Object tracking in unmanned aerial vehicle (UAV) videos using a combined approach

UAV videos are difficult to process because of fast abrupt motion, low resolution, noisy imagery, cluttered background, low contrast, and small target size. In particular, object size varies from several to thousands of pixels in different videos, which is difficult to handle well using a single algorithm. The paper proposes a switching/combined approach for object tracking in UAV videos, where small objects are tracked using a spatiotemporal segmentation method and larger objects are tracked using a modified statistical deformable model. The proposed snake model addresses the problems in existing statistical snakes that require a good initialization, manual parameter tuning and high computational complexity. It detects clutter, high gradient noise and partial occlusion, and corrects the object bounding box by automatically adjusting the snake parameters. The computational complexity is significantly reduced by performing the operations only on a small image region and using a fast snake deformation method. The effectiveness of the proposed method is demonstrated using real UAV video sequences.