Research On Vehicle Detection Method Based On Lidar Point Cloud

With the improvement of the level of economic development,the number of cars continues to rise,and the incidence of traffic accidents is increasing.The intelligent auxiliary driving' system can reduce the incidence of traffic accidents by reminding and guiding the drivers.Vehicle detection is the key of intelligent auxiliary driving system.It can not only provide clues for obstacle detection,but also help vehicle path planning.Vehicle borne laser radar scanning measurement is a new technology of Surveying and Mapping Surveying and mapping field after GPS,can get the 3D information of the object surface with rapidly and efficiently,provides a new means for the field of digital city,traffic simulation,environment measurement and navigation to achieve rapid 3D modeling.Therefore,how to effectively deal with the laser point cloud data is of practical significance.Under this background,this paper takes vehicle borne laser scanning system as a way to acquire three-dimensional information,and researches the algorithm of vehicle LIDAR point cloud data processing,and focuses on solving vehicle detection problem of laser point cloud.In order to improve the accuracy and robustness of existing vehicle detection algorithms,this paper proposes a vehicle detection algorithm based on 3D laser radar in complex road environment.First,the ground segmentation algorithm based on the color cloth model is used to separate the surface and non ground points,while reducing the computational complexity of the subsequent processing,and at the same time,it is conducive to the extraction of subsequent local features and subsequent target clustering and segmentation.Aiming at the problem of target adhesion in traditional complex clustering algorithm,this paper proposes a super body clustering algorithm to achieve non ground segmentation,and applies minimum cut algorithm to optimize segmentation accuracy.The experimental results show that the hyperbody clustering has a good effect on preserving the local boundary characteristics of the point cloud data.Finally,a vehicle identification method combining 3D CNN and license plate detection is proposed in this paper.A suitable 3D CNN network is designed to automatically learn the features of the sample from a large number of samples.In view of the incomplete information of some vehicles,the license plate detection method is proposed.The experimental results verify the effectiveness of the proposed vehicle characteristics and reduce the leakage rate of the vehicle detection algorithm.Quantitative and qualitative analysis is carried out on the open dataset.The experimental results show that the proposed method can effectively detect the surrounding vehicle information in complex scenarios,and has good accuracy and robustness.