Research On Obstacle Detection Algorithms For Binocular Vision Robots Discipline

In the unmanned material handling workshop,the transport robot needs to detect obstacles in the process of obstacle avoidance.Aiming at the problems existing in the application of obstacle detection algorithms to robot obstacle avoidance,such as high complexity,high cost and insufficient precision.This thesis proposes an obstacle detection algorithm based on binocular stereo vision technology,which can detect obstacles within 1500 mm?4000 mm.The obstacle detection algorithm in this thesis is mainly studied as follows:Acquisition of camera parameters;acquisition of parallax;detection of suspicious obstacles;obstacle detection.(1)In camera parameter acquisition stage,Zhang Zhengyou chessboard calibration method is used,and an improved fast corners detection algorithm is proposed to improve the accuracy of corner detection.The algorithm firstly convolves the checkerboard image with two kinds of corner prototypes and extracts the corner points roughly;Then the non-corner points are eliminated by the "threshold method";Finally,the orientation gradient of corner points is used for subpixel positioning.(2)In the parallax map acquisition stage,SGM semi-global stereo matching algorithm is used,and an optimization method of "normalized gray value+median filter+morphological processing" is proposed to optimize the initial parallax map,then a better parallax map is obtained.(3)In the detection stage of suspicious obstacles,the parallax map of suspicious obstacles is processed with pseudo-color firstly;Then the binary image of the suspicious obstacle is obtained according to the different color components of the suspicious obstacle in the pseudo-color parallax map;Finally,the maximum and minimum coordinates of the non-zero pixels in the binary image are calculated.According to the coordinates,the suspicious obstacles.(4)In the obstacle detection stage,according to the perspective projection model,the projected size of the obstacle area of the carrying robot in the image is firstly calculated,and the interference area of the suspicious obstacle at different distances is obtained;Then the suspicious obstacle and its interference region are interfered to determine whether the suspicious obstacle is an obstacleSince the size of the interference region is inversely proportional to the distance of the suspicious obstacle,the accuracy of the distance measurement is firstly verified in this paper,then a large number of ranging tests were conducted to obtain the original distance data within 1500 mm?4000mm,finally Matlab software is used to process the experimental data.The results show that the maximum ranging error is 34.7 mm and the average ranging error is 23.91 mm within the range of measurement;Then,in the simulated experimental environment and outdoor environment,five groups of suspicious obstacles were collected,and the suspicious obstacles were detected.The suspicious obstacles in the five groups of image pairs were extracted,and the distance information of different suspicious obstacles is obtained;Finally,the interference areas of different suspicious obstacles were calculated for interference discrimination.Experimental results show that the obstacle detection algorithm in this paper has good detection effect and can meet the requirements of the project within 1500 mm?4000 mm.