Three-dimensional reconstruction method based on target recognition for binocular humanoid robot

Abstract

According to the problems of the too many man-made specified parameters, huge computation and lower accuracy in complex target recognition and three-dimensional reconstruction of the target of the binocular humanoid robot, we put forward a new binocular robot target recognition and three-dimensional reconstruction method. This article is based on the basic principle of binocular vision. At first, binocular camera calibration uses the checkerboard calibration method; secondly, capturing images by binocular camera; thirdly, using Bayesian classifier to identify the target recognition; then, using improved SIFT algorithm to achieve feature point matching; furthermore, calculate the distance between a humanoid robot and the target based on the principle of triangulation; at last, build the target according to characteristics of the target by using three-dimensional reconstruction. The experiments demonstrate that this method reduce man-made specified parameters, and improve the efficiency of feature points matching, and the accuracy of the binocular distance measurement is higher, this method can effectively and reliably reach the goal of the three-dimensional reconstruction of the target ball.