Visual Pedestrian Tracking from a UAV Platform

Visual tracking is a challenging problem in computer vision, especially when the camera platform is moving. To track a target pedestrian from a UAV (Unmanned Aerial Vehicle) mobile platform, multiple techniques, such as motion control of UAV, visual detection and tracking are needed. This study presents a new tracking method, which mainly involves object tracking, pedestrian detection and online color feature matching. The algorithm of the proposed method has two main components: a median flow tracker and a pedestrian detection which is based on fast feature pyramids. The proposed tracking method is thoroughly evaluated on various public videos and real scene videos from UAV cameras, and compared with existing state-of-the-art tracking methods to show its effectiveness.