以影像處理為基礎的三次元量床自動量測策略設計

The Automatic Measuring Strategy Design for CMM Based on Image Processing

陳國源(Kuo-Yuan Chen)

指導教授 ： 戴兢志

大同大學/工程學院/機械工程學系所/碩士(2005年)

摘要

三次元量床是最常見的量測設備，通常是將測頭以固定間距對物體表面進行量測。但此量測策略會造成在物體表面平坦處產生多餘的資料點，而在曲面處卻缺少關鍵資料點。而且測頭通常無法沿著量測物之法線方向量測，因此曲面重建時，所有資料必須進行補正處理。
　　本研究擬以倒傳遞類神經網路替代整合光度立體攝影技術之照度演算法來規劃三次元量床之量測策略，光度立體攝影技術之量測策略主要是根據光源照射在待測物體上的反射光強度、待測物體之材料、表面性質及待測物形狀，運用三組由不同位置之光源所得之照度方程式，以數學演算法來求得量測點之法線向量，因為方程式上各係數多且並無標準值，在此以倒傳遞類神經網路來替代找出法線向量，以各量測點的法線向量來規劃量測點，則可以有效地縮短量測時間及減少曲面重建所需的資料量。倒傳遞類神經網路取代其數學演算法，以內在複雜的權值來連接所輸入灰階值與輸出法線向量間的關係，如此一來能夠解決利用數值方法求解數學方程式不一定能夠收斂的困擾，也可以更加縮短量測上的時間。而且各量測點的法線向量可以直接運用在量測時測頭的轉向，如此對於量測後的資料不須再做測頭補正處理。

關鍵字

倒傳遞類神經網路 ； 光度立體攝影技術 ； 影像處理

並列摘要

CMM is the general measuring equipment. The user can measure the surface of the object by using the probe in a fixed distance and this measuring strategy is the general measuring one. But, the surplus data point in the surface of the object will be caused with this method and the point points are not enough in the surface. Both of these is not good for the processing of the data and the re-establishment of the surfaces.
　　In this research the BPN is used to replace the integrated illumination with photometric stereo technique to plan the measuring strategy with the CMM. According the strength of the reflective light, the measuring materials, characters of the surface and shape of the measuring object, the measuring strategy of the photometric stereo technique is the illumination model from the three different positions of the lightsource and the normal vector of the measuring point is obtained by using mathematic algorithm. Because there are many parameters in the equation and no standard value, BPN is used to find out the normal vector and the normal vector of every measuring points are used to place the measuring points so that the measuring time and the number of the surface-reestablished can be reduced effectively. Instead of the mathematic algorithm, BPN connect the relation of the input gray value and the output normal vectors by the complicated weighted value so that the question that using numerical analysis used to solve the mathematic equation does not always can be solved convergence and the measuring time can also be reduced. Furthermore, the normal vectors of every measuring point can be used directly in the direction of the probe when measuring so that the measured data does not have to be enhanced.

並列關鍵字

Back-propagation neural network ； Image Processing ； photometric stereo technique

參考文獻

3. C. Menq and F. L. Chen, “Curve and Surface Approximation From CMM Measurement Data”, Computers. Ind. Engng., Vol.30, No.2, pp. 211-255, 1996.

4. C. K. Song and S. W. Kim, “Reverse Engineering : Autonomous Digitization of Free-form Surface on A CNC Coordinate Measuring Machine”, Int. J. Math. Tools Manufact, Vol.37, No.7, pp. 1041-1051, December, 1997.

6. M. Zerrog, R. Nevatia, “Part-based 3D Descriptions of Complex Objects from a single image, IEEE Transactions on Pattern Analysis and Machine Intelligence Vol.21 No.9 pp. 835-848, 1999.

7. Q. J. Peng, M. Loftus, “A new approach to reverse engineering based on vision information”, International Journal of Machine Tools & Manufacture 38 pp. 881-899, 1998.

8. Q. J. Peng, M. Loftus, “Using image processing based on neural networks in reverse engineering”, International Journal of Machine Tools & Manufacture 41 pp. 625-640, 2001.