Research On The Position Recognition Algorithm Of LED Chip Based On Vision

With the capacity and demand of LED expanding, people are increasingly concerned about the development and improvement of LED key equipments. LED die bonder is one of the key equipments in the LED packaging process,and its main function is to complete the positioning,picking,and fixing of the LED chips. The vision system of die bonder is to recognize the position of the chip,and the positioning speed and accuracy directly determines the performance of the die bonder.This thesis has an in-depth study for visual positioning algorithm of the LED chip,and developed a shape-based LED chip visual localization algorithm. The main work includes:(1) In order to improve the speed and accuracy of positioning,coarse and fine two template matching positioning strategy is proposed. Firstly,it builds multi-directional and multi-layered templates making use of the image pyramid. In the coarse positioning process,the similarity evaluation function is equipped with the template and dot product of the edge points on the normalized gray gradient vector with the image,and it matches according to the image pyramid from up to down so that can get pixel-level positions. In the fine positioning process,it is characterized by sub-pixel edge based Facet and initial value is the result of the coarse positioning,using ICP algorithm and K-D tree,positioning accuracy can achieve sub-pixel level.(2) Use projection segmentation method which achieve a fast segmentation of the LED chip image,and make the one-to-many matching problems convert into one-to-one problems between template and the image to be matched.(3) A new seed-based point matching search method is come up and it has effectively improved the coarse matching speed. In addition, during the sub-pixel edge extraction stage, using orthogonal base transform derived convolution template which accelerate the speed of the fine positioning.(4) Develop an LED chip visual positioning software.Positioning experiment is carried out for test and real images. The results of experiment show that both speed and accuracy of the algorithm can reach a appropriate level of foreign commercial software,can be used to partly replace foreign commercial software.