Research On Road Scene Measurement And 3D Reconstruction Method

With the rapid development of the transportation industry,advanced driver assistance systems have become the current focus of research in the field of smart cars.Road area detection and road obstacle detection,as two key components in this field,are important guarantees for improving road traffic efficiency,reducing traffic accidents,and ensuring safe driving.Based on the analysis of the advantages and disadvantages of the current commonly used methods and the current research status,a binocular vision sensor with high accuracy,strong anti-interference,low cost and rich scene information is selected.Although the method based on binocular vision is widely used in the field of road scene detection,the method still has problems such as poor real-time performance and low target detection accuracy.Based on the above consideration,this paper uses binocular vision as the basis to make corresponding improvements to the current problems,and realizes road surface segmentation and road obstacle detection and distance measurement in traffic.The specific research contents are as follows:Firstly,select a suitable binocular model is selected for the binocular vision three-dimensional reconstruction system.Aiming at the problem of poor real-time binocular vision,the road horizontal line information is used to realize the adaptive division of the ROI area of the acquired road pictures.In addition,the camera is calibrated using Zhang Zhengyou’s calibration method to obtain the camera’s internal and external parameters and distortion parameters,and then Bouguet’s method is used to correct the image.Secondly,it addresses the problems of poor road segmentation and low target detection accuracy.In road surface detection,the scene disparity map is obtained through semi-global stereo matching,and based on the traditional V disparity method,the maximum and minimum constraints and the removal of the remote area are combined to obtain the road information in the V disparity map.To improve the effect of road segmentation;In the obstacle detection,the target detection method based on YOLOv3 combined with binocular vision is used to detect and recognize the left view obtained by the binocular camera with the trained YOLOv3 model,and output the target type and target pixel coordinates Then,the disparity map obtained by the stereo matching of binocular vision is combined to realize the distance measurement of the target object.Finally,on the basis of binocular vision three-dimensional reconstruction,based on the standard data set KITTI,a comparative study of a variety of complex roads in traffic scenes is carried out to achieve the detection efficiency of the algorithm,the detection effect of the road area,and the accuracy of the obstacle detection.Analysis and evaluation.The experimental results show that the algorithm in this paper can effectively detect the travelable area of the road.Compared with the traditional algorithm,it improves the accuracy,recall and detection speed of road area detection.Among them,the accuracy rate can reach 90.36%;in the road obstacle detection experiment,the target recognition and ranging function are accurately and effectively realized,and the experimental results meet the current practical application requirements.