Research on Detecting and Tracking Algorithm of UAV Intrusion Based on YOLOv5+DeepSort
Huaizhou Yang, Yujie Ge · 2022 3rd International Conference on Computer Vision, Image and Deep Learning & International Conference on Computer Engineering and Applications (CVIDL & ICCEA) · 2022
In order to accurately detect and track intruding UAVs for the phenomenon of ‘black flight’ and ‘indiscriminate flight’, this paper proposes an improved YOLOv5 algorithm combined with the DeepSort tracking algorithm, which adds the convolutional block attention module (CBAM) to the Neck module of the YOLOv5s network to enhance the extraction of network features. DIOU-NMS is introduced to improve the problem of missed detection due to the obscured UAV targets. The tracking part uses the DeepSort algorithm, in which the Kalman filter is used for predicting and updating the next frame position of the target, and DIOU is used to solve the problem of secondary matching failure. Experiments show that the proposed method can effectively identify and track UAVs, and furthermore reduce the problems of missed detection and target ID misdetection that occur after obstacle occlusion.