Research On 3d Obstacle Detection System Based On Binocular Vision

Mobile welding robot can replace manual welding task in the harsh environment,but the traditional robot can only complete a relatively simple welding task,and can not completely replace manual welding,so it needs to be improved.The robot can accurately obtain the surrounding environment information is a prerequisite for the mobile welding robot to achieve intelligence,and the information obtained can be used as a basis for judging the follow-up movement of the robot.Aiming at how to detect the obstacles near the robot and obtain accurate three-dimensional information,this paper studies a detection scheme of obstacle distance and three-dimensional information based on binocular vision,and the feasibility of the scheme is verified by experiments.In this paper,an obstacle detection system is designed by analyzing the specific requirements of the existing mobile welding robot.Firstly,the image acquisition system is built,then the imaging principle of the binocular camera and the mathematical transformation formula of the coordinate system are introduced and deduced,and the stereo calibration and distortion correction of the camera are completed according to the binocular calibration experiment.Then,through the comparative experiment of feature point extraction,the limitations of the existing stereo matching algorithms are analyzed,and an improved FP＿SIFT algorithm is proposed.The algorithm extracts the outline of the obstacle through image enhancement technology and morphological processing,re-divides the pixel region in the feature description generation stage to reduce the dimension of the feature descriptor,uses the above methods to optimize the flow of the traditional SIFT algorithm,and compares the performance with the commonly used SIFT and SURF algorithms through stereo matching experiments.The experimental results show that the FP＿SIFT algorithm proposed in this paper has higher accuracy of feature point matching,faster matching speed and better robustness than SIFT algorithm.Finally,the disparity map is generated on the basis of the binocular stereo matching experiment,and the obstacle and background information are separated through the inverse relationship between the binocular parallax and the actual distance of the obstacle.The contour of the obstacle is separated and the feature points of the upper surface are screened by K-means clustering analysis algorithm,and the vertex coordinates of the upper surface are extracted and solved to get the distance and three-dimensional information of the obstacle.In this paper,the obstacle is arranged manually to simulate the working environment of the robot,and then the obstacle detection experiment is carried out.the experimental results show that the detection method proposed in this paper can solve the obstacle distance and three-dimensional information,and can meet the requirements of the project.it can provide a basis for the follow-up research work of autonomous obstacle avoidance of mobile welding robot.