Research On Pedestrian Detection Technology Based On Multi-line Radar And Video Fusion

In recent years,autonomous driving technology has provided a brand new solution to solve the problem of urban road congestion.A complete and accurate perception of urban road traffic environment is the basis and premise for the smooth operation of autonomous vehicles.Among them,pedestrian target detection still faces many challenges.The actual traffic application scenario is more complex.In the process of pedestrian target detection,pedestrians and pedestrians,pedestrians and the environment will interact with each other,which will affect the pedestrian target detection effect,and a single sensor information is far from meeting the application requirements.The laser radar has the advantages of strong anti-interference ability and can obtain more accurate Angle,distance and velocity information.The camera can obtain high resolution and high quality two-dimensional visual information,which has great advantages in environment perception and target detection.Based on the comprehensive analysis of the advantages and disadvantages of common pedestrian target detection algorithms,the Autoware open source software is used to calibrate Velodyne VLP-16 lidar and A31 monocular camera jointly,so as to realize the synchronization in time and space,and the results of the joint calibration are verified.At the same time,a pedestrian detection algorithm based on the fusion of multi-line radar and video is designed to integrate lidar point cloud data and camera image data to realize accurate detection of pedestrian targets and judge the direction and distance of pedestrian targets.The pedestrian detection algorithm based on the fusion of multi-line radar and video is composed of Voxel R-CNN and YOLOv5.At the same time,the attention mechanism is introduced to improve the detection speed and accuracy of pedestrian targets.During the fusion of lidar point cloud data and camera image data,the feature fusion of lidar point cloud features with attention weight value and camera image features is carried out by means of focus sparse convolution.Data set test and case test results show that,through the introduction of attention mechanism,the pedestrian detection algorithm based on the fusion of multi-line radar and video realizes the effective detection of pedestrian targets,and has a good detection effect on the situation of more pedestrian targets,far pedestrian targets,pedestrian targets occurring color barrier,pedestrian targets blocked and nighttime environment.Through comparison,it is found that the detection speed of pedestrian target based on point map fusion feature information is faster than that of pedestrian target detection based on point cloud feature information and image feature information,and the detection accuracy is about 3% and 2% higher than that of pedestrian target detection based on point cloud feature information and image feature information.After the introduction of the attention mechanism,all the evaluation indexes of the accuracy of pedestrian target detection were improved by about 2 percentage points.