Preprocessing method for large-scale scattered LiDAR point cloud in urban environment
Many sensors used in intelligent vehicles provide sensing information of driving scenes in the form of point clouds, such as RGBD cameras, LIDARs, and stereo cameras. The sensing information can be used to process point cloud data to achieve obstacles. Detection and tracking, lane edge detection, obstacle trajectory prediction, high-precision map drawing, real-time positioning and map construction, and other types of intelligent vehicle autonomous positioning and environment sensing functions are indispensable for intelligent vehicles to achieve complete autonomous driving. The urban road environment has become a research hotspot and a difficult point in the field of autonomous driving of intelligent vehicles with its complex and ever-changing characteristics. In the urban three-dimensional environment, the point cloud obtained by multi-line LiDAR scanning is huge, including tens of thousands of data points with a large number of outlier noises. These factors greatly increase the difficulty of implementing various types of sensing and localization algorithms, which seriously affect the accuracy and real-time performance of point cloud data utilization. Therefore, proper point cloud simplification and preprocessing methods to efficiently handle such a huge amount of data are needed. The purpose of point cloud data preprocessing is to provide point cloud data with less outlier noise points, small data size and significant local features for subsequent processing of ground point cloud segmentation, outlier noise rejection, and downsampling. In this way, the efficiency of subsequent point cloud data utilization is guaranteed.
Professor Xu Youchun from the Intelligent Vehicle Laboratory of the Army Military Transportation College is dedicated to research the related technologies of intelligent vehicle autonomous driving, including automatic control of intelligent vehicles, environmental awareness, planning decisions and autonomous positioning. In terms of environmental perception, the research results of the application of the LiDAR sensor are particularly significant. Due to the problem of large density, outlier noise and scattered distribution of LiDAR point cloud in urban three-dimensional environment, it is not conducive to the utilization of point cloud data. A preprocessing method for large-scale scattered LiDAR point cloud in urban environment is proposed. Firstly, the point cloud data is transformed into the mean elevation map, and the point cloud is ground segmented by the height gradient between the grids. Then, the DBSCAN clustering algorithm is improved by the three-dimensional Voxel Grid division method, and the improved VG-DBSCAN clusters the point cloud. After clustering, the target point cloud is separated from the outliers to eliminate outlier noise in the point cloud. Finally, the point cloud is downsampled by Voxel Grid filter. The results show that the proposed method can perform real-time preprocessing on point cloud data, and the average time consumption is only 132.1 ms. After preprocessing, the accuracy of point cloud frame matching is improved by 2 times, and the average consuming time of frame matching is only 1/6 of the value before.
Fig.1 Ground segmentation results. (a) Before segmentation; (b) After segmentation.
Fig.2 VG-DBSCAN filter denoising local effect. (a) Before denoising; (b) After denoising
Fig.3 Inter-frame registration results. (a) Before registration; (b) After registration
The Intelligent Vehicle Laboratory of the Army Military Transportation College owns 16 core members(4 senior and 7 middle-level staffs), 3 doctoral students, and 11 master students. The laboratory is mainly engaged in intelligent vehicle related technology research and software and hardware platform development. The team participated in the National Intelligent Vehicle Future Challenge sponsored by the National Natural Science Foundation of China for seven consecutive years and won the first place for three consecutive years; Participated in the 2015 Zhengkai Avenue smart car public trial and completed the first public road trial of the domestic intelligent vehicle; In 2016, it participated in the “Cross-risk-2016” Ground Unmanned System Challenge and won the double-group champions of Group A and Group B; In 2018, it participated in the “Cross-risk-2018” and continued to be the champion in Group A and Group B. The intelligent team designed a new intelligent vehicle system composed of hardware platform and software algorithm under the Great Wall H7, H8 and H9 platforms, breaking through the research on the key technologies of intelligent vehicles. In 2013, it won the first prize of China Intelligent Transportation Science and Technology Progress Award, and the third prize of military science and technology progress. In 2014, it was honored with the Tianjin Excellent Research Team. The team has published more than 100 academic papers and obtained 5 national invention patents.
Zhao Kai, Xu Youchun, Wang Rendong. A preprocessing method of 3D point clouds registration in urban environments[J]. Opto-Electronic Engineering, 2018, 45(12): 180266.