Research On Key Techniques Of Fiducial Marks Locating Vision System For Placement Machine

Vision system is one of the most important key parts of the placement machine which could influence the performance of placement machine directly. The fiducial marks locating subsystem plays an important role in initializing, calibrating the whole vision system and compensating the machining error, thus the performance of the whole vision system depends on the locating accuracy and speed of this subsystem largely. Based on the developing work of placement machine in research group, this paper did some research on the fiducial marks loacting subsystem. The main results were showed as following:The paper chose circular, rectangle, rhombus, triangle, cruciform, double rectangular, butterfly and sharp as the target fiducial marks based on the demond of the placement machine. By analying the characteritics of these fiducial marks, we found the way to obtain the edge pixels. The OTSU algorithm was used as image binarization and the contour detection based on morphology was used to determine the ROI(Region Of Interest) of the fiducial marks. The Canny operator was used to detect the edge of the fiducial marks. Adaptive selection methods for the threshold of Canny operator was proposed based on the image features and the method showed excellent robustness to light change.The detection algorithm based on the edge features was proposed. The lines and circles that consist of the edge pixels detected by the Canny operator were extracted by Hough Transform method. The Progressive Probabilistic Hough Transform method was chosen after the analysis of the performance the different Hough Transform methods and the transformation parameters sel ection method was also proposed. The improved hierarchical clustering method was proposed to merger the repeated lines that detected by Hough Transform. The edge equation and the center of the fiducial marks were obtained by regression fit. The algorithm showed that better robustness for general isolated interference and noise and the E-time was less than 10 ms, while the algorithm may not suitable for these fiducial marks whose edge is occluded partly.The detection algorithm based on the fiducial mark model was proposed. The parameters of each mark were analyzed and all the models for each mark were built. The matching criterion based on the Mean Hausdorff Distance was proposed to characterize the similar degree between the target to be detected and the tem plate. The computing method chose distance transform for matching similarity, which could solve the problem of demonding calculate target and Mean Hausdorff Distance repeatedly and decrease the computation time effectively. The Particle Swarm Optimization algorithm and the image pyramid structure were used to speed up the matching process. This method showed excellent detection effect for these partly incompleted and occlusion fiducial marks and better robustness for diferent isolatioin. The execution time was around 100 ms.The robustness to the salt and pepper noise and the maximum repeatability error of the two algorithms were tested from the experiments. Many repeated trials were carried out to analynize the detection precision, the maximum repeatability error and the execution time of the two algorithms under the same test condition. The results showed that the maximum repeatability error of both algorithms were less than 15μm which met the requirements of the system and the execution time is also the same.