Lidar-Based Road Detection In Adverse Weather ——In The Rain And Snow

As an important road detection sensor in the assisted driving system,lidar has the advantages of high precision and rich information,which can be used to detect and identify objects and provide driving environment information.Lidar is an active sensor that does not rely on external conditions to obtain information and can work day and night.Since the currently used 905 nm and 1550 nm bands are disturbed by the weather,raindrops,snowflakes,and smoke will attenuate the laser beam and hinder the return to the laser,which will affect the detection performance of the lidar system and reduce the detection accuracy.Therefore,it is of great significance to study the road detection of lidar in adverse weather and increase the clarity of point cloud images for increased driving safety.In this paper,road detection based on lidar under adverse weather conditions is studied.The lidar target imaging experiments in the rain and snow were carried out respectively,and the laser scattering by rain and snow was analyzed theoretically.According to the laser attenuation model in different environments,the farthest detection distance of lidar is obtained and verified by practice,and the target is accurately clustered and segmented by the analysis of point cloud.The main research contents are as follows:（1）Analysis and research of lidar detection mechanism and influencing factors.Starting from the basic concept of lidar,the composition,classification,structure,and detection mechanism of lidar are introduced,the working principle of lidar is analyzed,the detection power equation of lidar is obtained,and the attenuation model of laser in the process of atmospheric transmission is studied.Combined with theories such as Mie scattering,the functional expression of lidar detection ranged performance and attenuation coefficient is finally obtained.At the same time,an experimental platform is built based on this,and the point cloud of lidar under adverse weather conditions is obtained and compared with the theoretical model.Under adverse weather conditions,the relative error between the simulated detection distance and the maximum detection distance of the experimental data is within 2%-9%.It can be seen that the actual attenuation of the lidar is consistent with the calculation results of the theoretical model,which proves the rationality of the theoretical model.（2）Research on target detection processing of laser point cloud.Firstly,the source of noise in the lidar scanning data is analyzed,and the noise in the original point cloud is preprocessed in this paper.Considering that the detected targets are all far away from the lidar installation location and the lidar installation has a certain degree of accuracy.Therefore,the data onto the lidar installation origin 0.5 meters to the maximum detection distance of lidar is set as the valid data of the experiment,then the data onto 0.5 meters of the lidar is determined to be independent of the follow-up work and directly discarded.Since the sampling scenes are adverse and broad,such as rain and snow,there are many outliers in the original point cloud that deviate from the main point cloud.Based on the statistical outlier filtering method,a large number of outliers are further removed in this paper.Secondly,the method of cloth simulation is used to distinguish the ground points and non-ground points of the point cloud,which not only reduces the computational complexity of subsequent processing but also facilitates the extraction of subsequent local features and the subsequent detection of target clustering.The experimental results show that in the rainy and snowy weather environment,after processing,part of the influence caused by the weather can be eliminated.Finally,the dynamic and static objects in each sampling scene can be successfully distinguished according to their shape characteristics.