Research On Obstacle Detection And Avoidance Of Binocular Intelligent Vehicle

Binocular vision obstacle avoidance technology refers to a detection technology of the obstacles in the environment based on the binocular vision technology,which can obtain its position and avoid obstacles in accordance with the acquired position.This technology which has various advantages,such as fast information collection,rich perception in environment informationan and low cost,can be widely applied in various fields,like mobile robots and unmanned aerial vehicles.In this technology,how to accurately detect the foreground(obstacle)from the background is one of the key research objects.In the obstacle detection process,affected by the ground,there will be small areas of noise in the disparity map,and it is difficult to completely remove these noises while retaining the disparity information of the obstacles based on general image binarization processing,which increasing the difficulty to detect obstacles.In the obstacle depth detection process,the obstacle depth information obtained is inaccurate due to the influence of invalid points in the disparity map and detection abnormal points.The above-mentioned key issues are related to the feasibility and stability of binocular vision obstacle avoidance technology,so further research is needed.Based on the Raspberry Pi 4B intelligent vehicle and binocular stereo matching technology,the obstacle detection and obstacle avoidance control has been investigated in this thesis,the main research contents were as follows:1)The disparity map was obtained through camera calibration,left and right images stereo correction,distortion correction and SGBM stereo matching with this method;2)Hole filling,binarization,morphological corrosion operation processing,maximum contour detection,and minimum contour area constraints were performed on the disparity map.Obstacle parallax segmentation was performed based on the contour which met the constraint conditions;3)Invalid points in the parallax segmentation point were removed,and the median outlier removal method was proposed to eliminate the detection points with large deviations in the detection results to obtain accurate obstacle position information;4)Used three frames difference method to judge the motion state of the obstacle;5)The obstacle avoidance control was performed based on the obstacle position information and the motion state,and experimental tests were carried out for each functional module of the proposed algorithm.Experimental test results showed that the obstacle detection and obstacle avoidance algorithm proposed in this thesis effectively reduced the difficulty of obstacle detection and improved the accuracy and stability of obstacle depth detection.Through this algorithm,the Raspberry Pi intelligent vehicle can autonomously complete video collection and processing,obstacle detection and analysis,and obstacle avoidance decision-making and execution,which has certain reference value for the application of binocular vision technology in the field of intelligent vehicle obstacle avoidance.