A Model-Free Moving Object Detection and Tracking Framework Based on UAV Data

At present, object detection and target tracking play an important role in the development of unmanned aerial vehicle(UAV) applications. The detection and tracking algorithms based on deep learning show excellent performance on aerial data. However, these algorithms, depending on GPU and high-quality datasets, are difficult to migrate to terminals with microcomputers. In this paper, we develop a tracking method combining DeepSort and Histogram of Oriented Gradient (HOG), named HOGSort within a model-free and non-deep learning framework for detection and tracking of moving objects. Moving objects are detected by inter-frame difference algorithms and tracked by HOGSort. Experiments based on aerial videos show that most trackers can continuously track their respective targets, proving that our framework is effective.