Obstacle detection of indoor mobile robot based on binocular vision

Abstract

Aiming at the uncertainty of obstacles in indoor mobile robot operation environment, this paper designs a binocular vision obstacle detection system that does not depend on the a priori knowledge of obstacles and background. Specifically, we build a binocular vision system, calibrate the camera according to the binocular vision theory, and obtain the camera calibration parameters. The binocular image is corrected by calibration parameters and epipolar constraints to solve the non coplanar problem of image distortion. SGM (semi global matching) algorithm is used to obtain disparity map to ensure the rapidity and robustness of disparity acquisition. Finally, an obstacle detection algorithm is proposed. The ground detection model was established to constrain detection distance, introduce area threshold, filter interference, extract obstacles, and measure obstacles. Experimental results show that the system can effectively detect and measure obstacles ahead, and the relative error within 4m is 2.56%, which meets the requirements of indoor mobile robot obstacle detection.