Research On The Algorithm Of Robot Binocular Vision Obstacle Detection

Binocular vision has become a research branch and a research hotspot of machine vision,especially in the aspect of the combination of robot and vision technology,which has attracted more and more attention.The robot binocular vision recovers depth information from the parallaxes of points in two different pictures in the same space,thereby detecting the obstacles in preparation for subsequent actions.The advantages of binocular vision over other depth measurement methods are its advantages such as passiveness,non-contact,and wide application range.These advantages allow binocular robots to adapt to more difficult and complex environments such as space exploration,robot cruising and battlefield detection.The research work of this paper is based on the binocular stereo vision testing algorithm research under the robot’s background.The research focuses on the feature detection and stereo matching.The stereo matching is the core of this paper’s research.Currently popular stereo matching methods can be divided into two types,local and global.Although binocular stereo vision has achieved a series of research results in recent years,it still needs to be improved and strengthened on some key issues and technical processing.In this paper,a series of studies have been carried out on binocular stereo camera binocular calibration,feature detection,stereo matching,and depth information extraction.The main work and innovations are as follows:(1)Considering that the ORB feature point detection algorithm displays the ideal effect in the detection of feature points in terms of efficiency and accuracy,this paper selects it as the feature point detection part detection algorithm.However,ORB does not have scale-invariant characteristics.That is,when the proportions of the two pictures are not the same,the mis-matching of feature points will occur.Therefore,this paper combines the feature description sub-section of the ORB with the SIFT feature point detection algorithm,which makes up for this shortcoming of the ORB and makes it scale-invariant.(2)In the stereo matching section,this article addresses the core: matching cost problem and support window problem,proposed a multi-similarity measure fusion stereo matching algorithm..According to the process of stereo matching,firstly,several different feature description information are combined to form a matching cost function;then the cost aggregation is performed through the filtering method;then the parallax correction is performed on the WTA strategy and the LRC problem involved,and finally the ideal disparity map on the stereo matching is obtained.