| The most important part of the automatic driving system is environment perception.Depend on the processed information collected by the vehicle sensor,the automatic driving system can achieve correct and accurate discrimination and control.Basing on data fusion processing,the environmental information obtained by multiple cameras and radars is comprehensive and accurate.So multiple cameras and radars are widely used in modern autonomous vehicles.At present,the widely used visual target detection methods can not provide accurate information of anterior obstacles because they are often affected by occlusion,light,road conditions and other factors.The laser radar has a long detection range and can detect the three-dimensional information of the target.But the obtained feature points are relatively sparse,which is difficult to meet the needs of the current perception system.Therefore,how to detect the target accurately and quickly in the complex and changeable traffic environment the basing on the existing conditions is still the critical and difficult point in the research of environment perception.Basing on the highway environment,.the environment information data obtained by 3D lidar and camera is studied respectively basing on the point cloud data object detection algorithm and image information object detection algorithm,Then,an efficient and reliable data fusion detection method is established,which can give full play to the advantages of multi-dimensional heterogeneous sensor sensing system and realize accurate and real-time anterior obstacle detection.The main contents of this thesis are as follows:(1)In view of the problem of vehicle false detection caused by incomplete pavement segmentation on daytime backlight expressway,a vehicle detection method based on vehicle underbody shadow features is proposed.Firstly,the image is preprocessed,and the maximum variance method between three classes is used to segment the image.The anterior vehicle is detected by the bottom shadow feature of the vehicle.Then,in the candidate box of detection,the symmetry of the vehicle contour is used to verify whether the region is the vehicle region,so as to improve the accuracy of vehicle detection.Finally,from the experimental results,it can be seen that the image detection algorithm in this paper can better detect the anterior vehicle,and the detection rate reaches more than 90%.(2)According to the characteristics of the point cloud acquired on the highway,the object detection method based on the clustering of obstacle point cloud is defined.Firstly,statistical filtering and direct filtering algorithm are used to preprocess point cloud data,and a ground point cloud segmentation algorithm based on pavement plane fitting is improved.Then,KD tree is used to establish the topological relationship between point clouds,and constraints of clustering size and reflection intensity are added to improve the accuracy of clustering segmentation of point clouds.Finally,a three-dimensional bounding box is established to describe the spatial information of obstacles by principal component analysis.(3)First of all,the data of the two sensors must be registered in time and space,so that the point cloud data detected by lidar and the image data obtained by camera can be fused,,After the lidar points glow class target projection to the image,the formation of the corresponding point cloud detection rectangular area,coupled with the camera target detection of rectangular box as a result,the weighting factor to improve the conflict of D-S evidence theory is carried out on the two sensors decision level fusion.The experimental results show that the target detection accuracy of fusion algorithm is higher than that of Single sensor detection. |
