Thanks to the rapid development of computer vision technology, methods based on in-depth learning have gradually replaced counting methods based on traditional machine learning, and substantial progress has been made in the accuracy and real-time detection of abnormal crowd behavior detection. Firstly, it introduces the structure and application of the YOLO network of the one-stage detection system. Secondly, according to the development of the YOLO network model, it specifically introduces the abnormal behavior detection algorithms and research based on the YOLO v3 network, the YOLOv4 network and the YOLO v5 network. Finally, the summary The shortcomings of the current One-stage target detection network are analyzed, and the future research directions are prospected.

Unmanned Aerial Vehicle (UAV) has been widely used in military reconnaissance, smart transportation, public security and other fields. UAV-based person tracking is attracting incremental attention for its wide application requirements. Currently, some state-of-the-art visual tracking methods have achieved promising performance in common scenarios. However, in the scene of UAV-based person tracking, there will be long-term target disappearance and unpredictable dramatic target appearance changes, which still pose a huge challenge to UAV-based person tracking. In this work, a human-machine hybrid augmented tracking system based on eye tracking is proposed to cope with the challenge. During tracking, through the interaction between humans and machines, humans can provide real-time guidance and corrections to the tracker, and the tracker can also learn interesting targets from humans to enhance itself. The experimental results show that human-in-the-loop can remarkable improve the success rate and robustness of the tracking and our tracking system outperforms the state-of-the-art tracker in complex environments.