit 16, the insertion detection unit 79 sets the imaging unit 16 (S11). Specifically, in the imaging range 101 shown in the conceptual diagram of the imaging area in FIG. 11, the imaging unit 16 is set so that only a partial image of the first input detection area 102 is output, and each pixel of the image sensor 26 is set. Imaging is started after preparation for imaging such as initialization (reset) 61 is performed.
[0042]
Next, the illumination means 15 is caused to emit light for a short time at the timing when all the pixels 61 in the first input detection region 102 are in the light receiving state (S12). By emitting light only for a short time with respect to the passing speed of the coin 23, an image close to a stationary state can be obtained without stopping the coin 23, and the identification accuracy can be improved. Further, the LED 25 can be driven with a large current so that the illumination can be increased in illuminance or the life can be extended. Note that the illumination unit 15 causes both the second illumination unit 36 and the third illumination unit 37 to emit light. Unless otherwise specified, illuminating the illumination means 15 means that both the second illumination means 36 and the third illumination means 37 are caused to emit light.
[0043]
Then, the image data of the first input detection area 102 is input (S13). Next, when the image pickup means 16 detects the input, the same input is performed in the second input detection area 103, the next input in the third input detection area 104, and then the process returns to the first input detection area 102. The reason why the insertion detection is performed in the three insertion detection areas 102 to 104 is to reliably detect the insertion of the coin 23 even when the coin 23 and the passage 13 are dirty. Further, as shown in FIG. 11, the three insertion detection areas 102 to 104 are calculated backward from the time required for the pattern input illumination (S30) described later and the passing speed of the coin 23 with respect to the coin position 33 at the time of imaging. By disposing at the position, the coin 23 having an average passing speed can be imaged at substantially the center of the light receiving surface 27 on the illumination means 15 and the image sensor 26. In addition, the insertion detection regions 102 to 104 are separated from each other as much as possible so that they are not easily affected by the dirt on the coins 23 and the passages 13 at the same time, and the detection timing is not shifted by the outer diameter of the coins 23. Are placed in the vicinity of the center of the coin position 105 of the smallest coin to be identified and above and below it, and the insertion detection areas 102 and 104 arranged above and below these are the detection of insertion inserted in the vicinity of the center of the coin position 105 of the smallest coin. It is arranged slightly closer to the inlet than the area 103.
[0044]
Based on the image data of the input detection areas 102 to 104 input in this way, dew condensation is detected in the passage 13 including the optical transparent body 14, the lens 28, the light receiving surface 27 of the image sensor 26, and the like (S14). The determination of whether or not condensation is made is based on the fact that the input image is blurred due to defocusing on the surface of the coin 23 due to dew adhering to the optical transparent body 14 or the lens 28. If the variation of the image data in the three input detection areas 102 to 104, that is, the state in which all three variances are smaller than a certain value continues, it is determined that condensation has occurred. Dispersion can be easily obtained, and a shift in focus of the image pickup means 16 due to condensation can be detected by determining with the small dispersion. Furthermore, if even one of the input detection regions 102 to 104 has a large variance, the condensation determination is not performed, thereby preventing erroneous determination and improving the reliability of the condensation determination. If it is determined that condensation has occurred, the control processing of the image sensor 26 is not performed, so that malfunction due to abnormal image data can be prevented, condensation prevention means (not shown) can be operated, and condensation can be prevented from occurring in the vending machine. It is possible to communicate with the control unit and notify the operator of the vending machine by displaying. Condensation can be detected without using a dedicated sensor by using the imaging means 16, and an image for detecting condensation can be obtained in a short time by using the image sensor 26 capable of outputting a partial image. Can do.
[0045]
The input detection (S15) by the image pickup means 16 uses the fact that the illumination light irradiated from the illumination means 15 is reflected by the inserted coin 23 and the image data detected by the image pickup means 16 changes. If at least one of the average or variance of the image data in the three insertion detection areas 102 to 104 has a certain change from the standby state value, it is determined that the coin 23 has been inserted. By determining that the coin 23 has been inserted when at least one of the average and the variance changes, the insertion can be reliably detected even when no change occurs in either the average or the variance. In addition, the average and variance are calculated in a plurality of areas, and if there is a change in any one of the areas, it is determined that the coins are inserted. Can do.
[0046]
In the case where a sensor for detecting the insertion is provided separately, a space for providing this sensor is required, and if an optical sensor is used, interference with the illumination light of the illumination means 15 is prevented, and the sensor for detecting the input and the imaging are taken. The assembly of the means 16 causes variations in the distance between the two sensors, and an error occurs in the image detection timing. On the other hand, as in the present embodiment, it is possible to easily detect the input at a plurality of locations by using the image pickup means 16 to detect the input and also serve as the input detection sensor, reduce the space, and interfere with the illumination light. And accurate image detection that is not affected by assembly variations. Furthermore, by using the image sensor 26 capable of outputting a partial image, an input detection image can be obtained in a short time, and the input detection can be performed at high speed.
[0047]
Since it is not possible to detect the insertion of the coin 23 during the condensation determination or due to dirt in the passage 13 even if it is not condensation, the insertion is also detected by the thickness sensor 71 and the material sensor 75 other than the imaging means 16 (S16, S17). When the coin 23 inserted from the insertion slot 12 approaches the thickness / material combination sensor 17, the impedances of the coils 53 to 56 change, and the oscillation levels of the oscillation circuits 72 and 76 change accordingly. If there is a certain change from the oscillation level during standby, it is determined that a coin 23 has been inserted, and thickness and material detection (S20) is performed.
[0048]
In the thickness and material detection (S20), the thickness and material are detected from the change in the oscillation level of the oscillation circuits 72 and 76 corresponding to the impedance change of the coils 53 to 56 by the coin 23. As described above, the oscillation level of the thickness sensor 71 sensitive to the thickness of the coin 23 is connected to the thickness detection means 74 via the rectifier circuit 73 that converts the oscillation waveform from a sine wave to a signal indicating the oscillation level. The maximum amount of change in the oscillation level during passage is detected and output to the comparison means 94. Similarly, the material detection means 78 detects the maximum change amount of the oscillation level of the material sensor 75 sensitive to the material of the coin 23 when the coin passes and outputs it to the comparison means 95.
[0049]
When the imaging means 16 detects the insertion (S15), the unevenness and pattern detection (S21) and the thickness and material detection (S20) are performed in parallel. The unevenness and pattern detection (S21) is first performed from the adjustment step (S22) as shown in FIG . The image quality of image data to be detected changes depending on the material and dirt of the inserted coin 23, dirt on the passage 13, deterioration and temperature characteristics of the illumination means 15, and the brightness and contrast increase and decrease. On the other hand, every time the coin 23 passes, before performing the unevenness input and pattern input to be described later, the same image pickup means 16 and illumination means 15 as those used for these inputs are used in advance. Get a partial image. Based on the acquired image, the input conditions of the imaging unit 16, specifically the sensitivity, that is, the gain of the amplification circuit included in the image sensor 26, so that the image quality of the image data detected by the imaging unit 16 is as constant as possible. Adjust the offset level. As shown in FIG. 11, the input condition setting area 106 is set to a position and size that are completely included in the coin position 105 of the minimum coin with the assumed minimum speed and the coin position 107 with the assumed maximum speed. . In FIG. 11, the coin 23 is actually moved by the time required from the input detection to the input condition setting input, but the coin position 105 of the minimum coin with the minimum speed is It is the same as when detecting input. At the beginning of the adjustment step (S22), in order to speed up the readout time, the output image area of the imaging means 16 is set only to the input condition setting area 106, and imaging is started after preparation for imaging (S31). Subsequently, the illumination unit 15 emits light for a short time (S32), and the image data of the input condition setting area 106 is input (S33). In the present embodiment, the gain or offset level of the amplification circuit included in the image sensor 26 is changed according to the number of very bright pixels and the number of very dark pixels in the image data, and the setting of the imaging means 16 (S34). Therefore, the adjustment can be easily made only by comparing the image data and counting the number of pixels. However, the present invention is not limited to this, and it may be based on the average value or variation of the image data, or the A / D conversion means 80. The reference voltage is changed by the same means as changing the gain and offset level of the amplifier circuit included in the image sensor 26. Thus, by acquiring the image of the coin 23 in advance by the imaging unit 16, the input conditions of the imaging unit 16 and the like can be adjusted based on the acquired image, and image data with a constant image quality can be obtained. Further, by storing the adjustment amount and correcting the adjustment amount every time the coin 23 is inserted, it is possible to reduce the influence of changes over time such as component deterioration and maintain the identification performance. By inputting at an assumed position of the coin 23, an image of the coin 23 can be reliably obtained and appropriate adjustment can be performed. Further, by using the image sensor 26 capable of outputting a partial image, an adjustment image can be obtained in a short time and can be adjusted at high speed.
[0050]
Further, the same effect can be obtained by changing the illumination conditions such as the light emission intensity and the light emission time of the illumination means 15, but in the adjustment method of the present embodiment, when adjusting the illumination conditions, specifically, the illumination means 15 Compared with the case where the drive power supply (for example, the drive current of the LED 25) and the light emission time are changed, the adjustment can be performed without adding a power supply circuit or complicated control for changing the reading timing of the image sensor 26.
[0051]
A technique for providing a reflected light sensor different from the imaging unit 16 and performing adjustment according to the amount of light received by the reflected light sensor is disclosed in, for example, Japanese Patent Application Laid-Open No. 2000-33111. In this case, an image for adjustment is acquired by using the same image pickup means 16 used for unevenness input and pattern input, so that no adjustment dedicated sensor is provided and separate sensors are used for adjustment and identification. It is possible to adjust without being affected by the detection error.
[0052]
The subsequent process is an unevenness input process (S23). A fake coin (hereinafter referred to as a copy affixed fake coin) produced by pasting a copy of a true coin or a paper printed with a pattern of a true coin on a metal is sufficient with the true coin in the pattern verification (S51) described later. It is difficult to eliminate them without finding any significant difference. Therefore, in order to detect the degree of unevenness and eliminate the copy-pasted fake coins, as described above, the illumination means 15 is changed to the second illumination means 36 on the outlet 18 side and the third illumination means 37 on the insertion port 12 side. The second illumination means 36 and the third illumination means 37 can emit light independently. Then, as shown in FIG. 12, the first image data can be obtained when illuminated only by the second illumination means 36 that is generated if there is unevenness (see the cross-sectional view in the vicinity of the coin in FIG. 12A). The shadow position (see the schematic diagram of the image data in FIG. 12B) and the second when illuminated only by the third illumination means 37 (see the cross-sectional view in the vicinity of the coin in FIG. 12C) The difference between the positions of the shadows that can be formed in the image data (see the schematic diagram of the image data in FIG. 12D), that is, the difference when illuminated from different directions is used. As shown by broken lines in FIG. 11, the three uneven detection lines 108 to 110 are completely included in the assumed minimum coin position 105 of the minimum coin and the assumed minimum coin position 107 of the maximum speed. Since the position and length are set, an image of the coin 23 can be obtained with certainty, and appropriate unevenness detection can be performed. In FIG. 11, the coin 23 has actually moved for the time required from the input detection to the input of the unevenness detection, but the coin position 105 of the minimum coin with the minimum speed is detected for the sake of simplicity. Same as time. At the beginning of the concavo-convex input step (S23), the output image area of the imaging means 16 is sequentially set to these concavo-convex detection lines 108 to 110, and imaging preparation is started (S35). Subsequently, the second illuminating means 36 or the third illuminating means 37 are alternately switched with respect to one unevenness detection line to emit light for a short time (S36), and image data is input (S37). It repeats until the 1st and 2nd image data of the unevenness | corrugation detection lines 108-110 are input (S38). By performing the interval between the first image input and the second image input in a short time of about 500 microseconds, the coin movement between them is an average passing speed equivalent to one pixel, which will be described later. In addition, when the difference between the first image data and the second image data is calculated, the influence of coin movement can be suppressed by correcting the shift of one pixel. In the present embodiment, the second illumination means 36 and the third illumination means 37 are all configured by the LEDs 25 having the same wavelength characteristics. Therefore, the unevenness can be detected in the monochrome image. Further, the second illumination means 36 and the third illumination means 37 for detecting irregularities are configured by dividing the illumination means 15 into two groups as described with reference to FIG. That is, since the illumination means 15 for pattern detection is also used and the imaging means 16 is also used as the same, the cost and space can be reduced, and adjustment does not require a dedicated adjustment process. By causing the second illuminating unit 36 and the third illuminating unit 37 to emit light at different timings, interference of illumination light by the second and third illuminating units can be prevented, and an image of each reflected light can be obtained reliably. . By using the image sensor 26 capable of outputting a partial image, an image of reflected light can be obtained in a short time and irregularities can be input at high speed.
[0053]
In addition, although the technique which determines the unevenness | corrugation of a coin by irradiating with a different wavelength characteristic from a different direction is disclosed by Unexamined-Japanese-Patent No. 2002-183791, for example, since it irradiates with the same wavelength characteristic in this Embodiment. Unevenness can be detected without using illumination means and color image sensors or filters having different emission wavelengths. In addition, the degree of unevenness can be detected without being affected by variations in sensor sensitivity and filter transmission characteristics that vary depending on the wavelength and ambient temperature.
[0054]
Further, a technique for determining the pattern of a coin by irradiating separately from different directions and performing binarization based on each reflected light intensity is disclosed in, for example, Japanese Patent Laid-Open No. 7-210720. In the embodiment, as will be described later in the unevenness calculation (S44), since the detection is based on the difference between the reflected light intensities, the unevenness degree can be detected with multiple values and can be accurately determined.
[0055]
The next step is a pattern input step (S24). The output image area of the imaging means 16 is set to a position and size that completely include the maximum coin with the assumed minimum speed and the maximum coin with the assumed maximum speed, and imaging is started after preparation for imaging is performed. (S39). Subsequently, the illumination unit 15 emits light for a short time (S40), and the third image data is input (S41). In the pattern input, the coin input position in the image data differs due to variations in the assembly dimensions generated for each product and the characteristics of the lens 28, etc. Therefore, it is necessary to set a larger pattern input area accordingly. In contrast, in this embodiment, the position of the bottom surface 22 in the image data, that is, the lower end position of the coin 23 and the imaging magnification are variably stored, and the upper end position in the image data of the largest coin is calculated. By reducing the pattern input area, the input time of image data, the capacity of the storage circuit for storing data, and the processing time of data can be reduced. Further, the stored lower end position of the coin 23 and the imaging magnification can be reliably corrected by performing correction every time a genuine coin is inserted, and the performance deteriorates even with a change with time. And the initial discrimination performance can be maintained. The lower end position of the coin 23 and the imaging magnification are stored and corrected, the center position of the minimum coin is calculated from these, and the insertion detection areas 102 to 104, the input condition setting area 106, based on any one or more of the three, By setting the unevenness detection lines 108 to 110, the pattern input region, and detection lines 111 and 112 for detecting the center and outer diameter described later, the position of the coin 23 in the imaging means can be accurately captured, It is possible to accurately detect each of the assembling error of the image pickup means 16 and the dimensional / characteristic change with time and regardless of the size of the inserted coin 23, that is, the denomination. For example, Japanese Patent Application Laid-Open No. 10-105765 discloses a technique for storing the position of the coin 23 in the adjustment mode. However, the coin is inserted by variably storing it as in the present embodiment. It is possible to correct each time, and the identification performance can be maintained following the change over time.
[0056]
In the subsequent unevenness detection step (S25), the unevenness degree is calculated from the first and second image data input in the unevenness input step (S23), but before that, the center and the outer diameter are calculated from the third image data. And determining which denomination is close to the coin 23 inserted based on the detected outer diameter.
[0057]
In the center / outer diameter detection (S42), in order to detect the position of the image corresponding to the coin 23 in the third image data, the outer diameter is obtained together with the center as the reference point. Particularly when the identification target is a circular coin 23, the position of the coin 23 in the image data can be easily detected by obtaining the center and the outer diameter. Therefore, a plurality of candidate points corresponding to the outer periphery are selected, the distance between each candidate point and the temporary center is verified, the center is determined from the candidate points determined to correspond to the outer periphery, and the position of the coin 23 is determined. Detected. As shown in FIG. 13, for the third image data, first, a plurality of detection lines 111 parallel to the passing direction and a plurality of vertical detection lines 112 are set. In the present embodiment, each of the six detection lines 111 and 112 is always based on the lower end position of the stored coin 23 and the outer circumference of the minimum coin with the assumed minimum speed and the minimum coin with the assumed maximum speed. The layout and length are such that they intersect. By using the stored lower end position of the coin 23 and performing correction each time a correct coin is inserted, it is possible to reliably detect the position and to perform performance against changes over time. Deterioration can be prevented. In FIG. 13, for the sake of simplicity, four broken lines are shown. Then, smoothing and edge detection are performed on the image data on each of the detection lines 111 and 112. Specifically, for 9 pixel data including 8 pixels around the pixel of interest, the array data in (Table 1) and the absolute value of the sum of 9 elements integrated for each array element, and the array data in (Table 2) The sum with the absolute value of the sum of the nine elements integrated for each array element is obtained and used as data after smoothing and edge detection of the pixel of interest.
[0058]
[Table 1]

[0059]
[Table 2]

[0060]
The point after the smoothing / edge detection is divided into two at the center of the detection lines 111 and 112, and the point having the maximum value in the first half and the second half is set as the outer peripheral candidate point 113. At this time, by adding a correction value proportional to the distance from the center of the line and obtaining the maximum value up to the third, the outside of the image where the outer periphery of the coin 23 is likely to be copied, that is, from the center of the line It is possible to preferentially select a distant point, and even when the illuminance of the illumination unit 15 is uneven and the outer periphery of the coin 23 is darkly captured, the position of the coin can be detected with high accuracy. A part of the outer periphery candidate points 113 is indicated by white and black dots in FIG. For each line, a total of nine midpoints of two points selected from the first half candidate points and the second half candidate points are obtained. The point having the highest frequency from the frequency distribution of 54 middle points for 6 lines is set as the temporary center coordinate, and this is obtained in the direction parallel to and perpendicular to the passing direction to be the temporary center 114. The distance between the temporary center 114 and each outer peripheral candidate point 113 (72 points in total) is calculated, and the distance with the highest frequency is obtained from the frequency distribution of these distances. Then, out of the peripheral candidate points 113 within a certain allowable range with respect to the obtained distance, a set of two points belonging to only one point each in the first half and the second half of the same detection line is selected, and a black dot is shown in FIG. The outer peripheral point 115 is shown. For example, the outer periphery candidate point 116 is not set as the outer periphery point 115 because the distance from the temporary center 114 is within the allowable range, but there is no outer periphery candidate point 113 within the allowable range on the same detection line 112. Finally, an average coordinate parallel to the passing direction of the outer peripheral point 115 on the detection line 111 parallel to the passing direction and an average coordinate perpendicular to the passing direction of the outer peripheral point 115 on the detection line 112 perpendicular to the passing direction are obtained. The center. Further, the average distance between the center and the outer peripheral point 115 is obtained and doubled to obtain the outer diameter. In the present embodiment, a plurality of candidate points corresponding to the outer periphery are selected, the distance between each candidate point and the temporary center is verified, and the center is obtained from the candidate points determined to correspond to the outer periphery, thereby ensuring In addition, the position of the coin 23 can be detected. By setting a plurality of detection lines 111 and 112 and performing smoothing and edge detection to select a plurality of outer peripheral candidate points 113 on the detection lines 111 and 112 while correcting so that priority is given to the outside of the image. Thus, reliable center / outer diameter detection that is less susceptible to the effects of dirt on the coins 23 and the passages 13 and unevenness of the illumination means 15 becomes possible. Similar effects can be obtained by edge enhancement instead of edge detection. By performing smoothing and edge detection or enhancement simultaneously with the array data of (Table 1) and (Table 2), the center and outer diameter can be detected at high speed. In addition, when the identification target is other than a circle, for example, for a rectangular bill, etc., the long side and the short side dimension may be obtained together with the center and the outer periphery. It can also be detected as a reference point to detect the position of the identification target.
[0061]
In addition, although the technique which detects an outer periphery point with a some detection line is disclosed by Unexamined-Japanese-Patent No. 9-17821 or Unexamined-Japanese-Patent No. 10-63852, for example, this embodiment selects a plurality of outer periphery candidate points, Since the outer peripheral point is obtained by performing verification based on the distance from the temporary center, the outer diameter point is not erroneously selected even when the bottom surface 22 is reflected or there is noise as shown in FIG. The center and outer diameter can be detected.
[0062]
In the denomination temporary determination (S43), from the outer diameter output from the center / outer diameter detection means 85, it is temporarily determined which denomination the inserted coin 23 is close to. Two denominations close to the detected outer diameter are selected and output from a predetermined outer diameter reference value for each denomination and the stored imaging magnification. However, if the corresponding denomination is less than or equal to 1 denomination within a predetermined allowable range, that denomination is selected. Based on the outer diameter, the denomination can be easily provisionally determined. By using two denominations instead of focusing on one denomination, it is possible to misdetermine the denomination due to the influence of the passing position such as whether the coin 23 passes near the imaging means 16 in the passage 13 or far away. It is possible to improve the identification accuracy. For example, in Japanese one hundred yen coins (22.6 mm) and ten yen coins (23.5 mm), the outer diameter is about 4%, and particularly when a wide angle lens is used, the influence of the passing position is large. Since the outer diameter on the image data changes by ± 2% due to a change in the passing position of 5 mm, there is a risk of erroneous determination. Further, if it is tentatively determined to be more than 3 denominations, it takes a lot of time for the subsequent data processing, especially the rotation angle detection (S50) in the pattern matching step (S26) described later, so that the data processing time is constant for a maximum of 2 denominations. Although the upper limit time is not exceeded, three or more denominations may be used.
[0063]
A technique for determining a denomination as one denomination and performing a process according to the denomination is disclosed in, for example, Japanese Patent Laid-Open No. 10-63852. In this embodiment, a provisional determination is made on a plurality of denominations. By doing so, it is possible to prevent an erroneous determination of the denomination due to the influence of the passing position of the coin 23 or the like.
[0064]
In the subsequent concavo-convex calculation (S44), the concavo-convex degree of the coin 23 is detected by the concavo-convex detection means 83 from the first and second image data input in the concavo-convex input step (S23). For each unevenness detection line 108 to 110, the absolute value of the difference between the first image data and the second image data for each pixel is added over all the pixels on the unevenness detection line. When calculating the difference for each pixel, in order to cancel the influence of the movement of the coin 23 described above, the pixel of the first image data to be input first is one pixel behind the second image data (exit 18 side). ) To obtain the difference from the pixel data. Thus, the calculation is performed for the three unevenness detection lines 108 to 110, and the maximum value of the three is output. By setting a plurality of concavo-convex detection lines 108 to 110, the degree of concavo-convex can be reliably detected even with the coin 23 having a small concavo-convex pattern that is not easily affected by the dirt of the coin 23 and the passage 13.
[0065]
Further, a pattern matching step (S26) is performed. In the pattern matching step (S26), edge detection (S45) of the image data of the multi-value digital value is performed, and then the coordinate conversion from the orthogonal coordinates to the polar coordinates (S47) is performed while correcting the imaging magnification (S46), and binarization is performed. After (S49), the rotation angle of the image data of the inserted coin 23 with respect to the reference image is detected (S50), and collation with the reference image is performed in consideration of the detected rotation angle (S51). These steps are performed on the two front and back surfaces of the two denominations that were provisionally determined in the denomination denomination (S43). For example, a single-sided pattern is common, such as euro coins, and the other side is in 15 countries. If they are different, processing for a total of 16 planes may be performed.
[0066]
Each operation | movement of a pattern collation process (S26) is demonstrated in detail from edge detection (S45). In the edge detection means 84, edge detection that also serves as smoothing is performed on each pixel data of the third image data in the same manner as the center / outer diameter detection (S42), and the image data after the edge detection is used as the fourth image data. Output as data. Specifically, for 9 pixel data including 8 pixels around the pixel of interest, the array data in (Table 1) and the absolute value of the sum of 9 elements integrated for each array element, and the array data in (Table 2) The sum with the absolute value of the sum of the nine elements integrated for each array element is obtained and used as data after edge detection of the pixel of interest. Note that it is sufficient that the edge detection (S45) is performed for pixels located within the outer diameter range of the maximum outer diameter coin to be identified with reference to the center detected by the center / outer diameter detection (S42). By performing edge detection (S45) on the image data, it can be determined not by the brightness (brightness) of the reflected light but by the brightness difference with the adjacent portion, so that the overall brightness of the image data can be determined. It is difficult to be affected by various changes and local unevenness, and accurate identification becomes possible. Similar effects can be obtained by performing edge enhancement instead of edge detection. Here, the overall change in brightness of the image data and local unevenness are caused by the material and color of the coin 23, the entire passage 13 or the local contamination of the coin 23, and the like. These effects can be reduced.
[0067]
Next, the coordinate transformation (S47) will be described before the magnification correction (S46). The coordinate conversion means 88 converts the fourth image data from orthogonal coordinates to polar coordinates, and outputs the image data after the coordinate conversion as fifth image data. In the present embodiment, an optical design is performed in which a Japanese coin is an identification target and a 500-yen coin having the maximum outer diameter becomes N × N pixels (for example, 100 × 100 pixels). In addition, the imaging magnification is assumed to be 0.9 to 1.1 depending on the passing position of the coin, the assembly dimensions, and the variations in the component characteristics such as the lens 28. That is, a 500-yen coin may be imaged from a minimum of 0.9 N × 0.9 N to a maximum of 1.1 N × 1.1 N pixels (for example, 90 × 90 to 110 × 110 pixels) for each product or coin insertion. However, for simplification, the coordinate conversion (S47) and magnification correction (S46) will be described with a maximum of 7 × 7 pixels. FIG. 14A schematically shows the fourth image data. The X coordinate 122 (direction parallel to the passing direction of the coin 23) and the Y coordinate of the center 121 detected by the center / outer diameter detection (S42). 123 (direction perpendicular to the passing direction of the coin 23) is (0, 0), and 49 pixels from orthogonal coordinates (-3, -3) to (3, 3) are shown. FIG. 14B schematically shows the fifth image data after the coordinate conversion. The fifth image data is divided into four at non-uniform intervals in the radial direction of the polar coordinates (R coordinate 124), and the circumferential direction (S The polar coordinates (0,0) to (3,7) are shown by dividing the coordinates 125) into 8 at equal intervals to 0-7. This coordinate conversion is performed using the conversion tables shown in (Table 3) and (Table 4).
[0068]
[Table 3]

[0069]
[Table 4]

[0070]
(Table 3) represents the S coordinate of the polar coordinate corresponding to the X coordinate and the Y coordinate of the orthogonal coordinate, and (Table 4) represents the R coordinate of the polar coordinate corresponding to the X coordinate and the Y coordinate of the orthogonal coordinate. For example, a pixel with orthogonal coordinates (0, 0) is converted into polar coordinates (0, 0), and a pixel with orthogonal coordinates (-1, -3) is converted into polar coordinates (3, 7). In this way, all the pixels of the orthogonal coordinates are converted, and when a plurality of orthogonal coordinates correspond to the same polar coordinates, the average value of these pixels is used, and the polar coordinates having no corresponding orthogonal coordinates are adjacent polar coordinate data. Set based on adjacent data, such as copying. By converting the coordinates to polar coordinates, rotation angle detection (S50) and collation (S51), which will be described later, can be facilitated, the amount of data is reduced, and the processing speed can be increased. In the radial direction table (Table 4) for polar coordinate conversion, there is a coordinate where the R coordinate of the radial coordinate is 4 which is larger than the maximum value of 3, but this is outside the outer diameter range of the coin 23 and will be used thereafter. It is necessary to perform conversion while determining whether or not the R coordinate is 3 or less by preparing up to 4 R coordinates of the array for storing the image data after coordinate conversion, indicating that the pixel is not to be converted Therefore, all pixels can be handled in common and converted at high speed. Further, by performing coordinate conversion (S47) after edge detection (S45), the edge can be detected faithfully to the pattern of the coin 23 and the identification accuracy can be improved. This is because if the edge detection (S45) is performed after the coordinate conversion (S47) because the distance to the adjacent surrounding pixels (coordinates) is changed by the polar coordinate conversion, the adjacent neighbors used for edge detection or enhancement of the pixel of interest This is because the distance to the pixel differs depending on the coordinates of the pixel of interest and the coordinates of the adjacent pixels and is not constant.
[0071]
The magnification correction (S46) corrects the imaging magnification due to the passage position of the coin 23 in the passage 13, the assembly error of the imaging means 16, and the like. This correction is based on the values in the radial direction table (Table 4) for polar coordinate conversion based on the outer diameter detected in the center / outer diameter detection (S42) and the money type temporarily determined in the money type temporary determination (S43). Do by updating. The radial direction table is obtained from a distance table indicating the distance between each pixel and the center as shown in (Table 5) and an upper limit distance table indicating the upper limit of the distance for each radial coordinate as shown in (Table 6).
[0072]
[Table 5]

[0073]
[Table 6]

[0074]
For example, if the X coordinate is x, the Y coordinate is y, and the distance table value is represented by R ′ (x, y), R ′ (0, 0) is 0.00 and less than 1.75. R (0,0) is set to 0, and R ′ (3,1) is 3.16, which is 3.03 or more and less than 3.50, so R (3,1) is set to 3. Note that the distance table of the portion of X coordinate <0 and Y coordinate <0 is omitted by using objectivity to reduce data storage capacity and processing time, and when x <0 and y ≧ 0, R (x, y) = R (−x, y), R (x, y) = R (x, −y) when x ≧ 0 and y <0, and R (x, y) when x <0 and y <0. y) = R (−x, −y) is used to obtain the radial coordinate R.
[0075]
In the magnification correction (S46), for example, one of the denominations temporarily determined in the denomination temporary determination (S43) is 500 yen, and the outer diameter (eg, 90) detected in the center / outer diameter detection (S42) is 500 yen. In the case of 0.9 times the standard outer diameter (for example, 100), since the imaging magnification is 0.9 times, the upper limit distance value in the upper limit distance table is 0.9 times the standard as shown in (Table 7). Update to the value of.
[0076]
[Table 7]

[0077]
Then, when the value of the radial direction table for polar coordinate conversion is updated from the upper limit distance table and the distance table (Table 5), it becomes (Table 8).
[0078]
[Table 8]

[0079]
By performing coordinate conversion based on this conversion table, coordinate conversion can be realized simultaneously while enlarging the image data by 1.11 times in the radial direction, and the imaging magnification is easily corrected. By performing the magnification correction, accurate identification can be performed without being affected by variations in the passing position of the coin 23, assembly dimensions, and component characteristics. In addition, it is possible to suppress the variation in the identification performance for each product by storing the reference image in common without storing different reference images for each product, and to increase the allowable range of assembly and component characteristics.
[0080]
In the present embodiment, the value of the upper limit distance in the upper limit distance table is set to an unequal interval, with the interval being wider as the radius is smaller and the interval being larger as the radius is larger. As shown in (Table 9), they may be set at equal intervals.
[0081]
[Table 9]

[0082]
However, there is a large difference in the area of each polar coordinate, and the number of orthogonal coordinate pixels that are converted to the same polar coordinate at the time of coordinate conversion becomes non-uniform, resulting in a bias in the weight of each polar coordinate at the time of collation. In particular, if the upper limit distance is set so that each polar coordinate has the same area, the weight of each polar coordinate becomes equal, and the identification performance can be improved by setting the radial direction coordinate at unequal intervals.
[0083]
Japanese Patent Laid-Open No. 9-245213 discloses a technique for acquiring an image of a reference object, obtaining a magnification error, correcting the image, and cutting out the image. However, the reference object is a fixed denomination. There is no problem when only the denomination is input in the adjustment mode, but application is difficult when a plurality of denominations are input. On the other hand, this embodiment can also be applied to a case where a plurality of denominations are input by provisionally determining a plurality of denominations.
[0084]
Subsequently, threshold setting (S48) for binarization (S49) is performed. The binarizing means 90 converts each pixel (coordinate) of the fifth image data after coordinate conversion, which is a multi-valued digital value, into a binary digital value of 1 or 0 depending on whether or not it is greater than or equal to a threshold value. The binarized image data is output as sixth image data. By performing binarization (S49), it is possible to reduce the influence of noise in the image data, reduce the amount of image data, and speed up the processing. In addition, by performing edge detection (S45) or edge enhancement and coordinate transformation (S47) before binarization (S49), accurate edge detection or enhancement and coordinate transformation using multi-valued data can be performed and identified. Accuracy can be improved.
[0085]
The threshold setting unit 89 sets a threshold for binarization (S49), and outputs a threshold set based on the fifth image data and the denomination temporarily determined by the denomination temporary determination unit 86. . Specifically, in the sixth image data after binarization, the ratio of the number of pixels having a value of 1 to the total number of pixels is a ratio determined for each denomination and surface (front surface or back surface). In other words, the threshold is set so that the frequency distribution after binarization becomes constant. As a result, the image quality of the sixth image data after binarization is made uniform, and a reference image used for pattern matching (S51), which will be described later, is also created at a similar ratio, whereby the sixth image and the reference image are generated. Homogenization is possible and identification performance can be improved. In this embodiment, this ratio is set to 50% for all denominations and faces. However, it is possible to optimize the binarization according to each denomination and face pattern by changing it for each denomination and face. Based on the experiment, the pattern was clearly read at a rate of 30 to 70%, and high discrimination performance was achieved. For example, a threshold value may be obtained by creating a frequency distribution (histogram) of data values and the number of pixels from the fifth image data and obtaining a value at which the cumulative frequency from the larger data is 50%. In addition, the fifth image data used for setting the threshold value should naturally be limited to the inside of the coin portion, that is, the outer diameter, and the radius of the denomination whose radial coordinate is predetermined by performing magnification correction and polar coordinate conversion. This is easily possible if only the following parts are used. Further, in the present embodiment, the outer periphery / hole edge that differs for each coin 23 depending on the state of the coin 23 and the issue year, the inside of the hole that is the background instead of the coin 23, and the same coin 23 depending on the rotation angle and position. Also, different latent image / thin line patterns are excluded from use for threshold setting. FIG. 15 is a schematic diagram in which a reference image (to be described later) of a new 500-yen coin is overwritten with an excluded portion in gray and inversely converted into orthogonal coordinates. The outer peripheral edge 131, the latent image pattern portion 132, and the fine line pattern portion 133 are shown in FIG. Indicates. Wear and burrs are likely to occur on the outer periphery and edge of the hole, the background can be seen inside the hole, and the latent image and fine line pattern part have moire depending on the angle between the pixel arrangement direction of the image sensor 26 and the coin pattern. Appearance changes remarkably, and when used for threshold setting, it is affected by these partial images and hinders accurate binarization of the whole. For the same reason, the year part which is different for each coin may be excluded. Which part of the fifth image data is used for threshold setting can be determined by storing a binary digital value table in polar coordinate format for each denomination / surface in advance. Stable binarization suitable for the pattern can be performed. Further, when the portion used for setting the threshold value is not rotationally symmetric with respect to the center of the coin, the threshold value changes for each rotation angle of the coin. Therefore, the threshold value is set each time image data is rotated in the rotation angle detection (S50) described later. It requires a great deal of processing. In the present embodiment, the portion used for threshold setting is set so as not to include the excluded portion and to be rotationally symmetrical, that is, concentric with respect to the center of the coin. Therefore, every time image data is rotated in the rotation angle detection (S50). There is no need to re-determine the threshold value, and the table for storing the portion used for setting the threshold value may be one-dimensional data (hereinafter referred to as threshold setting area data) for the coordinates in the radial direction. In this way, the threshold value is set so that the ratio of the number of pixels having a value of 1 to the total number of pixels in the sixth image data becomes a predetermined ratio, that is, the frequency distribution after binarization is constant. Thus, stable identification can be performed without being influenced by the material of the coin 23, the state of the edge or surface, the rotation angle, and the difference in illuminance due to variations in the illumination means 15 and assembly dimensions.
[0086]
In addition, although the technique which sets the threshold value of binarization in a specific threshold value calculation area | region is disclosed by Unexamined-Japanese-Patent No. 2002-109596, for example, the specific part (coin 23 state in the coin 23) which interferes in binarization In addition, there is no mention of a technique for removing outer peripheries and hole edges that differ for each coin 23 depending on the issue year, and a latent image or thin line pattern that differs for each coin 23 depending on the rotation angle or position. In the present embodiment, stable threshold setting and binarization are possible by setting threshold values in areas other than these areas.
[0087]
In the rotation angle detection (S50), based on the sixth image data by the rotation angle detection means 91 and the reference image stored for each denomination / surface in the storage means 92, the coins 23 inserted into the reference image, that is, the first The rotation angle of the image data 6 is detected and output. Conversely, the rotation angle of the reference image with respect to the sixth image data may be detected. The reference image data is stored in advance as a binary digital value table in the polar coordinate format, like the sixth image data. The sixth image data is compared with the reference image data while rotating, and the most coincident angle is detected. Since the sixth image data is represented by polar coordinates, it can be easily rotated by shifting in the circumferential direction, that is, by changing the circumferential coordinate. The sixth image data is B (r, s) (r is the radial direction of polar coordinates, s is the circumferential direction), the reference image data is R (r, s), and the remainder obtained by dividing a by b is mod (a, b) When the number of divisions in the circumferential direction in polar coordinates is represented by Ns, B (r, mod (s + Δs, Ns)) and R (r, s) are compared by changing Δs by 2 from 0 to Ns−1. I will do it. The degree of matching is determined by the number of pixels (coordinates) at which the values of the sixth image data and the reference image data match. Since the sixth image data and the reference image data are both binary digital values, they can be easily determined by taking the exclusive OR of these values and the number of pixels of which the result is zero. The reason why Δs is changed by two is to shorten the detection time, and Δs having the highest degree of coincidence is obtained by changing by two. Then, the rotation angle adjacent to Δs, that is, the degree of coincidence between mod (Δs−1, Ns) and mod (Δs + 1, Ns) is also obtained, and the rotation angle with the highest degree of coincidence is selected from the three rotation angles. Thus, the rotation angle can be detected with high accuracy even if the rotation angle is thinned out. Not only the method of changing by two, but the accuracy is maintained by calculating the degree of coincidence by thinning the rotation angle, such as increasing the amount of change if the degree of coincidence is low and decreasing the amount of change if the degree of coincidence is high. The speed of rotation angle detection can be increased. Further, in the present embodiment, the degree of coincidence is calculated for the same reason as when the threshold value is set for the outer periphery / edge of the coin, the inside of the hole, and the latent image / thin line pattern portion excluded from use in the threshold setting (S48) ( This is also excluded in the rotation angle detection and verification). Specifically, data indicating pixels (regions) used for rotation angle detection (S50) and collation (S51) (hereinafter referred to as collation region data) is expressed in a polar coordinate format similar to the reference image data and the sixth image data. The binary digital values are stored in advance as a table. Then, an exclusive logical sum of the sixth image data and the reference image data and a logical product of the collation area data are calculated, and it can be determined that the degree of matching is higher as the number of mismatched pixels whose result is 1 is smaller. In order to express the degree of coincidence, not relative, in this embodiment, the number of pixels having a value of 1 in the collation area data is used as the number of collation pixels, and the number of mismatch pixels is subtracted from the number of collation pixels. Is divided by the number of matching pixels to obtain the degree of matching. Thereby, even if the number of matching pixels is different, the degree of matching can be easily compared.
[0088]
In the pattern matching (S51), the sixth image data and the reference data at the rotation angle detected by the rotation angle detecting means 91 are checked. First, the sixth image data is shifted in the circumferential direction by the rotation angle detected by the rotation angle detecting means 91, that is, the rotation angle is corrected by changing the circumferential coordinate to obtain seventh image data. Then, an exclusive OR of the seventh image data and the reference image data is obtained, and a logical product with the collation area data used in the rotation angle collation is obtained as collation result data. The value of the pixel (coordinate) where the seventh image data and the reference image data do not match is exclusive OR, and if it is within the collation area, the logical product with the collation area data is 1, so the result is 1 The number of pixels represents the number of mismatched pixels. In addition, since the number of pixels having a value of 1 in the matching area data represents the number of matching pixels, the quotient obtained by dividing the difference between the number of matching pixels and the number of mismatched pixels by the number of matching pixels is the seventh image data and the reference image. Indicates the degree of match.
[0089]
In this embodiment, instead of obtaining the degree of coincidence of the entire coin, the coin area is divided into a plurality of areas, and the number of areas in which the degree of coincidence for each area is equal to or greater than a certain value, that is, the number of coincidence areas is obtained. . By comparing the degree of coincidence with a constant value, the number of coincidence areas can be easily obtained. Specifically, referring to FIG. 16, the example in which the number of pixels in the radial direction is 32 and the number of pixels in the circumferential direction is 16 will be described. This is a schematic diagram illustrating the region division in FIG. 16B for the image data illustrated in FIG. As shown in the figure, the radial direction 141 in polar coordinates is divided by a divisor of the number of pixels in the radial direction (the division number in this example is 8), and the circumferential direction 142 is similarly divided by a divisor of the number of pixels in the circumferential direction (this The number of divisions in the example is 4), and a plurality of (in this example, 32) regions 143 are obtained. By being based on polar coordinates, it is possible to easily divide the region, and when divided at equal intervals, the weight of each region 143 becomes equal. Then, for each region, the degree of matching is obtained with the pixels belonging to that region. Further, the pattern matching means 93 outputs this value as the matching area ratio, which is the ratio between the number of areas whose degree of matching for each area is equal to or greater than a predetermined lower limit value and the total number of areas. Note that it is difficult to accurately determine the degree of coincidence in the region 143 in which the amount of information used for calculating the degree of coincidence of each region is insufficient (the collation region is narrow, that is, the number of collation pixels is small). The region 143 in which the number of matching pixels is less than a certain value is excluded, and the matching region is accurately determined. In this way, instead of obtaining the degree of coincidence of the pattern of the entire coin, the coin area is divided into a plurality of areas 143 and collation is performed based on the degree of coincidence of each area. Even if it exists locally, it is not easily affected and accurate identification is possible. In addition, by simply changing the lower limit value of the degree of matching for each area determined to be a matching area and the allowable range of the matching area ratio in the comparison unit 97, it is possible to easily change the priority with respect to dirt resistance and fake coin elimination performance. . For example, in order to be able to discriminate even if a 1/3 portion of a coin cannot be accurately read due to dirt or the like, the allowable range of the matching area ratio may be set to 2/3 or more.
[0090]
Furthermore, the number of coincidence areas is the number of areas whose degree of coincidence for each area is equal to or greater than a predetermined lower limit value. This lower limit value is determined for each denomination / surface, and if it is variable and not fixed, 23 and the influence of dirt on the passage 13 can be reduced and accurate identification can be performed. This is because if the coins 23 and the passage 13 are dirty, the degree of coincidence decreases, the number of coincidence areas decreases, and the possibility that the coins are not determined as genuine coins increases. This is because the matching degree of the entire matching area is almost uniformly decreased at any rotation angle in the rotation angle detection. On the other hand, for example, the matching degree at the rotation angle at which the matching degree of the entire matching area is the smallest is obtained. By using the difference between the maximum matching degree and the minimum matching degree, or by adding a certain value to the minimum matching degree, the improvement can be made. In this example, when the contamination is so severe that accurate identification is difficult, the maximum matching degree becomes a very low value and there is a risk of erroneous identification. However, if the absolute value of the maximum matching degree is low, a lower limit value is determined not to be determined as a true coin, or the lower limit value calculated by adding a certain value to the minimum matching degree is less than a predetermined constant value. For example, by setting the lower limit value to this constant value, it is possible to avoid erroneous identification.
[0091]
The pattern matching (S51) is first performed on the surface of the first denomination that is temporarily determined to be approximate in the denomination temporary determination (S43) based on the reference data and the matching area data. Next, since the threshold setting area data is different for each surface, the threshold setting (S48) to the pattern matching (S51) are similarly repeated for the back surface. Furthermore, since the magnification correction differs for each denomination, the magnification correction (S46) to S52 are performed for the second denomination that is provisionally determined to be approximate in the denomination temporary determination (S43). Therefore, the pattern matching means 93 outputs the matching area ratio of 2 denominations × 2 planes.
[0092]
In the comparison (S61), the outputs of the thickness detection means 74, material detection means 78, unevenness detection means 83, and pattern matching means 93 input to each of the comparison means 94 to 97 are compared with the reference stored in the storage means 92. A comparison is made. If the values match within the allowable range, a signal indicating the type of the genuine coin is output, and if the reference value does not match any type, a signal indicating that it is a false coin is output. The determination (S62) outputs a signal indicating the type of the true coin only when the signals from the comparison means 94 to 97 input to the determination unit 99 all indicate the same type of the correct coin, and in other cases Outputs a signal indicating a false coin as the determination signal 100.
[0093]
A comparison (S71) between the case where the thickness sensor detects the insertion (YES in S16) and the case where the material sensor detects the insertion (YES in S17), that is, the case where dew condensation is being detected (S71) is shown in FIG. Except for the comparison means 96 and 97 to which the pattern matching means 93 is connected, the outputs of the thickness detection means 74 and the material detection means 78 input to each of the comparison means 94 and 95 are used as the reference stored in the storage means 92. A comparison is made. If the values match within the allowable range, a signal indicating the type of the genuine coin is output, and if the reference value does not match any type, a signal indicating that it is a false coin is output. Similarly in the determination (S72), the comparison means 96 and 97 related to the image pickup means 16 are excluded, and only when the signals from the comparison means 94 and 95 inputted to the determination means 99 all indicate the same genuine coin type. A signal indicating the type of coin is output, and a signal indicating a false coin is output as the determination signal 100 in other cases.
[0094]
In order to determine which of the determinations S62 and S72 is to be performed by the determination unit 99, the output of the dew condensation detection unit 82 is connected to the determination unit 99. The output of the switching means 98 such as a switch is also input to the determination means 99, and the identification items (thickness and material) excluding the identification items (unevenness and pattern) related to the imaging means 16 for the determination during the dew condensation detection. To avoid accepting sales opportunities as a vending machine, etc. during dew condensation, or forcing the inserted coins during dew condensation because it is difficult to identify with the imaging means 16 Although not shown in FIG. 9, it is possible to select whether to give priority to the counterfeit coin rejection performance without accepting it by the switching means 98. Thus, by providing the switching means 98, it is possible to select whether to give priority to accepting coins or give priority to fake coin elimination performance, and to easily change the selection even at the installation site of a vending machine or the like. In addition, although not shown, a dew condensation prevention unit using a heat generation unit is provided, and by operating this dew condensation prevention unit when dew condensation is detected, it is possible to always perform identification related to the imaging unit 16 while saving power consumption. . As the heat generating means, an electronic component having a large heat generation amount can be used in the vicinity of the passage 13 without using a dedicated component, or an inexpensive heater can be used.
[0095]
In the determination (S62), although not shown in FIG. 9, the identification items (thickness and material in the present embodiment) that are not related to the imaging means 16 are the same type of true coins, and the identification items related to the imaging means 16 ( In the case of unevenness and pattern), if the false coin continues to be a different type of genuine coin, it is determined that the image quality has deteriorated due to dirt on the passage 13 or the like. If all the identification items continue to be the same type of true coins, it is possible to detect the degree of deterioration in image quality by determining that there is no problem in image quality. When a drop in image quality is detected, it is possible to communicate with the control unit of the vending machine and display it to the operator of the vending machine to inform the user that the image quality has deteriorated and to encourage cleaning. It is.
[0096]
In addition to the switching means 98, it is possible to select whether or not to perform identification (hereinafter referred to as image identification) itself by the imaging means 16 by a second switching means (not shown). By providing the second switching means, it is possible to select the one in which image identification is not performed when the imaging means 16 or the optical transparent body 14 has a failure, a serious scratch or dirt, and the dew condensation detection (S14) and the imaging means By not performing the input detection (S15) and the unevenness and pattern detection (S21) in FIG. 16 (not shown in FIG. 9), it is possible to prevent the operation stop of the product due to malfunction and the deterioration of the identification performance.
[0097]
Note that the coin-insertion detection method, the passage configuration described with reference to FIGS. 2 and 4, and the dew condensation detection, dew condensation prevention, and image quality detection can be applied to a pattern identification device having a conventional configuration. Reliable input detection, reliable identification that is not easily affected by dirt, and accurate identification with reduced effects of condensation, dirt, and the like are possible.
[0098]
As described above, according to the present embodiment, it is possible to prevent occurrence of unevenness in an image by performing identification at a non-incident portion of the regular reflection light 29 from the coin 23. Further, by providing the imaging unit 16 vertically above the passage 13, adhesion of dirt to the imaging unit 16 can be reduced. Since the imaging unit 16 includes the image sensor 26 that can output a partial image, the image data of the specific area 66 can be read out in a short time. The imaging means 16 includes an array of light receiving elements arranged in a matrix, a control circuit 68 for supplying a voltage for sensitivity control to the rows of the array, and a neural network for processing a current flowing from a column of the array to the ground. 69, the product-sum operation of the filter matrix and the image can be executed in the image sensor 26. By providing the condensation detection means 82 for detecting whether or not the passage 13 in the vicinity of the imaging means 16 is condensed, it is possible to detect whether or not the passage 13 is condensed. Then, by omitting the control process of the image pickup means 16, it is possible to prevent malfunction due to abnormal image data, to operate the condensation prevention means, or to notify the occurrence of condensation. By detecting that the coin 23 has been inserted using the image pickup means 16, it is possible to reliably detect that the coin 23 has been inserted. Image data with a constant image quality can be obtained by adjusting the reading conditions based on the image read in advance by the imaging means 16. A second illuminating means 36 and a third illuminating means 37 for irradiating the coin 23 with light from different directions having substantially the same wavelength characteristics, are provided. By identifying the coin 23 based on the difference between the reflected light and the reflected light from the coin 23 when irradiated by the third illumination means 37, the unevenness of the coin 23 can be detected. By verifying and detecting a plurality of outer peripheral candidate points 116 as a reference point (center 114) of the position of the coin 23 in the third image, the position of the coin 23 can be reliably detected. By tentatively determining and identifying one or more denominations, it is possible to prevent misdetermination of denominations due to the influence of the passing position of the coins 23. By providing the image pickup means 16 on the vertically upper side of the passage 13, the adhesion of dirt to the image pickup means 16 and the like can be reduced. The edge can be detected faithfully to the pattern of the coin 23 by performing coordinate conversion and identifying after the edge detection enhancement of the third image. By setting a threshold value based on an image excluding a specific portion of the coin 23 and performing binarization with this threshold value, stable binarization can be performed. By performing collation with the reference image based on an image excluding a specific portion of the coin 23, stable collation can be performed. By detecting the rotation angle by thinning out the rotation angle, the rotation angle can be detected at high speed while maintaining accuracy. Since the image of the coin 23 is divided into a plurality of regions 143 and identification is performed based on the degree of coincidence of each region with the reference image, the influence of dirt on the coins 23 and the passage 13 is reduced.
[0099]
Therefore, since the pattern of coins can be accurately identified, a pattern identification device with high identification performance that can prevent unauthorized use of fake coins can be realized.
[0100]
【The invention's effect】
As described above, according to the present invention, the imaging unit includes a passage, an illuminating unit that irradiates light to the identification target that passes through the passage, and an imaging unit that receives reflected or transmitted light from the identification target. In the pattern identifying apparatus for identifying the pattern to be identified based on the image acquired in step (b), the imaging unit performs at least imaging of the identification target and adjustment of input conditions of the imaging unit based on the captured image. After the adjustment step, the pattern identification device is characterized in that the identification target is further imaged by the same imaging means and the pattern of the identification target is identified. It is possible to improve.
[Brief description of the drawings]
FIG. 1 is a front view showing an outline of a coin identification device according to an embodiment of the present invention. FIG. 2 is a cross-sectional view in the vicinity of an imaging means. FIG. 3 is a front view in the vicinity of an illumination means. Sectional view of the passage [FIG. 5] Sectional view of the vicinity of the thickness / material combined sensor [FIG.6] Structure of the control circuit of the image sensor [FIG.7] Structure of the control circuit of the image sensor [FIG. FIG. 8 is a block diagram showing the configuration of the control circuit. FIG. 9 is a flowchart for explaining the operation. FIG. 10 is a flowchart for explaining the operation of detecting irregularities and patterns. FIG. FIG. 12A is a cross-sectional view of the vicinity of a coin when illuminated with only the second illumination means. FIG. 12B is a schematic diagram of image data when illuminated with only the second illumination means. (C) Sectional view of the vicinity of the coin when illuminated by only the third illumination means (d) Fig. 13 is a schematic diagram of image data when illuminated only by the third illumination means. Fig. 13 is a schematic diagram of third image data explaining center / outer diameter detection. Fig. 14A is a coordinate transformation and magnification. Schematic diagram of fourth image data explaining correction (b) Schematic diagram of fifth image data explaining coordinate conversion and magnification correction FIG. 15 is a diagram of a new 500-yen coin explaining threshold setting FIG. 16A is a schematic diagram of sixth image data explaining pattern matching. FIG. 16B is a schematic diagram of sixth image data explaining area division of pattern matching. Explanation of symbols]
DESCRIPTION OF SYMBOLS 11 Coin identification apparatus main body 12 Slot 13 Passage 14 Optical transparent body 15 Illumination means (1st illumination means)
16 Imaging means 17 Thickness / material combined use sensor 18 Outlet 21 Side wall 22 Bottom surface 23 Coin 24 Side wall 25 LED
26 Image sensor 27 Light receiving surface 28 Lens 29 Regular reflection light 31 Coin position 32 Coin position 33 Coin position during imaging 34 Regular reflection position 35 Straight line 36 Second illumination means 37 Third illumination means 41 Discharge auxiliary means 51 Core 52 Core 53 Coil 54 Coil 55 Coil 56 Coil 61 Pixel 62 Horizontal scanning circuit 63 Vertical scanning circuit 64 Horizontal scanning line 65 Vertical scanning line 66 Specific area 67 Output 68 Control circuit 69 Neural network 71 Thickness sensor 72 Oscillation circuit 73 Rectification circuit 74 Thickness detection means 75 Material sensor 76 Oscillation circuit 77 Rectification circuit 78 Material detection means 79 Input detection means 80 A / D conversion means 81 Adjustment means 82 Condensation detection means 83 Concavity and convexity detection means 84 Edge detection means 85 Center / outer diameter detection means 86 Denomination provisional determination means 87 Magnification correction means 88 Coordinate conversion means 89 Threshold setting means 90 Valuation means 91 rotation angle detection means 92 storage means 93 pattern matching means 94 comparison means 95 comparison means 96 comparison means 97 comparison means 98 switching means 99 determination means 100 determination signal 101 imaging range 102 first input detection area 103 second Insertion detection area 104 Third input detection area 105 Coin position 106 Input condition setting area 107 Coin position 108 Concavity and convexity detection line 109 Concavity and convexity detection line 110 Concavity and convexity detection line 111 Detection line 112 Detection line 113 Peripheral candidate point 114 Temporary center 115 Peripheral point 116 Peripheral candidate point 121 Center 122 X-coordinate 123 Y-coordinate 124 R-coordinate 125 S-coordinate 131 Peripheral edge 132 Latent image pattern 133 Fine line pattern 141 Radial direction 142 Circumferential direction 143 Region

Claims (4)

The identification object includes: a passage; an illuminating unit that irradiates light to the identification target that passes through the passage; and an imaging unit that receives reflected or transmitted light from the identification target, and the identification target based on an image acquired by the imaging unit In the pattern identification apparatus for identifying a pattern of the image, the imaging means further includes the same imaging means after an adjustment step of performing imaging of at least the identification target and adjustment of input conditions of the imaging means based on the captured image. A pattern identification device that images the identification target and identifies the pattern of the identification target. The imaging of the identification target in the adjustment step is to image a predetermined region so that the coin position of the assumed minimum speed and the assumed maximum speed of the coin are included. The pattern identification apparatus according to 1. The pattern identifying apparatus according to claim 1, wherein in the adjusting step, the input condition of the imaging apparatus is adjusted by determining brightness and darkness of each image obtained by capturing the identification target for each pixel. In the adjustment process, each time the identification target passes, the adjustment amount of the input condition of the imaging device is stored, and the adjustment amount of the input condition when the identification target passes next is corrected based on the stored adjustment amount. The pattern identification device according to claim 1, wherein:

Images