Research on infrared unmanned aerial vehicle tracking method combining deep learning and kernel correlation filtering

The use of UAVs in warfare is becoming more and more widespread, and they pose an increasing threat to soldiers and military targets. Drones are low-cost, portable and stealthy. The ability of existing detection systems to detect UAV-type targets is obviously insufficient. Infrared imaging detection technology for detecting low-altitude small targets is more widely used, but there are also problems such as difficult target tracking and slow speed. Deep learning is widely used in target detection, but it is time-consuming and less used in the tracking task with high real-time requirements. The ultrahigh detection and tracking speed of correlation filters, however, can make up for this defect. Therefore, this paper combines deep learning target detection methods with kernel correlation filters for solving the real-time problem of infrared UAV detection and tracking. In this paper, we use open-source datasets to conduct experiments, using Faster RCNN as the detection method, combined with the kernel correlation filter method, to detect and track infrared UAV videos, and the tracking processing speed is about 160 frames/s. The tracking processing speed is about 20 frames/s using Faster RCNN as the tracking method. The results show that this paper's method can solve the real-time problem of infrared UAV detection and tracking.