JP3922608B2 - Height measuring method and apparatus - Google Patents

Description

【００４３】
を算出することにより平坦度Ｆが演算される。演算回路６４は、各３×３の画素ブロックの画素の最大値と階調の最小値の差ΔＢを求め、演算回路６５はＦ／ΔＢを演算する。比較器６６はＦ／ΔＢがしきい値６６’より小さいときには、画像データが平坦でないので、マルチプレクサ６７を切り替える。これにより帯域除去フィルタ回路６８で注目画素に対して
【００４４】
【数２】

【００４５】
の帯域除去フィルタ処理のかけられたデータが出力され、一方ΔＢ＝０の時またはＦ／ΔＢがしきい値より大きい時は遅延回路６９からのデータが選択され、帯域除去フィルタ処理されないデータが出力される。
【００４６】
このように階調データのフィルタ処理が行なわれた画像データは、二値化処理ブロック７０で二値化処理される。そのために、演算回路７１、７２は各ラインの階調の平均値ｘ（上にバー付き）と標準偏差ρを計算する。比較器７５は、１ラインバッファ７３の各画素毎にその画素の階調データがそのラインの階調の平均値ｘ＋ρ×１．５より大きい時は現在の画素の値を１に、以下の時は現在の画素の値を０にして二値化する。
【００４７】
この各二値化された画像データは、二値化データフィルタ処理ブロック８０に送られ、ノイズ除去処理回路８３は各３×３画素ブロック毎に小突起データ並びに孤立データをノイズとして除去する。このノイズ除去は、図７に示したようなａ〜ｆのフィルタ処理を行うことに対応している。３×３の中心の画素を注目画素として、図７のパターンが現れた時、その注目画素の値を０にする。ｆ以外の５種類のフィルタは、９０度づつ回転させて実行する。このようにノイズ処理された二値化画像データは、１ラインバッファ８５、８６に送られる。判定回路８７は、１ラインの全ての画素が０の場合には、前後のラインを参照して穴埋めを行なう。例えば、第２ラインの画素が全て０であった場合、その前後のライン（第１と第３ライン）に１の画素がある場合には、その１の画素のあるところを１にする。
【００４８】
このように処理された画像データは重心位置演算処理ブロック９０に送られ、演算回路９６で重心位置（平均値）が演算される。この重心位置は、図１９に示すように、画素列Ａ、Ｂ、Ｃ・・・に対して１、２、３・・・のような連番を付けることにより行なわれる。この例では１、２行目に関してはＩ列、Ｊ列の画素の値が１であり、Ｉ列の番号は９、Ｊ列の番号は１０なので、１、２行目の重心値は９．５となる。また３、４、５行目に関してはＩ列、Ｈ列、Ｇ列の画素が１であり、各行の重心値は各列に付された番号と同じ値の９、８、７となる。以下同様にして各行の重心値を求める。
【００４９】
このようにして縦軸を重心値、横軸をライン番号としてグラフを描くと図８のようになる。この結果がステップＳ４８で高さデータとしてメモリ１００に格納される。
【００５０】
次に、メモリ１００から図８の高さデータを読み出して、微分を行なって平坦部（レジスト面）の高さを求め、この平坦部を基準としたクリーム半田の高さを求める。
【００５１】
図９の（ａ）は図８の高さデータの一次微分、（ｂ）はその一次微分を更に微分した二次微分のグラフである。微分は左側から行っている。従って、図８を左から右に走査した時、上方への突起部がクリーム半田部の立ち上り部、下方への突起部が立ち下がり部である。立ち下がり部の突起レベルは小さいので、立ち上り部のみ使用した。また右方向からの微分も行い、右から走査した時の立ち上り部（左から走査したときの立ち下り部に相当）も使用した。以下に説明する処理により平坦部とクリーム半田の高さがそれぞれ測定される。
【００５２】
まず、図１０のステップＳ５０で高さ補正メモリ（不図示）を設け、これをクリアしておく。続いて、ステップＳ５１で変数ｘを２とし、ステップＳ５２でそのときの高さデータｈxの前のラインに対する高さデータの比ｈ'xを求め、それをｈ'メモリ（不図示）に格納する（ステップＳ５３）。ｈ'xがしきい値１．２５より大きい場合には（ステップＳ５４）、ステップＳ５５でｈx-1を（ｘ−１）をアドレスとして高さ補正メモリに格納する。ステップＳ５６でｘを＋１して以上の処理をｘ＝ｉ＋１になるまで繰り返す（ステップＳ５７）。
【００５３】
一次微分が終了すると、ステップＳ５８でｘ＝３とし、ステップＳ５９で二次微分値ｈ"xを求め、この二次微分値がしきい値１．２５より大きい場合には（ステップＳ６０）、ステップＳ６１でｈx-1を（ｘ−１）をアドレスとして高さ補正メモリに格納する。続いてステップＳ６２でｘを＋１して以上の処理をｘ＝ｉ＋１になるまで繰り返す（ステップＳ６３）。
【００５４】
図１０が左からの微分であるのに対して、図１１は右からの微分を示すもので、ステップＳ７１で変数ｘをｉ−１とし、ステップＳ７２でそのときの高さデータｈxの次のラインに対する高さデータの比ｈ'xを求め、それをｈ'メモリに格納する（ステップＳ７３）。ｈ'xがしきい値１．２５より大きい場合には（ステップＳ７４）、ステップＳ７５でｈx+1を（ｘ＋１）をアドレスとして高さ補正メモリに格納する。ステップＳ７６でｘを−１して以上の処理をｘ＝０になるまで繰り返す（ステップＳ７７）。
【００５５】
一次微分が終了すると、ステップＳ７８でｘ＝ｉ−２とし、ステップＳ７９で二次微分値ｈ"xを求め、この二次微分値がしきい値１．２５より大きい場合には（ステップＳ８０）、ステップＳ８１でｈx+1を（ｘ＋１）をアドレスとして高さ補正メモリに格納する。続いてステップＳ８２でｘを−１して以上の処理をｘ＝０になるまで繰り返す（ステップＳ８３）。
【００５６】
図１２には、図８のクリーム半田部の測定された高さデータとその一次微分並びに二次微分データの一部（図８のライン４５近辺のデータ）が表として図示されている。図１２の１列目が高さデータ、２列目が左からの一次微分、３列目が左からの二次微分、４列目、５列目が右からの一次微分、二次微分、６列目が抽出されたデータである。例えば、２行目の一次微分値ｈ'xは９．５／９．４＝１．０１となる（図１０のステップＳ５２）。二次微分ｈ"xは一次微分の比１．０１／０．９９＝１．０２となる（図１０のステップＳ５９）。図１２の高さデータ１９の行をみると、一次微分は１．８１、二次微分は１．７２でありしきい値１．２５以上である（ステップＳ５４、Ｓ６０）。そこでこの行の前の行の高さデータ１０．５を抽出する（図１２の６列目に例示されている）。
【００５７】
このように微分値が大きくなるところの高さデータは、図１０のステップＳ５５、Ｓ６１あるいは図１１のステップＳ７５、Ｓ８１で高さ補正メモリに格納され、それが図１１のステップＳ８４で平均値ｈAVとして求められる。図８に示す画像データでは、この平均値は９．３３となった。この平均値はクリーム半田が印刷される平坦部（レジスト面）の高さデータに相当するので、ステップＳ８５で高さデータ（ｈx）からこの平坦部の高さデータ（ｈAV）を減算することにより平坦部からのクリーム半田の実際の高さ（平均高さは１０２μｍ）が算出される。これが図１３に図示されている。予め別の計測器で測定したクリーム半田の平均高さも１０２μｍであったので、上記処理によるクリーム半田の高さが正確であることが確認できた。
【００５８】
なお、上記処理では、一次微分と二次微分値を用いているが、一次微分だけでも平坦部の高さデータを抽出することができる。また、左側あるいは右側の一方の微分値だけで高さデータを抽出するようにしてもよい。
【００５９】
[他の実施形態]
上述した処理だけでは、クリーム半田の立上がり部分以外でも、微分値が大きくなり、平坦部の高さデータを抽出してしまう場合が有ること、クリーム半田の初期立上りではなく、立上り途中でも微分値が大きくなり、高さを抽出してしまう場合が有ること、基板にそりがある場合、光切断画像が傾いて撮像され、高さを正しく計算できないことなどの問題が発生する場合がある。そのために以下のような処理でこれらの問題を解決する。
【００６０】
最初の問題点は、高さデータが抽出された時に、その高さデータの前後の平坦度を調べて、平坦であればその高さは採用せず、平坦でなければ採用することにより解決する。また、２番目の問題点は、注目高さを採用する際に、複数個前迄にすでに高さデータが抽出されているかどうかを調べ、採用されていなければ、現在の注目高さデータを新たに採用し、採用されていれば、現在の注目高さデータは放棄するようにする。そのために、図１０のステップＳ５４（Ｓ６０）とステップＳ５６（Ｓ６２）の間に図１４（ａ）、（ｂ）に示した処理を設ける。
【００６１】
図１４（ａ）のステップＳ９０では、ステップＳ５４（Ｓ６０）で一次微分値あるいは二次微分値がしきい値以上で高さデータが平坦部の高さデータとして抽出された場合、その近辺で平坦度Ｆ’
【００６２】
【数３】

【００６３】
を計算する（なお、二次微分のときはｈに代わりｈ’を用いる）。続いてステップＳ９１でこの平坦度がしきい値（２）よりも大きいかを判断し、大きい場合には、抽出した注目高さデータを採用し、それ以外は不採用とする。図１２の表のデータで実際に計算してみると、抽出した高さデータは１０．５であるから、数３による平坦度Ｆ’は、Ｆ’＝（｜９．５−１０｜＋｜１０−１９｜＋｜１９−２０｜＋｜２０−９．５｜）／４＝５．２５＞２（しきい値）となる。従って、注目高さデータとして採用され、ステップＳ５５（Ｓ６１）で高さ補正メモリに格納する。
【００６４】
２番目の問題点は、ステップＳ９１とステップＳ５５（Ｓ６１）の間に図１４（ｂ）に示したステップＳ９２の処理を設けることにより解決できる。同ステップでは、例えば５つ前まで、すなわち（ｘ−２）から（Ｘ−６）に高さデータが採用されているかどうかを調べる。採用されてない場合には、注目高さデータとして採用され、ステップＳ５５（Ｓ６１）で高さ補正メモリに格納する。
【００６５】
図１５は、図１４が左から微分したときのに対して右側から微分したときに対応する処理であり、平坦度Ｆ”が
【００６６】
【数４】

【００６７】
により計算され、２番目の問題点を解決するために、ステップＳ９６で（ｘ＋２）から（Ｘ＋６）に高さデータが採用されているかを調べているところが図１４と異なる。
【００６８】
３番目の問題点の解決としては、前記微分処理と最初と２番目の問題点の解決手段で抽出採用された高さデータを多重線形回帰によって計算し、近似直線を求めて傾きの補正を行う。このために、図１６に図示したように、図１１のステップＳ８３の次に、高さ補正メモリ内のｈxの数ｎ、すなわち抽出採用された高さデータの数を調べる（ステップＳ１００）。続いて、ステップＳ１０１で高さ補正メモリ内のｈx＝ｙとして、
【００６９】
【数５】

【００７０】
の各値を求め、次にステップＳ１０２で
【００７１】
【数６】

【００７２】
よりａ0、ａ1を求め、近似式ａ0＋ａ１ｘ＝ｙを求める。
【００７３】
例えば、抽出した注目高さデータが（６２，９）（１５１，１０）（１８４，１０．５）（２３８，１１．５）（３６２，１２）（４４９，１３）とすると、ｎ＝６，Σｘ＝１４４６，Σｙ＝６６，Σｘ＾２（ｘの２乗のΣ）＝４４９７９０，Σｘｙ＝１６９１８から、ａ0＝８．５９２４７４，ａ1＝０．００９９９が算出される。従って、近似式は：８．５９２４７４＋０．００９９９ｘ＝ｙとなる。
【００７４】
次に、ステップＳ１０３からＳ１０６において近似式のｘを1から高さデータの数（ｉ＋１）までのｙの値を求め、ステップＳ１０４でｈxの補正をする。今、高さデータ数を４８０個とすると、前記４８０個の高さデータと４８０個のｙの差を求めれば、これが傾きを補正した４８０個の高さデータとなる。図１７（ａ）、（ｂ）はその演算結果を示している。
【００７５】
図１８はそりのある基板で測定し、傾き補正をしないで０点の補正を行った例であり（図１１ステップＳ８５）、高さデータのグラフは右肩上がりとなっているのが分かる。これに図１６の処理に沿った傾き補正を行うと、図１３と同じような結果が得られる。
【００７６】
［表計算ソフトによるデータ処理］
上述した実施形態では、ＶーＲＡＭ画像メモリ４０の画像データは、図４に示す回路構成で画像処理されたが、ＶーＲＡＭ画像メモリ４０の画像データを表計算ソフトに取り込んで行なうこともできる。ＶーＲＡＭ画像メモリ４０の画像データは、横６４０画素×縦４８０画素のビットマップ画像であるので、これを各画素を２５６階調の階調データに変換した後、６４０列×４８０行のセルの表計算ソフトに取り込む。画素間の分解能は１０μｍであるので、表計算ソフトに読み込んだ場合は前記画素がセルに相当することから、セル間のピッチは１０μｍとなる。ただし、実際の表計算ソフトは最大列数が２５６列という機能上の制約が有るので、２００列×４８０行の階調データを取り込んで処理を行う。
【００７７】
まず、取込んだ階調データは３×３のセル毎に取り出されて、階調の平坦度を調べてフィルタ処理が行なわれる。３×３のセルの中心のセルを注目セルとし、その周りの階調の平坦度を計算する。注目セル周囲のセル間差の絶対値の平均値を求める。列をＡ、Ｂ、Ｃ・・・・、行を１、２、３・・・・として、例えば、注目セルをＢ２として、数１に従い平坦度Ｆを算出する（図４の演算回路６３による演算に対応）。次に３×３のセルの中の階調の最大値と階調の最小値の差ΔＢに対する比を求め（演算回路６５に対応）、Ｆ／ΔＢがしきい値より小さい時（比較器６６に対応）に注目セルに対して、数２の帯域除去フィルタ（フィルタ回路６８に対応）をかける。もしΔＢ＝０の時またはＦ／ΔＢがしきい値より大きい時は何もしない。次に注目セルをＢ３に移し以上の処理を実行し、Ｂ４、Ｂ５・・・・と処理をする。そして次の行に移行してＣ２、Ｃ３・・・のように順次処理を行う。
【００７８】
以上のように階調データのフィルタ処理が終わると、次の二値化処理に移る。各行の階調の平均値と標準偏差を計算する（演算回路７１、７２に対応）。そして各セル毎にそのセルの階調データがその行の階調の平均値＋１．５×標準偏差より大きい時（比較器７５に対応）は現在のセルの値を１に書換え、以下の時は現在のセルの値を消去する。セルＡ１、Ｂ１、Ｃ１・・・に対しては１行目の平均値と標準偏差を用い、セルＡ２、Ｂ２、Ｃ２・・・に対しては２行目の平均値と標準偏差を用いる。各セルは１か空白の状態になる。
【００７９】
このように各セルの値が二値化されたあと、小突起データ並びに孤立データはノイズと考えられるので、ノイズ除去のためにこれらのデータの消去処理を行う。すなわち、図７に示す様なａ〜ｆのフィルタ処理を行う。３×３の中心のセルを注目セルとして、図７のパターンが現れた時、その注目セルの値を消去する。ｆ以外の５種類のフィルタは、９０度づつ回転させて実行する（ノイズ除去処理回路８３に対応）。
【００８０】
次は各行を参照して、全て空白の場合は上下の行の数値１のセルの配列を参照して、穴埋めを行う（判定回路８７に対応）。次に、各行の数値１のセルに対して重心値を計算する。これは、図１９に示すように、列Ａ、Ｂ、Ｃ・・・に対しては、１、２、３・・・と連番を付けるとこの数値が重心の値となる（重心位置演算回路９６に対応）。この例では１、２行目に関してはＩ列、Ｊ列のセルの値が１であり、Ｉ列の番号は９、Ｊ列の番号は１０なので、１、２行目の重心値は９．５となる。また３、４、５行目に関してはＩ列、Ｈ列、Ｇ列のセルが１であり、各行の重心値は各列に付された番号と同じ値の９、８、７となる。以下同様にして各行の重心値を求める。
【００８１】
このように、図４の各処理ブロック６０、７０、８０、９０をソフトウェアで処理することもできる。
【００８２】
また、図１４のステップＳ９０の平坦度の計算は、表計算では、例えばＨ３のセルを注目高さとして抽出した時に、Ｆ’＝（｜Ｈ１−Ｈ２｜＋｜Ｈ２−Ｈ４｜＋｜Ｈ４−Ｈ５｜＋｜Ｈ５−Ｈ１｜）／４として計算する。しきい値をＳとするとＦ’＞Ｓの時に抽出した注目高さデータを採用し、それ以外は不採用とする。実施例ではＳ＝２とした。その他２番目あるいは３番目の問題点を解決するための処理も表計算を用いて同様に行なうことができる。
【００８３】
【発明の効果】
以上説明したように、本発明では、撮像された突出物の像から突出物の高さデータを求め、その高さデータを一次微分と二次微分あるいは一次微分し、その微分値から平坦部の高さデータを求めるようにしているので、突出物が形成される前の平坦部の高さデータを求める必要がなく、平坦部に突出物を形成するごとに突出物の高さを測定しなければならないような場合に、測定のタクトタイムを顕著に向上させることができる。
【００８４】
また、本発明での高さデータ処理では、撮像された画素の階調データに対して平坦化処理がなされたあと二値化されるので、安定した高さデータが得られる。また二値化データに対してフィルタ処理を行なって、その結果空白行が発生した場合、その前後の行の二値化データの配列に応じて空白行にデータの埋め込みが行なわれるので、信頼性のある高さデータが求められる。
【００８５】
また、本発明では、突出部の立ち上がり部以外で微分値が大きくなっても、抽出された高さデータ近辺の平坦度を調べ、所定よりも平坦でなければ抽出された高さデータを平坦部の高さデータとしているので、信頼性のある高さデータが得られる。
【００８６】
また、突出部の立ち上がりの途中で、微分値が大きくなっても、それ以前の複数個の高さデータの中からすでに平坦部の高さデータが採用されているときは、その抽出された高さデータを破棄することにより同様に信頼性のある高さデータが保証される。
【００８７】
更に、被測定物のライン光に沿った方向に被測定物にそりがある場合には、平坦部の高さデータとして採用された複数個の高さデータに基づいて多重線形回帰を用いて平坦部の高さデータの近似式を求め、その近似式から平坦部のライン光に沿った方向の傾き補正を行なっているので、突出物の形成される基板に傾きがあってもそれを補正して正確な高さ測定を行なうことができる。
【図面の簡単な説明】
【図１】本発明に用いられる高さ測定装置の構成を示す斜視図である。
【図２】図１の高さ測定装置の側面図である。
【図３】ライン光を投光して得られる像から画像データを取得する回路構成を示した回路図である。
【図４】取得された画像データを処理する回路構成を示した回路図である。
【図５】クリーム半田部分にライン光を投光した場合に得られる画像の処理の流れを示したフローチャート図である。
【図６】クリーム半田部分にライン光を投光した場合に得られる画像を示した説明図である。
【図７】画像データのフィルタ処理に用いられるフィルタデータを示す説明図である。
【図８】クリーム半田の高さデータを示す線図である。
【図９】図８のクリーム半田の高さデータの一次微分及び二次微分を示す線図である。
【図１０】高さデータを左側から微分して微分値を求める流れを示したフローチャート図である。
【図１１】高さデータを右側から微分して微分値を求める流れを示したフローチャート図である。
【図１２】微分値から配線基板の平坦部の高さデータを求める過程を示した表図である。
【図１３】微分値に基づいて得られた配線基板の平坦部から測定したクリーム半田の高さデータを示した線図である。
【図１４】クリーム半田の立ち上がり部分以外で抽出された信頼性のないデータを除去するための処理を示したフローチャートである。
【図１５】図１４と同様な処理で逆側から微分を行なった場合の処理を示すフローチャートである。
【図１６】基板のそりを補正する流れを示したフローチャートである。
【図１７】基板にそりがあるときの処理データを示す表図である。
【図１８】基板にそりが有る場合の高さデータを示す線図である。
【図１９】画素あるいはセルの情報から重心値を求めるための例を示した説明図である。
【図２０】従来の三次元測定装置の構成を示した斜視図である。
【符号の説明】
１ レーザ光源
５ ラインジェネレータ
６ ＣＣＤカメラ
９ ライン光[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a height measuring method and apparatus, and more particularly, to a height measuring method and apparatus for measuring a fine height such as cream solder printed by a cream solder printer used in a surface mounting system.
[0002]
[Prior art]
Conventionally, a three-dimensional recognition device using a light cutting method is known as a device for recognizing a three-dimensional shape and measuring the height. FIG. 20 shows a three-dimensional recognition apparatus using this light cutting method. The line light 122 from the line light generator 121 that is a light source is projected at a predetermined angle from obliquely above the object to be measured 123, and an image formed along the surface shape formed on the surface of the object to be measured 123 is vertical. The image is taken by the CCD camera 124 from above. The image captured by the CCD camera 124 is A / D converted by the CCD camera controller 125 and captured by the image capturing device 126. Then, the acquired data is converted into the three-dimensional coordinates of the measurement object 123 by the coordinate calculator 127.
[0003]
In a cream solder height measuring apparatus that is used by being incorporated in a cream solder printer, a portion (measurement unit) 128 surrounded by a dotted line in FIG. 20 is incorporated in an XY moving gantry (XY moving mechanism). Is done. First, when a printed wiring board is carried into the cream solder printing machine, after the positioning of the wiring board and the stencil is completed, the measurement unit 128 is moved from the initial expected avoidance position to the target measurement position by the XY moving gantry. Then, the measurement unit captures an image of the line light formed on the pad surface (resist surface) on the wiring board, which is the object to be measured, by the CCD camera 124 and then retreats to the initial expected avoidance position. Next, cream solder is printed on the pad surface of the wiring board. After the printing on the pad surface of the wiring board is completed, the measurement unit 128 is moved again to the target measurement position by the XY moving gantry, and the image of the line light formed along the shape of the cream solder is displayed on the CCD camera 124. Then, the image is moved back to the first expected avoidance position.
[0004]
From the above imaging data, the center-of-gravity position coordinates in the height direction of the pad surface of the wiring board and the center-of-gravity position coordinates in the height direction of the cream solder portion are calculated. Then, from the subtraction of the center of gravity position coordinate in the height direction of the pad surface of the wiring board and the center of gravity position coordinate in the height direction of the cream solder part, the height of the cream solder part after printing is determined with reference to the pad surface of the wiring board. calculate. And the average height of the cream solder part over each pad surface is calculated.
[0005]
In order to obtain a three-dimensional shape, a plurality of data with different light cutting positions are required. For example, assume that light cutting is performed at a pitch of 50 μm in order to obtain a three-dimensional shape of cream solder printed on a pad having a length of 2 mm. In this case, before printing the cream solder, an image of line light formed on the pad surface of the wiring board is taken 40 times by the CCD camera while changing the position of light cutting. Furthermore, after the cream solder is printed, it is necessary to capture the image of the line light formed along the shape of the cream solder 40 times while changing the light cutting position with a CCD camera. Therefore, the total number of times of imaging by the CCD camera is 80 times. Similarly, the total number of micro movements of the measurement unit is 80 times.
[0006]
[Problems to be solved by the invention]
However, in a cream solder height measuring device that is used by being incorporated in a conventional cream solder printing machine, the image of the line light before printing and the image of the line light after printing are picked up by a CCD camera and the cream solder is used. The height of must be calculated. For this reason, there is a problem that the tact time becomes long and the performance of the entire semiconductor surface mounting system is lowered.
[0007]
Furthermore, in order to change the light cutting position minutely, the heavy measurement unit equipped with the CCD camera must be moved. For the XY moving gantry, the skip function to the target measurement position and the measurement purpose There is a problem that the servo system for driving the XY moving gantry is complicated because it must have two functions of minute movement at the position.
[0008]
Accordingly, the present invention has been made to solve such a problem, and provides a height measuring method and apparatus capable of measuring the exact height by greatly improving the tact time. Let that be the issue.
[0009]
[Means for Solving the Problems]
  In the present invention, in order to solve this problem,
  Line light is projected onto the object to be measured that has a protrusion protruding from the flat part, and the height of the protrusion from the flat part is measured by the light cutting method by imaging the flat part and the protrusion cut by the line light. In the height measurement method to
  Find the height data of the protrusion from the image of the captured protrusion,
  First and second derivative or first derivative of the height data,
  The differential valueExtract the previous height data when is over the threshold,
  The flatness in the vicinity of the extracted height data is checked, and if it is not flatter than a predetermined value, the extracted height data is used as the height data of the flat portion.The configuration is adopted.
[0010]
  In the present invention,
  Line light is projected onto the object to be measured that has a protrusion protruding from the flat part, and the height of the protrusion from the flat part is measured by the light cutting method by imaging the flat part and the protrusion cut by the line light. In the height measuring device to
  A light projecting device that projects line light onto the protrusion;
  Means for picking up an image of the projection projected with the line light;
  Means for processing the image data of the protrusion by the line light to calculate the height data of the protrusion;
  Means for first and second derivative or first derivative of the height data;
  Means for calculating the height of the flat portion from the differential value;
  Means for calculating the height of the protrusion based on the height of the flat portion.And
  The means for calculating the height of the flat part extracts the previous height data when the differential value exceeds a threshold value, checks the flatness near the extracted height data, and is flatter than a predetermined value. Otherwise, the extracted height data is used as the height data of the flat part.The configuration is also adopted.
[0011]
In such a configuration, the height data of the flat portion of the object to be measured is obtained from the primary differential value and the secondary differential value or the primary differential value of the height data obtained by processing the image data of the object to be measured. Since the height data of the protrusion is calculated based on the height of the portion, there is no need to obtain the height data of the flat portion before the protrusion is formed, and every time the protrusion is formed on the flat portion. The tact time can be remarkably improved when the height of the protrusion must be measured, for example, when cream solder is printed on the wiring board and the height of the cream solder must be measured. .
[0012]
The present invention is based on the idea that the differential value increases at the rising portion of the protruding portion, but there is a problem that the differential value also increases in other cases. This problem is solved by checking the flatness in the vicinity of the extracted height data, and if the flatness is not flatter than a predetermined value, the extracted height data is used as the height data of the flat portion.
[0013]
In addition, the differential value may increase even during the rise of the protrusion, but this problem is that when the height data is extracted, the height of the flat part is already selected from the plurality of previous height data. When data is adopted, the problem is solved by discarding the extracted height data.
[0014]
Furthermore, accurate measurement cannot be performed when there is a warp in the direction along the line light of the object to be measured. However, this problem is caused by a plurality of heights adopted as the height data of the flat part. This is solved by obtaining an approximate expression of the height data of the flat portion using multiple linear regression based on the data, and correcting the inclination in the direction along the line light of the flat portion from the approximate expression.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, the present invention will be described in detail based on embodiments shown in the drawings.
[0016]
[Basic configuration]
FIG. 1 is a basic configuration diagram of main optical components of a three-dimensional measuring apparatus showing an embodiment of the present invention, and FIG. 2 is a side view thereof. In each figure, reference numeral 1 denotes a laser diode as a laser light source. The laser light emitted from the laser diode 1 is collimated by a collimator lens 2 into a parallel light beam 1a parallel to the optical center axis. The laser parallel light beam 1a enters a light projecting unit 7 in which a focus lens 3, a light projecting mirror 4, and a line generator lens 5 are incorporated. The laser beam is narrowed down to become spot light by the focus lens 3 in the light projecting unit 7, reflected by the light projecting mirror 4 at an angle of 45 degrees with respect to the vertical axis (Z direction), and by the line generator lens 5. The line light 9 having a width of 14 to 20 μm and a length of 10 mm is formed, and the line light 9 is formed in the X direction on the object to be measured (cream solder or a wiring board on which it is printed) 11.
[0017]
Scattered and reflected light from the object to be measured 11 is imaged by a non-interlaced CCD camera 6 having a field of view of 6.4 mm × 4.8 mm arranged so that the optical axis 6a is a vertical axis. Further, the light projecting unit 7 is attached to the shaft 8a of the linear motor 8, and the linear motor 8 performs linear motion with a stroke of about 10 mm in the Y direction, as shown by double arrows. It reciprocates parallel to the light beam 1a. With this movement of the light projecting unit 7, the line light 9 moves in the direction perpendicular to the X direction in which the line light extends.
[0018]
[Circuit configuration (data acquisition)]
FIG. 3 is a block diagram showing a circuit configuration for acquiring image data of an object to be measured in the three-dimensional measuring apparatus. In the figure, a linear motor drive command device 31 receives a start signal from the CPU 44, outputs a command pulse train for driving the linear motor to the linear motor driver 32, and moves the linear motor 8 by 0.25 μm per pulse. The linear motor drive command unit 31 simultaneously sends a forward / reverse signal to an LD (laser diode) on / CCD trigger timing decoder (hereinafter referred to as timing decoder) 35, and the linear motor 8 moves in either the forward direction or the reverse direction. Let them know.
[0019]
The linear motor driver 32 receives the command pulse train to drive the linear motor, receives the signal indicating the actual position from the position encoder 8b built in the linear motor 8, adjusts the supply voltage to the linear motor 8, and The position is feedback controlled.
[0020]
The position counter 34 receives an A-phase signal and a B-phase signal having a phase difference of 90 degrees from the position encoder 8b of the linear motor 8, and outputs a position signal indicating the position of the linear motor 8 as digital data. The position counter 34 is reset by an origin reset signal from the position encoder 8b.
[0021]
The timing decoder 35 receives the position data from the position counter 34 and outputs an LD ON timing signal (160 μm pitch). In response to the rising edge of this signal, the one-shot multivibrator (MS) 46 outputs a 2 ms wide LD on pulse to the laser diode driver 36 and turns on the laser diode 1 in pulses. The laser diode 1 incorporates a light amount monitor photodiode (not shown), whereby the light amount of the laser diode 1 is controlled to be constant.
[0022]
Since the linear motor 8 is moving even while the LD is on, the correction is made to shift the line light position depending on the moving direction, and the half pitch (80 μm in the embodiment) is shifted in the reverse direction to the forward direction. Therefore, the forward / reverse direction signal is input from the linear motor drive command device 31 as described above.
[0023]
The position latch 37 latches the position data output from the position counter 34 at the rising edge of the LD on timing, and the position in the Y direction of the line light at that time is synchronized with the falling timing of the LD on pulse from the MS 46. To tell. Even when the LD is on, the line light moves and causes a deviation from the actual line light position. This is corrected in the CPU 44 by the line light moving speed, the LD on time, and the forward / reverse signal. Yes.
[0024]
On the other hand, the synchronization signal timing generator 39 receives the CCD camera synchronization timing signal from the timing decoder 35 via the OR circuit 35 ', and outputs a VD vertical synchronization signal that is synchronized with the HD horizontal synchronization signal. In association with the VD vertical synchronization signal, reading of the light amount data of each pixel in the previous frame is started. At the same time, accumulation of light quantity at each pixel starts, and an image by the line light 9 is acquired on the CCD area image sensor 38 of the CCD camera 6.
[0025]
The image on the CCD area image sensor 38 is read as a light amount value (analog value) of each pixel via the driver 33 by a vertical register transfer clock, a horizontal register transfer clock, etc. from the synchronization signal timing generator 39. This is input to the A / D converter 48 via the amplifier 47 and input to the V-RAM image memory 40 as digital data.
[0026]
The horizontal address counter 42 is reset after a predetermined number of horizontal clocks from the falling edge of the horizontal synchronizing signal from the synchronizing signal timing generator 39, and then the horizontal position of the current pixel in the effective screen is counted by counting the horizontal clocks thereafter. (Horizontal address) is output. The horizontal address value is input to the horizontal address of the V-RAM image memory 40 via the multiplexer 41.
[0027]
On the other hand, the vertical address counter 43 is reset after a predetermined number of horizontal synchronization signals from the falling edge of the vertical synchronization signal from the synchronization signal timing generator 39, and then counts the pulses of the horizontal synchronization signal, thereby counting the number of horizontal synchronization signals. The vertical position (vertical address) of the current pixel is output. This vertical address value is input to the vertical address of the V-RAM image memory 40 via the multiplexer 45.
[0028]
The write timing generator 49 writes a write signal to the V-RAM image memory 40 in synchronization with the horizontal clock between the effective horizontal scanning section signal from the horizontal address counter 42 and the effective vertical scanning section signal from the vertical address counter 43. Is output. Thus, the V-RAM image memory 40 stores each pixel data in the effective screen determined by the effective horizontal scanning section signal and the effective vertical scanning section signal.
[0029]
The horizontal / vertical address generator 50 outputs a horizontal address signal to the multiplexer 41, a vertical address signal to the multiplexer 45, and a read signal to the V-RAM image memory 40. The multiplexers 41 and 45 are switched to the reading side in response to the switching signal from the CPU 44, and the image data in the V-RAM image memory 40 having addresses determined by the horizontal address signal and the vertical address signal are sequentially read in synchronization with the reading signal. The reading of the image data is performed after the writing of the image data to the V-RAM image memory 40 is completed. This is ensured by switching the multiplexers 41 and 45 from the writing mode to the reading mode by the switching signal from the CPU 44.
[0030]
[Circuit configuration (data processing)]
4 shows a circuit configuration for processing the image data stored in the V-RAM image memory 40. The circuit configuration centered on the V-RAM image memory 40 is that shown in FIG. The same is shown.
[0031]
Image data read from the V-RAM image memory 40 in synchronization with the read signal when the multiplexers 41 and 45 are switched to read by the switching signal is input to the gradation data filter processing block 60 and noise is removed. . The gradation data filter processing block 60 is provided with two one-line buffers 61 and 62, whereby image data for three lines can be obtained simultaneously. The image data for these three lines is input to the arithmetic circuit 63, the flatness F of the image data is calculated, and also input to the arithmetic circuit 64, and the maximum gradation value MAX and the minimum gradation value MIN. The difference ΔB is calculated. The image data for three lines is also input to the band elimination filter 68, where the band elimination filter is applied. The output of the 1-line buffer 61 is input to the delay circuit 69, and a delay corresponding to the calculation processing time is applied. Note that the processing of the arithmetic circuits 63 and 64 and the band elimination filter 68 is performed in units of 3 × 3 pixel blocks.
[0032]
The arithmetic circuit 65 calculates the ratio of the flatness F of the image data to the difference ΔB, and the comparator 66 compares the calculation result with a threshold value 66 '. If it is less than the threshold value, the multiplexer 67 selects the image data subjected to the band removal filter processing by the band removal filter 68, and if it is greater than the threshold value, the delay circuit 69 applies a delay corresponding to each computation time. The selected image data is selected and output to the binarization processing block 70.
[0033]
In the binarization processing block 70, the average value calculation circuit 71 calculates an average value x (with a bar above) for each line of the image data from the gradation data filter processing block 60, and a standard deviation calculation circuit. 72, the standard deviation ρ is calculated for each line, and the threshold value calculation circuit 74 calculates the threshold value (x + 1.5ρ). The comparator 75 compares this threshold value with the image data for one line held in the one line buffer 73, and binarizes the image data.
[0034]
The binarized image data from the binarization processing block 70 is input to the noise removal processing circuit 83 and the two one-line buffers 81 and 82 of the binarization data filter processing block 80. The noise removal filter processing circuit 83 receives three lines of image data simultaneously from the two one-line buffers 81 and 82 on the input side and the direct image data, and small protrusions and isolated data are generated for each 3 × 3 image block. If it exists, if it exists, the process which removes the data is performed. The output of the noise removal processing circuit 83 is input to the determination circuit 87 and the 1-line buffer 85. The determination circuit 87 determines whether all of one line is 0. If all the one lines are 0, the multiplexer 89 selects the output of the OR circuit 88, and otherwise, selects the output of the 1 line buffer 85. Is input to the 1-line buffer 86. The image data in the 1-line buffer 85 corresponds to the current image data, and the output from the noise removal processing circuit 83 and the image data in the 1-line buffer 86 correspond to the image data before and after that. The image data filled in by OR processing of the data at the same horizontal position of the preceding and following lines is output.
[0035]
The binarized image data from the binarized data filter processing block 80 is input to the centroid position calculation processing block 90, and the centroid position is calculated for each line. The rising edge detection circuit 91 of the center-of-gravity position calculation processing block 90 detects that the binarized image data changes from “0” to “1”, and the horizontal address value of the horizontal address counter 93 at that time is latch circuit 94. Latch on. The falling edge detection circuit 92 detects that the binarized image data changes from “1” to “0”, and latches the horizontal address value of the horizontal address counter 93 at that time in the latch circuit 95. The barycentric position calculation circuit 96 calculates the barycentric position by averaging the horizontal address values at the time of rising and falling, and stores the value in the barycentric position calculation result memory 100. The horizontal address counter 93 counts a horizontal clock for reading from the V-RAM image memory in order to obtain a horizontal address value. The horizontal address counter 93 is reset when the image data 1 line is switched.
[0036]
[Measurement of height data]
Next, a process for obtaining height data of a wiring board or cream solder will be described with reference to the flow of FIG. 5 taking an example of cream solder printed on the wiring board as an object to be measured.
[0037]
First, after confirming that cream solder is printed in step S40, in step S41, the CPU 44 generates a position command signal and a start signal in the linear motor drive command device 31, and the cream solder is printed on the linear motor 8. Move to the height measurement position. When this actual height measurement position is reached, the CPU 44 sends out a CCD camera synchronization timing pulse via the OR circuit 35 '(step S42), and generates an LD ON signal to light the laser diode 1 for 2 ms, for example ( Step S43).
[0038]
The laser light emitted from the laser diode 1 is condensed by the collimator lens 2, becomes a parallel light beam 1 a parallel to the optical center axis, and is narrowed down to become spot light by the focus lens 3. The laser spot light is reflected by the light projecting mirror 4 in a direction of 45 degrees with respect to the incident angle, and enters the line generator lens 5. The laser spot light is stretched in one direction (X direction) by the lens 5 by the lens 5 and becomes line light 9 having a width of 14 μm and a length of 10 mm on the object to be measured 11. This line light is imaged by a non-interlaced CCD camera 6 with a visual field of 6.4 mm × 4.8 mm.
[0039]
After waiting for the time T1 in step S44, a CCD camera synchronization timing pulse is sent in step S45, the synchronization signal timing generator 39 is driven, and the image data of the image sensor 38 of the CCD camera 6 is output from the write timing generator 49. In synchronism with this, it is read into the V-RAM image memory 40 (step S46). The image data obtained in this way is shown in FIG. In the figure, a high brightness image protruding to the right side is the cream solder portion 113a, and a straight line portion having a low brightness therebetween is the resist surface or pad surface 113b of the wiring board.
[0040]
This image data is subjected to each image processing in step S47. First, the multiplexers 41 and 45 are switched to the read mode by the switching signal of the CPU 44, and the image data is read from the V-RAM image memory 40 in synchronization with the horizontal address and the vertical address according to the read signal from the horizontal / vertical address generator 50. .
[0041]
The read image data is subjected to gradation data filter processing in a gradation data filter processing block 60. The arithmetic circuit 63 calculates the flatness F of gradations around the pixel at the center of each 3 × 3 pixel block as the target pixel. The flatness F is obtained as an average value of absolute values of differences between pixels around the target pixel, and pixel columns are A, B, C,..., Pixel rows are 1, 2, 3,. For example, if the pixel of interest is B2,
[0042]
[Expression 1]

[0043]
The flatness F is calculated by calculating. The arithmetic circuit 64 calculates a difference ΔB between the maximum pixel value and the minimum gradation value of each 3 × 3 pixel block, and the arithmetic circuit 65 calculates F / ΔB. When F / ΔB is smaller than the threshold value 66 ′, the comparator 66 switches the multiplexer 67 because the image data is not flat. As a result, the band elimination filter circuit 68 applies to the target pixel.
[0044]
[Expression 2]

[0045]
On the other hand, when ΔB = 0 or when F / ΔB is larger than the threshold value, the data from the delay circuit 69 is selected, and the data not subjected to the band removal filter processing is output. Is done.
[0046]
The image data on which the gradation data has been filtered in this way is binarized by the binarization processing block 70. For this purpose, the arithmetic circuits 71 and 72 calculate the average value x (with a bar above) and the standard deviation ρ of the gradations of each line. The comparator 75 sets the current pixel value to 1 when the gradation data of each pixel of the line buffer 73 is larger than the average value x + ρ × 1.5 of the gradation of the line, and when Will binarize the current pixel value to 0.
[0047]
Each of the binarized image data is sent to the binarized data filter processing block 80, and the noise removal processing circuit 83 removes the small protrusion data and the isolated data as noise for each 3 × 3 pixel block. This noise removal corresponds to performing the filter processes a to f as shown in FIG. With the pixel at the center of 3 × 3 as the target pixel, when the pattern of FIG. 7 appears, the value of the target pixel is set to zero. The five types of filters other than f are rotated by 90 degrees and executed. The binarized image data subjected to noise processing in this way is sent to the 1-line buffers 85 and 86. When all the pixels in one line are 0, the determination circuit 87 performs hole filling with reference to the preceding and following lines. For example, if all the pixels in the second line are 0, and there is 1 pixel in the preceding and succeeding lines (first and third lines), the place where the 1 pixel exists is set to 1.
[0048]
The image data processed in this way is sent to the centroid position calculation processing block 90, and the centroid position (average value) is calculated by the calculation circuit 96. As shown in FIG. 19, the center of gravity is determined by assigning sequential numbers such as 1, 2, 3,... To the pixel rows A, B, C. In this example, for the first and second rows, the pixel values in the I and J columns are 1, the I column number is 9, and the J column number is 10. Therefore, the centroid values in the first and second rows are 9. 5 For the third, fourth, and fifth rows, the pixels in the I, H, and G columns are 1, and the center-of-gravity value of each row is 9, 8, and 7, which is the same value as the number assigned to each column. In the same manner, the center of gravity value of each row is obtained.
[0049]
When the graph is drawn with the vertical axis as the center of gravity and the horizontal axis as the line number in this way, the graph is as shown in FIG. This result is stored in the memory 100 as height data in step S48.
[0050]
Next, the height data of FIG. 8 is read from the memory 100, and differentiation is performed to obtain the height of the flat portion (resist surface), and the height of the cream solder with reference to the flat portion is obtained.
[0051]
(A) of FIG. 9 is a graph of the first derivative of the height data of FIG. 8, and (b) is a graph of the second derivative obtained by further differentiating the first derivative. Differentiation is performed from the left side. Therefore, when FIG. 8 is scanned from left to right, the upward protruding portion is the rising portion of the cream solder portion, and the downward protruding portion is the falling portion. Since the protrusion level of the falling part is small, only the rising part was used. Also, differentiation from the right direction was performed, and a rising portion when scanning from the right (corresponding to a falling portion when scanning from the left) was also used. The height of the flat portion and the cream solder is measured by the processing described below.
[0052]
First, in step S50 of FIG. 10, a height correction memory (not shown) is provided and cleared. Subsequently, in step S51, the variable x is set to 2, and in step S52, the ratio h′x of the height data to the previous line of the height data hx at that time is obtained and stored in the h ′ memory (not shown). (Step S53). If h'x is larger than the threshold value 1.25 (step S54), hx-1 is stored in the height correction memory using (x-1) as an address in step S55. In step S56, x is incremented by 1 and the above processing is repeated until x = i + 1 (step S57).
[0053]
When the primary differentiation is completed, x = 3 is set in step S58, and a secondary differential value h "x is obtained in step S59. If this secondary differential value is larger than the threshold value 1.25 (step S60), step In step S61, hx-1 is stored in the height correction memory using (x-1) as an address, and then x is incremented by 1 in step S62 and the above processing is repeated until x = i + 1 (step S63).
[0054]
FIG. 10 shows the differentiation from the right, whereas FIG. 11 shows the differentiation from the right. In step S71, the variable x is set to i−1, and in step S72, the next height data hx is obtained. The ratio h′x of the height data to the line is obtained and stored in the h ′ memory (step S73). If h'x is larger than the threshold value 1.25 (step S74), hx + 1 is stored in the height correction memory using (x + 1) as an address in step S75. X is decremented by 1 in step S76, and the above processing is repeated until x = 0 (step S77).
[0055]
When the primary differentiation is completed, x = i−2 is set in step S78, and a secondary differential value h ″ x is obtained in step S79. If this secondary differential value is larger than the threshold value 1.25 (step S80). In step S81, hx + 1 is stored in the height correction memory using (x + 1) as an address, x is decremented by 1 in step S82, and the above processing is repeated until x = 0 (step S83).
[0056]
FIG. 12 is a table showing the measured height data of the cream solder portion of FIG. 8 and a part of the first and second derivative data (data near line 45 in FIG. 8). The first column of FIG. 12 is the height data, the second column is the first derivative from the left, the third column is the second derivative from the left, the fourth column, the fifth column is the first derivative from the right, the second derivative, The sixth column is the extracted data. For example, the first-order differential value h′x in the second row is 9.5 / 9.4 = 1.01 (step S52 in FIG. 10). The secondary differential h ″ x is the primary differential ratio of 1.01 / 0.99 = 1.02 (step S59 in FIG. 10). Looking at the row of the height data 19 in FIG. 81, the second derivative is 1.72, which is equal to or greater than the threshold value 1.25 (steps S54 and S60) Therefore, the height data 10.5 of the previous row is extracted (six columns in FIG. 12). Exemplified in the eye).
[0057]
The height data where the differential value becomes large is stored in the height correction memory in steps S55 and S61 in FIG. 10 or steps S75 and S81 in FIG. 11, and this is the average value hAV in step S84 in FIG. As required. In the image data shown in FIG. 8, this average value was 9.33. Since this average value corresponds to the height data of the flat portion (resist surface) on which the cream solder is printed, the height data (hAV) of the flat portion is subtracted from the height data (hx) in step S85. The actual height of the cream solder from the flat portion (average height is 102 μm) is calculated. This is illustrated in FIG. Since the average height of the cream solder measured with another measuring instrument in advance was 102 μm, it was confirmed that the height of the cream solder by the above treatment was accurate.
[0058]
In the above process, the first and second derivative values are used. However, the height data of the flat portion can be extracted only by the first derivative. Alternatively, height data may be extracted using only one of the left and right differential values.
[0059]
[Other embodiments]
With the above-described processing alone, the differential value becomes large even at a portion other than the rising portion of the cream solder, and the height data of the flat portion may be extracted. In some cases, the height may be extracted and the height may be extracted, and when the substrate is warped, the light-cut image may be tilted and captured, and the height may not be calculated correctly. Therefore, these problems are solved by the following processing.
[0060]
The first problem is solved by examining the flatness before and after the height data when the height data is extracted, not adopting the height if it is flat, and adopting it if it is not flat. . The second problem is that when adopting the attentional height, it is checked whether or not the height data has already been extracted before several times. If not, the current attentional height data is updated. If it is adopted, the current attention height data is abandoned. For this purpose, the processing shown in FIGS. 14A and 14B is provided between step S54 (S60) and step S56 (S62) in FIG.
[0061]
In step S90 of FIG. 14A, when the primary differential value or the secondary differential value is equal to or greater than the threshold value and the height data is extracted as the height data of the flat portion in step S54 (S60), the flatness is flat in the vicinity. Degree F '
[0062]
[Equation 3]

[0063]
(Note that h 'is used instead of h for the second derivative). Subsequently, in step S91, it is determined whether this flatness is larger than the threshold value (2). If it is larger, the extracted attention height data is adopted, and otherwise, it is not adopted. When actually calculating with the data in the table of FIG. 12, the extracted height data is 10.5, so the flatness F ′ according to Equation 3 is F ′ = (| 9.5-10 | + | 10−19 | + | 19−20 | + | 20−9.5 |) /4=5.25> 2 (threshold value). Therefore, it is adopted as the attention height data and stored in the height correction memory in step S55 (S61).
[0064]
The second problem can be solved by providing the process of step S92 shown in FIG. 14B between step S91 and step S55 (S61). In this step, for example, it is checked whether or not the height data is adopted up to five previous times, that is, from (x-2) to (X-6). If not adopted, it is adopted as attention height data, and is stored in the height correction memory in step S55 (S61).
[0065]
FIG. 15 shows a process corresponding to the case where FIG. 14 is differentiated from the left side with respect to the case where it is differentiated from the right side.
[0066]
[Expression 4]

[0067]
14 is different from FIG. 14 in that it is checked in step S96 whether height data is adopted from (x + 2) to (X + 6) in order to solve the second problem.
[0068]
To solve the third problem, the height data extracted and adopted by the differential processing and the first and second problem solving means is calculated by multiple linear regression, an approximate straight line is obtained, and the inclination is corrected. . For this purpose, as shown in FIG. 16, after step S83 in FIG. 11, the number n of hx in the height correction memory, that is, the number of extracted height data is checked (step S100). In step S101, hx = y in the height correction memory is set.
[0069]
[Equation 5]

[0070]
Next, in step S102
[0071]
[Formula 6]

[0072]
Thus, a0 and a1 are obtained, and an approximate expression a0 + a1x = y is obtained.
[0073]
For example, if the extracted target height data is (62,9) (151,10) (184,10.5) (238,11.5) (362,12) (449,13), n = 6 From Σx = 1446, Σy = 66, Σx ^ 2 (Σ of square of x) = 449790, Σxy = 16918, a0 = 8.592474 and a1 = 0.00999 are calculated. Therefore, the approximate expression is: 8.592474 + 0.00999x = y.
[0074]
Next, in steps S103 to S106, x in the approximate expression is obtained as a value of y from 1 to the number of height data (i + 1), and hx is corrected in step S104. Now, assuming that the number of height data is 480, if the difference between the 480 height data and 480 y is obtained, this becomes 480 height data with the inclination corrected. 17A and 17B show the calculation results.
[0075]
FIG. 18 shows an example in which measurement is performed on a substrate having a warp and zero correction is performed without correcting inclination (step S85 in FIG. 11), and it can be seen that the height data graph rises to the right. If the tilt correction is performed along the processing in FIG. 16, the same result as in FIG. 13 is obtained.
[0076]
[Data processing by spreadsheet software]
In the embodiment described above, the image data in the V-RAM image memory 40 has been subjected to image processing with the circuit configuration shown in FIG. 4, but the image data in the V-RAM image memory 40 can also be imported into spreadsheet software. . Since the image data in the V-RAM image memory 40 is a bitmap image of 640 pixels wide × 480 pixels high, after each pixel is converted to gradation data of 256 gradations, a cell of 640 columns × 480 rows is obtained. Into the spreadsheet software. Since the resolution between the pixels is 10 μm, the pixel corresponds to the cell when read into the spreadsheet software, and the pitch between the cells is 10 μm. However, since the actual spreadsheet software has a functional restriction that the maximum number of columns is 256 columns, processing is performed by fetching gradation data of 200 columns × 480 rows.
[0077]
First, the acquired gradation data is extracted for each 3 × 3 cell, and the filter processing is performed by checking the flatness of the gradation. The center cell of the 3 × 3 cells is set as the target cell, and the flatness of the gradation around the cell is calculated. An average value of absolute values of differences between cells around the target cell is obtained. .., The rows are 1, 2, 3,..., For example, the target cell is B2, and the flatness F is calculated according to Equation 1 (by the arithmetic circuit 63 in FIG. 4). Corresponding to calculation). Next, a ratio to the difference ΔB between the maximum gradation value and the minimum gradation value in the 3 × 3 cell is obtained (corresponding to the arithmetic circuit 65), and when F / ΔB is smaller than the threshold value (comparator 66). 2), the band elimination filter (corresponding to the filter circuit 68) of Formula 2 is applied to the target cell. If ΔB = 0 or F / ΔB is larger than the threshold value, nothing is done. Next, the target cell is moved to B3, the above processing is executed, and B4, B5,... Are processed. Then, the process proceeds to the next line, and processing is sequentially performed as C2, C3.
[0078]
When the gradation data filtering process is completed as described above, the process proceeds to the next binarization process. The average value and standard deviation of the gradation of each row are calculated (corresponding to the arithmetic circuits 71 and 72). For each cell, when the gradation data of the cell is larger than the average value of the gradation of the row + 1.5 × standard deviation (corresponding to the comparator 75), the value of the current cell is rewritten to 1, and at the following times Erases the value of the current cell. The average value and standard deviation of the first row are used for the cells A1, B1, C1,..., And the average value and standard deviation of the second row are used for the cells A2, B2, C2,. Each cell is 1 or blank.
[0079]
After the values of the respective cells are binarized in this way, the small protrusion data and the isolated data are considered to be noise. Therefore, these data are erased to remove noise. That is, the filter processes a to f as shown in FIG. 7 are performed. With the 3 × 3 center cell as the target cell, when the pattern of FIG. 7 appears, the value of the target cell is erased. The five types of filters other than f are rotated by 90 degrees (corresponding to the noise removal processing circuit 83).
[0080]
Next, reference is made to each row, and if all are blank, reference is made to an array of cells with a numerical value of 1 in the upper and lower rows, and hole filling is performed (corresponding to the determination circuit 87). Next, a centroid value is calculated for a cell of value 1 in each row. As shown in FIG. 19, for columns A, B, C..., This number becomes the value of the center of gravity when serial numbers 1, 2, 3,. Corresponding to circuit 96). In this example, for the first and second rows, the cell values in the I and J columns are 1, the I column number is 9, and the J column number is 10. Therefore, the centroid values in the first and second rows are 9. 5 For the third, fourth, and fifth rows, the cells in the I, H, and G columns are 1, and the center-of-gravity value of each row is 9, 8, and 7, which is the same value as the number assigned to each column. In the same manner, the center of gravity value of each row is obtained.
[0081]
Thus, each processing block 60, 70, 80, 90 of FIG. 4 can also be processed by software.
[0082]
Further, the flatness calculation in step S90 in FIG. 14 is calculated by F ′ = (| H1−H2 | + | H2−H4 | + | H4− in a table calculation when, for example, a cell H3 is extracted as the target height. Calculate as H5 | + | H5-H1 |) / 4. If the threshold value is S, the attention height data extracted when F ′> S is adopted, and the others are not adopted. In the example, S = 2. Other processes for solving the second or third problem can be similarly performed using a spreadsheet.
[0083]
【The invention's effect】
As described above, in the present invention, the height data of the protrusion is obtained from the imaged image of the protrusion, the height data is subjected to the primary differentiation and the secondary differentiation or the primary differentiation, and the flat value is determined from the differential value. Since the height data is obtained, it is not necessary to obtain the height data of the flat part before the protrusion is formed, and the height of the protrusion must be measured every time the protrusion is formed on the flat part. In such a case, the tact time of measurement can be remarkably improved.
[0084]
Further, in the height data processing according to the present invention, since the gradation data of the imaged pixel is flattened and then binarized, stable height data can be obtained. If binary data is filtered and a blank line is generated as a result, data is embedded in the blank line according to the binarized data array in the preceding and subsequent lines. A certain height data is required.
[0085]
Further, in the present invention, even if the differential value becomes large except at the rising portion of the protruding portion, the flatness in the vicinity of the extracted height data is checked, and if it is not flatter than the predetermined value, the extracted height data is Therefore, reliable height data can be obtained.
[0086]
Also, even if the differential value increases during the rise of the protruding part, if the height data of the flat part is already adopted from the plurality of previous height data, the extracted height By discarding the length data, reliable height data is similarly guaranteed.
[0087]
Furthermore, when there is a warp in the direction along the line light of the object to be measured, the object is flattened using multiple linear regression based on a plurality of height data adopted as the height data of the flat part. Since the approximate expression of the height data of the part is obtained and the inclination correction in the direction along the line light of the flat part is performed from the approximate expression, it is corrected even if the substrate on which the protrusion is formed has an inclination. Accurate height measurement.
[Brief description of the drawings]
FIG. 1 is a perspective view showing a configuration of a height measuring device used in the present invention.
FIG. 2 is a side view of the height measuring device of FIG.
FIG. 3 is a circuit diagram showing a circuit configuration for acquiring image data from an image obtained by projecting line light.
FIG. 4 is a circuit diagram showing a circuit configuration for processing acquired image data.
FIG. 5 is a flowchart showing a flow of processing of an image obtained when line light is projected onto a cream solder portion.
FIG. 6 is an explanatory diagram showing an image obtained when line light is projected onto a cream solder portion.
FIG. 7 is an explanatory diagram showing filter data used for filter processing of image data.
FIG. 8 is a diagram showing height data of cream solder.
9 is a diagram showing first and second derivatives of cream solder height data in FIG. 8; FIG.
FIG. 10 is a flowchart showing a flow of differentiating height data from the left side to obtain a differential value.
FIG. 11 is a flowchart showing a flow of differentiating height data from the right side to obtain a differential value.
FIG. 12 is a table showing a process of obtaining height data of the flat portion of the wiring board from the differential value.
FIG. 13 is a diagram showing cream solder height data measured from the flat portion of the wiring board obtained based on the differential value.
FIG. 14 is a flowchart showing a process for removing unreliable data extracted at a portion other than the rising portion of the cream solder.
FIG. 15 is a flowchart showing processing when differentiation is performed from the reverse side in the same processing as in FIG. 14;
FIG. 16 is a flowchart showing a flow for correcting warpage of a substrate.
FIG. 17 is a table showing processing data when a substrate is warped.
FIG. 18 is a diagram showing height data when a substrate has a warp.
FIG. 19 is an explanatory diagram showing an example for obtaining a centroid value from pixel or cell information.
FIG. 20 is a perspective view showing a configuration of a conventional three-dimensional measuring apparatus.
[Explanation of symbols]
1 Laser light source
5 Line generator
6 CCD camera
9 line light

Claims (6)

Line light is projected onto the object to be measured that has a protrusion protruding from the flat part, and the height of the protrusion from the flat part is measured by the light cutting method by imaging the flat part and the protrusion cut by the line light. In the height measurement method to
Find the height data of the protrusion from the image of the captured protrusion,
First and second derivative or first derivative of the height data,
Extract the previous height data when the differential value exceeds the threshold,
A flatness measurement in the vicinity of the extracted height data is examined. If the flatness is not flatter than a predetermined level, the extracted height data is used as the height data of the flat portion . When the height data is extracted, if the height data of the flat part is already adopted from the plurality of previous height data, the extracted height data is discarded. The height measuring method according to claim 1 . Based on a plurality of height data adopted as the height data of the flat portion, an approximate expression of the height data of the flat portion is obtained using multiple linear regression, and along the line light of the flat portion from the approximate expression The height measuring method according to claim 1 or 2 , wherein a direction inclination correction is performed. Line light is projected onto the object to be measured that has a protrusion protruding from the flat part, and the height of the protrusion from the flat part is measured by the light cutting method by imaging the flat part and the protrusion cut by the line light. In the height measuring device to
A light projecting device that projects line light onto the protrusion;
Means for picking up an image of the projection projected with the line light;
Means for processing the image data of the protrusion by the line light to calculate the height data of the protrusion;
Means for first and second derivative or first derivative of the height data;
Means for calculating the height of the flat portion from the differential value;
Have a, means for calculating by the height of the projections relative to the height of the flat portion,
The means for calculating the height of the flat part extracts the previous height data when the differential value exceeds a threshold value, checks the flatness near the extracted height data, and is flatter than a predetermined value. Otherwise, the extracted height data is used as the height data of the flat portion . The means for calculating the height of the flat portion is extracted when the height data of the flat portion is already adopted from the plurality of height data before the height data is extracted. The height measurement apparatus according to claim 4 , wherein the height data is discarded. Based on a plurality of height data adopted as the height data of the flat portion, an approximate expression of the height data of the flat portion is obtained using multiple linear regression, and along the line light of the flat portion from the approximate expression 6. The height measuring apparatus according to claim 4, further comprising correction means for correcting a direction inclination.

Images