Research On Target Tracking Method Based On Fusion Of Millimeter Wave Radar And Visual Information

Environmental perception is the primary step of an intelligent assisted driving system and the foundation of decision-making and control parts of the system.Due to the performance limitations of individual sensors,the development of environmental perception technology is focused on the fusion of information from multiple sensors.This thesis proposes a targettracking method based on millimeter-wave radar and visual sensors to address the development needs of environmental perception technology.The main research contents are as follows:(1)Research on target tracking methods based on millimeter-wave radar.The selection of millimeter-wave radar and data analysis and acquisition are completed.Noise signals are filtered out by setting an effective detection area and using the DBSCAN clustering method.The extended Kalman filter and the joint probabilistic data association algorithm are combined,and a counting module is added to propose a multi-target tracking algorithm framework for data association and trajectory management of multiple targets.The experimental results show that the algorithm can effectively and continuously track targets.(2)Research on improved YOLOv5s-based target detection algorithm.To address the problem of low detection accuracy of the YOLOv5 s algorithm in complex road scenes,the network structure is improved by introducing the coordinate attention mechanism and bidirectional feature pyramid network to enhance the feature extraction ability of the network.Meanwhile,a manually-labeled self-built training dataset is used to train the neural network,and ablation experiments and road experiments are conducted on the trained model.The experimental results show that the improved network model has increased detection accuracy by 5.19%,6.05%,and 6.56% compared to the original algorithm in sunny weather,rainy weather,and nighttime scenes,respectively.(3)Research on fusion target detection algorithm based on millimeter-wave radar and visual information.A spatial and temporal synchronization model between the millimeter-wave radar and the camera is established,and the camera’s intrinsic and extrinsic parameters are solved.The radar data points are projected onto the image,and the target detection box size for the projection of millimeter-wave radar data onto the image is determined.Finally,a fusion target detection algorithm based on millimeter-wave radar data and visual information is proposed,and a comparative experiment is conducted to evaluate the performance of the target detection algorithm.The experimental results showed that the fusion detection algorithm improved the detection accuracy of single radar and single vision algorithms by 7.80% and 3.59%respectively in sunny weather.The fusion detection algorithm improved the detection accuracy of single radar and single vision algorithms by 2.03% and 31.47% respectively in nighttime scenes without street lights.These results verify the effectiveness of the proposed fusion target detection strategy.