THE UAV REMOTE SENSING VEHICLE TRACKING BASED ON IMPROVED FAIRMOT

Multi-object tracking (MOT) is an important problem in Earth-observation remote sensing task and has wide application. In recent years, UAV has been widely used in middle to high altitude remote sensing observation tasks, undertaking various tasks such as object detection and tracking, natural disaster monitoring and so on. The application of multi-object tracking to UAV platform has become one of the research trends. In this paper, a vehicle tracking algorithm based on UAV platform is proposed. In view of the size conflict between the network and the input image, bilinear interpolation method is adopted to resize the image and evaluate the loss. UAVDT dataset is used to train the algorithm and evaluate the model, thus achieving high-precision detection and tracking. At the same time, it has a high processing speed and achieves near-real-time performance.