A Real-Time Visual Tracking for Unmanned Aerial Vehicles with Dynamic Window

Visual object tracking is one of the basic topic in computer vision that has been studied for decades. As deep neural network has been applied in trackers, their tracking accuracy has been greatly improved and achieved state-of-the-art performance in recent years. Due to the special viewing-angle of UAV, the video sequences captured are high resolution and large in size. So, when applying these trackers to UAVs for tracking a target, the processing speed is hard to reach real-time requirement. In this paper, we propose a visual object tracking framework, DyWinSiam, which can speed up the tracking process significantly compared with state-of-the-art deep neural network trackers, yet it still achieve compatible accuracy.