F-E Fusion: A Fast Detection Method of Moving UAV Based on Frame and Event Flow

In recent years, the widespread application of UAVs has caused threats to public security and personal privacy. This paper presents a fast and low-cost method for UAV detection and tracking from fixed-position cameras. In our method, we capture event data and video frames through Dynamic Vision Sensor (DVS) and conventional camera respectively. We use the combination of Dynamic Neural Field (DNF) and clustering algorithm to locate the moving objects in the scene from the event data collected by DVS. Then we obtain high-resolution images from the corresponding regions of the video frame according to the calculated positions for classification. Compared with YOLO or RCNN, our proposed method reduces the computational overhead by calculating the location of moving objects through event flow. Experimental results show that our method has more than 40 times faster recognition speed on the same platform than YOLO v3. The data and the code of the proposed method will be publicly available at https://github.com/Xiaoxun-NUDT/F-E-fusion.