Detection and Tracking Method of Small-Sized UAV Based on YOLOv5

In recent years, accidents such as casualties and economic losses have occurred due to unmanned aerial vehicles (UAVs) flying in violation of regulations. The monitoring and countermeasures of UAVs with low flying altitude, slow flying speed and small size have become the current research hotspot. Vision-based methods are still one of the most mainstream methods, but due to the characteristics of UAVs such as small size, large attitude changes, and low flight altitudes, there are difficulties such as poor imaging contrast, complex background, and small proportion of targets. Aiming at the above difficulties, this paper proposes an improved YOLOv5 UAV detection algorithm and tracking method. The detection probability of drones is improved by adding a detection heads and attention module, and high-speed tracking performance is achieved by training a low-resolution detector combined with the Kalman algorithm. This paper deploys this method on NVIDIA Jetson Xavier NX, resulting in an output rate of 200 FPS.