Research On Vehicle Detection And Tracking Algorithm Using Radar And Vision Fusion

With the development of automobile technology and the arrival of fast-paced life,automobile plays a more and more important role in people's life.Traffic accidents caused by automobile also make people pay more attention to automobile safety.As one of the core technologies of advanced driving assistance system(ADAS),environmental sensing technology collects and processes the environmental information of vehicle and its surroundings by vehicle-borne sensors,providing a basis for the decision-making and control of the vehicle system.Multi-sensor information fusion can obtain more abundant surrounding environment information and it has good environmental adaptability,which improves the accuracy and robustness of environmental sensing system.Accurate and real-time vehicle detection technology provides effective environmental sensing information for ADAS system,which is of great significance for improving driving safety and comfort of vehicle.Accordingly,the environmental sensing technology in ADAS system was took as the research object in this thesis.The preceding vehicle detection methods based on millimeter wave(MMW)radar and machine vision were studied.A sensor fusion method using MMW radar and machine vision was proposed to detect preceding vehicle,which can detect the preceding vehicle accurately and in real time.The main works of the thesis are as follows:(1)Vehicle detection and tracking of MMW radar.Firstly,the radar data was preprocessed by filtering out the empty target and setting the thresholds of relative speed and lane range,which aims to selected the effective vehicle initially.In addition,a multi-target tracking algorithm was proposed,which combined the unscented Kalman filtering algorithm,considering the multi-target data association and track management.Finally,the tracking algorithm was used to track the effective vehicle target continuously.(2)Vehicle detection and tracking based on machine vision.Firstly,a vehicle detection algorithm based on shadows underneath the vehicles and symmetry feature was proposed,which was used to generate vehicle hypothesis area quickly.Then,a vehicle classifier was trained using a large number of positive and negative samples with the method of AdaBoost algorithm combined with haar-like features.It was used to validate vehicle hypothesis area to identify the effective vehicle targets.Finally,the kernel correlation filtering algorithm was used to track effective vehicle targets,which aims to improve the adaptability of algorithm for various environments and reduce the false detection rate and the missed detection rate of algorithm.(3)Data fusion of MMW radar and machine vision.The data fusion algorithm of MMW radar and machine vision was proposed.The information fusion of MMW radar and machine vision in space and time was realized by camera calibration,coordinate system transformation and sensor sampling time unification.(4)Based on VS 2015 C++ platform and OpenCV algorithm library,a sensor fusion algorithm of MMW radar and machine vision was proposed.The real vehicle data was collected to verify the algorithm.The results showed that the proposed algorithm can meet the needs of the preceding vehicle detection and recognition under the conditions of good daytime illumination and structured roads.It had good environmental adaptability and provided accurate and effective road environment information for vehicle ADAS system.