Research On Image Recognition And Image Location Of Micro Assembly System

Due to the small size of micro flat parts and its high assembly precision, traditional manual assembly is difficult to meet market demand. With the rapid development and wide application of machine visual technology, applying it to micro flat parts assemblies can cover the shortage of manual assemblies and greatly improve production efficiency. The difficulty that machine vision is used for assembly is how to develop proper image processing method for system requirements. The variation of parts’ machining quality can make much difference on imaging quality and image processing. This paper proposed an approach to recognizing and positioning assembly features of a fuse, and a method for robot visual location and unloading objects with machine visual technology. The detail is as follows:(1) The geometric model and lens distortion model were built separately. Camera calibrations such as direct linear transformation method, Tsai’s calibration method and Zhang’s plane calibration method were present and their advantages and disadvantages were made analysis of. With the fundamental of Zhang’s method, the camera calibration was accomplished with Matlab toolbox. The image was corrected on visual C++ in combination with image processing library of Open CV.(2) Linear and non-linear filtering method was present systematically, and the comparison of processing results with different filtering algorithm was made to choose proper filtering method. Image segmentation and edge detection were made with typical edge detection operator. A double least-square line fitting method was proposed. This method extracted posture information of assembly parts precisely, which could be used to guide posture alignment.(3) Imaging error caused by manufacturing error of aligning prism was measured experimentally. Mathematical model of imaging error caused by prism’s angle installation error, and imaging error caused by lens’ optical axis not being vertical to working surface was discussed with pin-hole imaging theory. Analysis of edge points’ recognition error and edge line’s extraction error was made based on the principle of image processing algorithm. The model of coaxial alignment error was set up with those errors.(4) A robot auto-locating system with monocular vision was designed. The pixel equivalent was computed. Projective matrix mapping robot coordinate and image coordinate was represented and the location of fixture’s center in image coordinate was determined. According to the result of robot’s autonomous location test, the positioning accuracy could reach 0.6 mm.According to the above research, a method of automatic assembly based on machine vision is obtained. The feasibility of applying the principle of coaxial alignment with machine vision technology to assembly process is verified in this research.