| With the development of machine vision, the three dimensional reconstruction technologies of object are increasingly important. In recent years, binocular stereo vision has developed rapidly, high efficiency, high accuracy system, and low cost are its characteristics, very suitable for online non-contact, product testing and control in manufacturing site. So3D reconstruction technologies based on binocular vision has been one of the central points of machine vision field nowadays, and feature point detection and matching technology of3D reconstruction technique is the key and difficult point, the study is very necessary. With the development of robot technology, its application in manufacturing industry has become more and more mature, robot has many advantages, such as it can work for24hours, won’t influence products’quality because of fatigue, also it is stable and precise.3D reconstruction technology developments did the foreshadowing for robot glue automation.This paper mainly did research on following aspects:feature extraction algorithm, stereo matching algorithm, calibration of binocular camera, and generation robot glue path’s initial data. Innovation points are shown as followed.(1) An improved adaptive Chord-to-Point distance accumulation algorithm or ACPDA for short has been proposed which can handle with adjacent corner, obtuse corner and round corner. The main advantages of our implementation include:firstly, the neighborhood of the candidate corner is re-detected for avoiding loss of the adjacent corner which makes the localization of corners more precisely. Secondly, an adaptive threshold for each curve is set for removing the false corner for avoiding loss of the obtuse corner. Finally, a local adaptive threshold is constructed to remove the round corners effectively. The experimental results have evaluated the performance of ACPDA.(2) For solving the incorrect stereo matching of edge points, a stereo matching algorithm based on constraint of corner distance has been proposed. Firstly, ascertain the candidate corner matching set through the epipolar constraint and the corner’s feature value constraint. Secondly, put forward the "edge correlation" constraint, calculate the contribution value of candidate corner based on corner distance, and fulfill the precise matching by relaxation method. Then constructs featured vector of the corner, checkout its corner matching further. Finally, guide the edge points matching by corner matching. The experimental results show that this matching algorithm has high accurate rate, and solve the wrong matching problem of edge points effectively, it’s quite suitable for the applications of edge-based stereo.(3) Though analysis of binocular camera model, doing calibration and correction, getting binocular camera internal and external parameters, and combined with parallax of corresponding points to get glue trajectory’s3D coordinates, we can get glue path’s initial data then convey data to robot, and it will analog the edge trajectory gluing operation. |
