The Research Of Obstacle Detection Method Based On Binocular Stereo Vision

Binocular stereo vision can both obtain the image and depth information.Compared to others,such as lidar and radar,vision sensors have a wider field of view,more accurate lateral accuracy,and independent work among devices.At present,stereo vision has been widely used in industrial surveying,space exploration,adaptive cruise control and robotic autonomous obstacle avoidance navigation.This paper taken the obstacle detection as the application background,and used the binocular stereo vision as the sensor device to detect the potential obstacles and obtain the distance between the obstacle and the vehicle.This research is of great significance for the development of sensors on self-driving cars.The content of this article is as follows:First,the camera calibration and image correction portion was studied.The camera was calibrated using Zhang Zhengyou's calibration method.The Bouguet method is used to correct stereo image pairs and remove image distortion.Secondly,this paper proposed a stereo matching algorithm based on occlusion information.In the cost aggregation phase,the occlusion area pixel points are assigned a smaller weight;in the disparity optimization stage,selected the little disparity as the parallax of the occlusion area pixel.The accuracy in depth discontinuous regions was improved.Finally,an obstacle detection scheme based on U-V disparity map was designed.In the road surface extraction,the random sampling consistency(RANSAC)combined with principal component analysis(PCA)is used to predict the road paramaters.The histogram counting grid is used to extract the projection of the obstacle in the V disparity map,and the density-based clustering method(DBSCAN)is used in the U-disparity map obstacle extraction.Obstacle detection requires high accuracy and safety in the autonomous driving field.Compared with the less stable method of deep learning,the scheme can easily separate the road surface from obstacles.Potential obstacles above the road surface,and the algorithm has low computational complexity,and has good robustness and real-time performance.This paper proposed the adaptive weight stereo matching method based on occlusion.And designed a U-V parallax map obstacle detection scheme applied to obstacle detection in 3D road driving environment analysis.Detect results meet real-time requirements.