Detection and Tracking of Low-Altitude Unmanned Aerial Vehicles Based on Optimized YOLOv4 Algorithm

With the popularization of nonmilitary unmanned aerial vehicles (UAVs), UAV-detection technology has become a hotspot in security research. This study proposes a low-altitude UAV-detection and -tracking method based on the optimized YOLOv4. This method combines detection technology based on convolutional neural networks with a tracking algorithm for the first time to achieve dynamic detection of low-altitude UAVs. First, the original YOLO network structure is optimized based on multiscale feature fusion. Thereafter, in combination with the DeepSORT multitarget tracking algorithm, the detection and tracking model is constructed. Training and comparative experiments are performed on the self-built LARotorcraft dataset. The experimental results show that the proposed model can effectively reduce the miss detection rate for small targets. Good real-time performance is obtained with an average detection accuracy of up to 77.2%, and stable tracking of visual targets is realized.