Research On Obstacle Detection Of Mobile Welding Robot Based On Binocular Vision

With the continuous progress of science and technology,the welding industry is developing towards intelligence and automation,and welding robots have been widely used in industrial production and other fields.The mobile welding robot needs to identify the surrounding obstacles in the process of traveling and operating in the welding environment to provide the basis for the next motion control of the mobile welding robot.This paper develops a vision sensing system on the self-developed mobile welding robot platform,establishes a binocular vision-based obstacle detection and localization scheme,and verifies the feasibility of the scheme through experiments.The main research contents are as follows.(1)This paper built a binocular vision system,designed the camera clamping mechanism and 3D head,and designed the overall scheme based on the analysis of the workflow of the system implementation.(2)The imaging principle of binocular vision is studied,and the limitations of the traditional camera calibration method are analyzed,and a camera calibration scheme with higher accuracy is designed.Firstly,the commonality between neural network and camera calibration is analyzed,and BP neural network is applied to camera calibration.In order to solve the problems of slow convergence and easy to fall into local optimum of BP neural network,a camera calibration method based on BAS-BP algorithm is proposed,which is compared with the traditional Zhang's calibration method,and the feasibility and robustness of the scheme are verified through experiments.(3)The images acquired by the binocular camera are pre-processed with image enhancement,filtering and denoising,equalization,etc.The SIFT algorithm and SURF algorithm are used to extract feature points from the images,and they are applied to stereo matching,and a stereo matching method based on feature points is proposed.(4)Using the results after binocular stereo matching,a binocular parallax map is generated,an obstacle detection method based on the parallax map is studied,and the depth information of the obstacle is solved by using the parallax of parallel binocular vision according to the relationship between parallax and depth information.At the same time,the location of obstacles is realized by calculating the height and horizontal distance of the points with the same name,and propose a method based on K-means cluster analysis to eliminate the mis-matching points,and then complete the detection of obstacles and the solution of 3D information.The experimental results show that the method of obstacle detection and its 3D geometric information reconstruction based on binocular vision introduced in this paper has good feasibility and meets the experimental requirements,which lays the foundation for the subsequent work of mobile welding robot such as obstacle avoidance.