Accurate Specified-Pedestrian Tracking From Unmanned Aerial Vehicles

Recently, accurate target tracking is widely used in the field of Unmanned Aerial Vehicles (UAV). In this paper, we focus on the application of detecting and following a walking pedestrian in real time from the moving platform with many interferences. We present a scheme that uses CNN model (YOLO-V2) to detect pedestrian and matches the walking pedestrian with a postprocessing and feature queue and Locality constrained Linear Coding algorithm. After that the ground station receives and analyses the video stream from the parrot and sends back commands to control the motion of UAV. At the beginning of the tracking process, the UAV is hovering when one pedestrian will be selected as the special target. Visual information is acquired only through a front camera without assistant sensors. A parrot Bebop 2 is adopted in the experiment, which is the basis for doing experiments outdoors and experimental result verify the effectiveness of our solution.