Using YOLO-based pedestrian detection for monitoring UAV

Pedestrian detection (PD) is an important application domain in computer vision and pattern recognition. Conventional PD in real life scene is usually based on a fixed camera, which can detect and track the pedestrians in the monitoring region. However, when the pedestrian leaves the visible area of the fixed camera, it is usually difficult, if not impossible, to monitor the pedestrian. In response to the limitations of the conventional pedestrian detection application scenarios, a four-rotor unmanned aerial vehicle (UAV) system equipped with a high-definition (HD) camera is designed and implemented to detect human targets. Considering the size of human body in aerial image is small and easily to be occluded, we draw on the advanced research results in the field of target detection and propose a robust pedestrian detection method based on YOLO (You Only Look Once) network. The flow of the proposed approach is as follows. Firstly, the HD camera, which is installed on the monitoring UAV, is used for capturing images of the designated outdoor area. Secondly, image sequences are collected and processed using the airborne embedded NVIDIA Jason TX1 and Ubuntu as the core and operating system, respectively. Finally, YOLO is used to train the pedestrian classifier and perform the pedestrian detection. Experimental results show that our method has good detection results under the complicated conditions of detecting small-scale pedestrians and pedestrian occlusion.