Uav anti-collision visual detection algorithm

• Published in: 2023 4th International Conference on Computer Vision, Image and Deep Learning (CVIDL) pp. 214 - 217
• Main Authors: Han, Jun Yu, Li, Zhe, Li, Sheng Hou, Zhao, Bo Shen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12.05.2023
• Subjects: Computer vision; Deep learning; Feature extraction; Filtering; Object detection; Real-time systems; UAV anti-collision; Visualization
• DOI: 10.1109/CVIDL58838.2023.10166842

In order to solve the problem of unsafe flight of traditional "low, slow and small" civil UAVs, and reduce pres sure on air traffic control personnel., an improved YOLOv5m target detection algorithm is proposed for active target detection of UAVs. According to the characteristics of low-altitude UAV detection, an attention mechanism was added to the YO LOv5 feature extraction network, and the original weighted NMS was optimized to SOFT_NMS Finally, the accuracy and real-time performance of the model are verified by constructi ng real UAV conflict target detection data set. The experimen tal results show that the average accuracy, recall rate, accura cy are 2.31 %, 3.66% and 5.59% higher than before the impr ovement, meet the anti-collision detection requirements of civi 1 UAV s in low altitude airspace.