Multi-vehicle Detection and Tracking Based on Kalman Filter and Data Association

Environment perception is an important issue for autonomous driving applications. Vehicle detection and tracking is one of the most serious challenges and plays a crucial role for environment perception. Considering that the convolutional neural network (CNN) can provide high recognition rate for object detection, the vehicles are detected by utilizing Yolo v3 algorithm trained on ImageNet and KITTI datasets. Then, the detected multiple vehicles are tracked based on the combination of Kalman filter and data association strategy. Experiments on the publicly available KITTI object tracking datasets are conducted to test and verify the proposed algorithm. Results indicate that the proposed algorithm can achieve stable tracking under normal conditions even when the object is temporarily occluded.