An Obstacle 3D Location Method Of Mobile Welding Robot Based On Binocular Vision

In recent years, the popularity of computer technology and artificial intelligence lead a great change in the development of modern processing and manufacturing industries. Therefore, welding robot with necessary mobility and flexibility are produced and widely used in aerospace, shipbuilding, automobile manufacturing and other industries. With the development of industrial, there are more and more application of large welded structure, which a lot of welding work need on-site operations, so the requirements on welding robot's auto ability are increasing. One of the most difficulties on mobile robot is obstacle problems during welding operation. That is, the robot should sense the environment automatically without human control, analysis obstacle information and make some response, eventually complete the task independently. Therefore, the research has important significance and broad application prospects.The purpose of this project is, by matching images from binocular camera in front of the robot and 3D calculation, to recognize the environment in the front, return obstacle's height and horizontal distance if an obstacle is found. Based on existing theory and results, the author makes a comprehensive discussion and verification on three-dimensional localization theory and technology, establishes and improves a set of obstacles stereo localization algorithm framework based on the principle of binocular vision.The major work in author's research includes the following aspects:1. Extract feature points and match images from binocular camera based on SIFT algorithm, then remove wrong matches in the method of NCC relevance constrain, and finally get stable match points.2. Calibrate the inside and outside parameters in the method of Zhang Zheng-you plane template calibration, and analysis the accuracy with calibration experiments.3. Calculate the robot operation ground plane equation and corresponding point's 3D coordinate based on calibration data and parallel binocular vision parallax theory. By calculate the height of corresponding points to the ground plane to determine the existence of obstacle.4. For pathological value in the height data caused by wrong matches not fully filtered, will removed by a Maximum-Minimum Distance clustering method, then calculate obstacle's height and horizontal distance through corresponding points on the obstacle's surface.Experimental results show that, the obstacle 3D localization method based on binocular vision has good feasibility. Simultaneously, the method described in the article is a relatively general algorithm, and the work provides basis for further research on welding automation and mobile robot applications.