Research On Pedestrian Detection System Based On Millimeter Wave Radar And Vision

With the rising of automobile ownership amount in China,traffic safety issues have become increasingly severe.And pedestrian detection and recognition technology,which belongs to advanced driver assistance area,have received increasingly attention.When detecting pedestrian on the travel trajectory ahead of automobile,it is necessary to extract pedestr ian information from the complicated traffic background to obtain their position and motion state.There are two commonly used pedestrian detection sensors,camera and millimeter wave radar.The former has the ability to identify and classify targets,but it is easily influenced by environmental background.Besides,it is difficult to position target object accurately.By contrast,Millimeter wave radar is robust to weather conditions,and it is capable of observing the position and motion state of obstacle s accurately,while it cannot classify different kinds of obstacles.Therefore,the multi-sensor detection technology that combines machine vision with millimeter wave radar can effectively improve the accuracy and reliability of pedestrian detection.Based on mentioned above,this thesis takes multi-sensor information fusion methods as the research point,and combines millimeter wave radar with machine vision to accurately identify and detect pedestrian in front of the vehicle.The main focus and innovation of this dissertation are as follows:1.Verified the validity of the detected information from millimeter wave radar.In view of the non-pedestrian object,false target object and discrete object information in the feedback of millimeter wave radar,the research comprehensively takes,the reflection cross-sectional area of the target object,the detection ability of the radar for different targets at different angles as well as the effective target life cycle into account.Thereby filtering the feedback information and getting stable and effective pedestrian information out of the redundant radar detection information.2.Based on machine learning,achieved the detection of pedestrians in front of the vehicle.The HOG feature and the SVM classifier are ado pted to detect pedestrians in the image,and the PSA method is utilized to reduce the rank of feature vector,thereby decreasing the detection time.Aiming at the problem of excessive detection window,a method of extracting the useful region based on sky and road segmentation is proposed,which improves the detection accuracy and speed up the detection procedure.By integrating the target object position and its corresponding pixel coordinate information in the map,a monocular ranging method that meets th e accuracy requirements is obtained.3.Integrated the feedback information from millimeter wave radar and machine vision.The data fusion model of radar and machine vision is built to realize the fusion of these two sensors in time dimension and space dimension.With the help of multi-feature information for target matching and feature information fusion according to sensor characters,an accurate target object matching method is obtained.Then,Kalman filter is used to track the disposed target information,thereby getting target object motion trajectory information effectively and stably.It is verified by experiments that the fusion method proposed in this paper can accurately detect pedestrian information in front of the vehicle and returns effective and stable pedestrian position information.And thereby providing pedestrian information on the road ahead of the vehicle for advanced driver assistance systems.