Road Localization And 3D Map Comstruction Based On Stereo Vision

Simulating human eyes,using computer vision to achieve autonomous navigation of unmanned motion platforms is of great significance.This paper aims at the problem of autonomous navigation of unmanned platforms in the urban structured roads and needs to obtain the road's driving area.Stereo vision system method applied to combine simultaneous positioning with map construction and road positioning methods to project roads into the constructed maps.In order to realize the relative positioning of unmanned platforms and roads,it can better provide navigation information for unmanned sports platforms.The main research contents are as follows:1.In the study of the binocular camera system model,the mathematical connection among the pixel,image,camera,and world coordinate system analyzed,and a correction model created for the distortion generated by the camera perspective imaging according to the imaging principle of the camera.The mathematical model of the binocular vision system was established base on the positional relationship between the two cameras.2.Propose lane detection based on multi-channel threshold fusion and lane-tracking method based on Kalman filter.The image is binarized based on horizontal and vertical edge features of the lane line,saturation and brightness information in the color space,and the lane pixels are extracted accurately using perspective transformation,image histogram statistics and sliding window search,and the polynomial is used.The Least Squares method fits the lane pixels,and uses Kalman filtering to track the lane line to improve the detection efficiency.It ensures that the algorithm is in the urban structured road environment including tree shadows,new and old road surface abrupt changes,and complex lighting conditions has a good detection effect.3.An algorithm for lane alignment based on lane detection information and extreme line constraints is proposed,and the lane line information projected into an environmental map constructed by the SLAM algorithm.Firstly,the improved algorithm used to extract ORB feature points and generate descriptors.The improved RANSAC matching optimization algorithm is used to improve the accuracy and efficiency of feature point matching,and a platform motion and pose track model is established.Then the tree-based Bow used in the closed-loop detection part.The word bag model applied to accelerate detection,and the Bundle Adjustment chart optimization method used to optimize the tracking pose.Finally,the extreme line constraint relationship and the lane line fitting information applied to quickly find the left and right image matching pixels and obtain the3 D information of the road and project it to sparse point map.4.Through the construction of stereo vision system,we collect the picture in the outdoor environment and verify the algorithm.The positioning accuracy is determined by comparing the actual distance from the lane to the camera and the distance of the lane.At the same time,the effectiveness of closed loop detection and closed loop optimization of the SLAM algorithm verified.