Target accurate positioning based on the point cloud created by stereo vision

Abstract

To solve the problem of workpiece target accurate positioning in industrial environment, we proposed a method that used object's surface point cloud which created by stereo matching of binocular vision. Firstly, found out the ROI(Region Of Interest) of the stereo matching point cloud by template matching, and used an algorithm of noise reduction to gain a clean ROI point cloud. And then, extracted the perfect point cloud of object's surface from object's 3d-model, and used improved iterative closest point algorithm to do point cloud registration that can gain the accurate pose of object. Experiment shows that positioning accuracy is less than 1.5 mm(Euclidean Distance) which can meet the needs of industrial robot doing sorting or accurate grabing.