視覺輔助型鋼橫向剖面輪廓線上量測系統

An On-line Vision-Based Measurement of the Cross-Sectional Contour of H-Shaped Steel

楊雨霖(Un-Lin Yong)

指導教授 ： 林穀欽

崑山科技大學/工程學院/機械工程研究所/碩士(2006年)

https://doi.org/10.6828/KSU.2006.00012

摘要

我國雖然有生產各種型鋼的能力，但型鋼之尺寸量測大多仰賴人力進行離線量測，當生產過程中型鋼製造精度出現問題時，待工作人員發現再停止生產線停止時，往往已經造成不小的損失，本系統研發目的即是發展一視覺輔助型鋼橫向剖面輪廓線上量測系統，期以達到線上檢測型鋼生產尺寸，自動修改工作設定或適時地警告操作員中斷生產流程。
本研究系統之硬體設備包括8 台CCD 攝影機、4 具線性雷射光發射器、伺服控制元件、執行影像擷取、分析及控制軟體之電腦整合單元，本系統控制雷射光投射主動光源，以產生H 型鋼輪廓外型之特徵影像，另由電腦控制H 型鋼輪廓外形影像之萃取。
本論文引用Zhang[Zhang, 1998]校正法以估算攝影機之參數，本研究建立一攝影機校正用之立體效正器，各面之校正點座標位置關係為已知，再以Epipolar Geometry以尋找特徵點對應之投影座標，進而計算特徵點之世界座標，藉由三維空間座標轉換及疊合重建H型鋼之橫向輪廓，以判斷型鋼尺寸精度良窳，達到線上檢測的目的。
實驗結果證明本型鋼尺寸量測系統誤差百分比約在2%以下，較一般型鋼製造業者要求精度3%為低，因此本量測技術可滿足業界需求，未來期望本技術能實際應用於鋼廠生產線自動化作業之檢測。

關鍵字

攝影機模型 ； 立體視覺 ； 非接觸式量測 ； 攝影機校正

並列摘要

Although manufacturers are able to produce various kinds of H-shaped steel in Taiwan, they conduct the task of checking the steel cross-sectional size off line and by manpower. Significant economic loss could arise if the
production precision of the steel is out of control.
The system built for this research include 8 CCD cameras, 4 line laser projectors, a servo controller, an image capturer, and computer softwares.
The projectors shed line laser light on H-shaped steel to generate characteristic image for measurement.
The camera models and calibration algorithm proposed by Zhang [Zhang, 1998] are utilized. A block with control points on its surface is
designed for calibrating the cameras. The control points are the points with known coordinates. The study implements epipolar geometry to find the image coordinates of the control points between two correspondent images
to determine the intrinsic and extrinsic camerca parameters.
The cross-sectional size of H-shaped steel is measured by determing the world coordinates of the laser light on the steel surface. The
correspondence of laser images is determined based on the epipolar geometry, using Zhang′s models and the camera parameters achieved in the
previous calibration process.
Experimental results prove that the measurement accuracy using the proposed approach is less than 2%. The accuracy is acceptable for the
industrial need. We expect to apply the proposed approach to a real line production in the future.

並列關鍵字

camera model ； stereo vision ； non-contact measurement ； camera calibration.

參考文獻

[27] 林建昌，“視覺輔助H型鋼橫向尺寸量測系統之研發”，崑山科技大學機械研究所碩士論文， 2004年。

[33] 陳彥良，“即時立體視覺物體追蹤系統”，中原大學機械工程研究所碩士論文，2003。

[1]A-Nasser Ansari, “Automatic facial feature extraction and 3D face modeling using two orthogonal views with application to 3D face recognition ,” University of Miami, Department of Electrical and Computer Engineering,, FL 33124, USA 26 September 2003。

[2] Banard, S.T. and W.B. Thompson , “Disparity analysis of images”, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 2, No.4, July 1980, pp.333-340.

[3] Boykov, Y., O. Veksler, and R. Zabih , “Disparity component matching for visual correspondence”, Proc. IEEE Conf. Computer Vision and Pattern Recognition. 1997.