Research On Vehicle Target Recognition Technology In Highway Scene Based On Airborne Lidar

Airborne LiDAR has the advantages of wide scanning range,highflexibility and strong anti-jamming ability.It is widely used in surface mapping,urban planning,highway monitoring and other fields.According to statistics,most of all traffic accidents are caused by motor vehicles,among which large vehicle accidents are the most serious.Based on airborne LiDAR,through the research of data acquisition,point cloud preprocessing,vehicle target recognition and other technologies,this paper aims to identify vehicles on highway efficiently and accurately,and detect large vehicles.The main research contents of this paper are as follows:(1)Point cloud data collection.This paper introduces the mainstream LiDAR,analyzes the scanning principle of Livox LiDAR,studies the point cloud data acquisition program,and obtains the continuous frame point cloud files in the highway scene.(2)Point cloud pre-processing technology based on highway scene.The pre-processing process is divided into three steps:denoising,ground removal and clustering.Through experiments,four common denoising algorithms are compared,and finally the conditional filter and radius filter are used for joint denoising.The RANSAC algorithm is improved,an adaptive fast ground removal algorithm(LP-RANSAC)is proposed.The algorithm can quickly remove multiple road surface.Experiments verify the effectiveness and reliability of the algorithm.Through experiments,the segmentation effects of density clustering and Euclidean clustering are compared.Aiming at the problem of region over segmentation,a clustering algorithm based on Euclidean distance and region judgment is proposed.Using this algorithm,a large number of invalid targets are eliminated.(3)Vehicle target recognition technology based on machine learning.Various feature descriptors are analyzed.The principles of various machine learning algorithms are described in detail.A multi-feature fusion descriptor is designed.The performance of different features under the same classifier is tested.The effectiveness of multi-feature fusion descriptor are verified.The vehicle recognition function is realized.The performance of the same feature under different classifiers is tested.The stability of multi-feature fusion descriptor is verified.(4)Large vehicle detection technology.A large vehicle detection algorithm based on single frame is realized.Aiming at the problem of recognition discontinuity,a large vehicle target detection algorithm based on continuous frame is proposed.The effect of large vehicle target detection is improved.