Front Obstacle Detection Based On Millimeter Wave Radar And Camera Data Fusion

Advanced Driving Assistance System(ADAS)technology and Autonomous Driving technology are the key parts of environmental awareness.It is the main task of environmental awareness research to enable vehicles to obtain more abundant and accurate target information.Because a single sensor has its own advantages and disadvantages in detection,it is difficult to obtain comprehensive target information only by relying on a single sensor,and multi-sensor fusion detection technology can solve this problem.In many automotive sensors,millimeter wave radar detection has the advantage of target motion information,and the camera target classification and contour detection has the advantage of sex,to data fusion of millimeter wave radar and cameras,can be complementary to the pros and cons of each detection performance,thereby gaining goal detection is more comprehensive,accurate information,also can reduce the error check each single sensor targets.Therefore,this paper proposes a forward obstacle detection method based on millimeter-wave radar and camera data fusion to detect obstacles in the road environment ahead of vehicles.The specific contents are as follows:First,parsing the raw data collected in millimeter wave radar and data reliability screening,and according to the safety area selection by millimeter wave radar point cloud data,after the screening of millimeter wave radar data after the research of target tracking algorithm,according to the properties of millimeter wave radar detection proposed improved DBSCAN(Density--based Spatial Clustering of Applications With Noise)Clustering algorithm and the gate size change With target distance,The millimeter-wave radar target tracking algorithm based on the near neighborhood method and Kalman filter algorithm is realized,and the effective target information of millimeter-wave radar point cloud data is tracked and obtained.Finally,the target tracking algorithm is verified by screening the data collected by the millimeter-wave radar vehicle.Secondly,the paper analyzes the visual target detection method based on the deep learning relative advantage of traditional visual target detection method,put forward in this paper,based on YOLOv4(You Only Look Once version 4)target detection method of deep learning model,and the nu Scenes original target detection analysis,the distribution of training data set for the tag number after processing and optimization,this paper got the label distribution more reasonable number of training data set.The training of YOLOV4 model and SSD(Single Shot Multibox Detector)model has been completed.The training results of the two models,as well as their advantages and disadvantages,have been compared,and the video data collected by real vehicle driving has been verified and analyzed.The analysis results show that YOLOV4 has stronger target detection ability,more accurate target positioning and target extraction ability.Finally,the actual data of the monocular distance measuring model is verified,and the accuracy of the estimated distance is obtained under different distance segments.Then,design the millimeter wave radar and camera to collect the data of time synchronization algorithm,and completed the millimeter wave radar data projection to the camera image pixel coordinates algorithm,at the same time to reduce the number of code data fusion arithmetic,in view of the two sensor detection range and performance,the design of the principle of screening target data joint detection area,based on the above work is put forward in this paper,the millimeter wave radar and camera data fusion target detection algorithm,including the detection of target information matching algorithm and both the output of the information strategy,offer more perfect in front of the vehicle obstacle information.Finally,offline data verification is also carried out for the target detection algorithm of millimeter-wave radar and camera data fusion proposed in this paper,which proves the effectiveness of the fusion algorithm.Compared with a single sensor,the fusion algorithm can obtain more comprehensive and accurate information about the road target ahead,and can also reduce the missed detection when detecting by a single sensor.