Multi-Sensor Fusion Pedestrian Detection Based On Fast R-CNN

Recently,Pedestrian detection is a hot research area in the field of intelligent vehicle,how to detect pedestrian correctly and efficiently is the main points.Different pedestrians have great changes in their figure,posture,angle of view,clothing and lighting.In addition,complex background scenes and camera's own motion and sway are all difficulties of pedestrian detection.In view of the above problems,this paper set up the integration of millimeter wave radar and camera model,successfully improves the robustness of the detection system;base on “Fast Region-based Convolutional Neural Networks” object detection framework,can achieve rapid and accurate identification of pedestrians.Based on the research status of target detection and recognition technology,this paper describes and compares the traditional machine learning method and the deep learning method.In order to improve the efficiency of the detection system,the target detection framework of Fast R-CNN is adopted in this paper.In view of the characteristics of the detection target,the network structure is redesigned,and the method of making the negative sample is proposed,which improves the applicability of the detection model.The millimeter wave radar can make up for the characteristics that the camera can not obtain the distance information of the target,and is not affected by the light and bad weather.This paper combines the millimeter wave radar with the camera in space and time.In order to achieve the spatial integration of millimeter wave radar and camera,the millimeter wave radar and camera are combined to calibrate the target area.The target area obtained by millimeter wave radar is projected onto the image through the coordinate transformation to get the region of interest(ROI).In order to achieve time fusion,this paper builds a fusion model based on the theory of thread synchronization.The fusion of millimeter wave radar and camera can reduce the dependence of a sensor on the complementary information provided by the two types of sensors.In order to verify the feasibility of the algorithm,the experimental verification of the fusion model is carried out in this paper.The accuracy of pedestrian detection in common scenes is compared and analyzed.Among them,the recall rate is 98%,the rate of missed detection is 2%,the false detection rate is 1.33%,and the detection precision is 96.1%.In order to verify the advantages of the fusion algorithm,the accuracy and efficiency of the fusion model and the accuracy and efficiency of the single sensor are compared and analyzed.The detection accuracy is increased by about 9.3% compared with the millimeter wave radar's detection accuracy,the false detection rate and the missing detection rate all decrease significantly.Compared with the detection accuracy of the camera only,the detection accuracy is increased by 3.9%,the false detection rate is decreased by 1.34%,and the missed detection rate has decreased by 3%.The detection rate is about 10%.