SETh: The method for long-term object tracking

The article presents a novel long-term object tracking method called SETh. It is an adaptive tracking by detection method which allows near real-time tracking within challenging sequences. The algorithm consists of three stages: detection, verification and learning. In order to measure the performance of the method a video data set consisting of more than a hundred videos was created and manually labeled by a human. Quality of the tracking by SETh was compared against five state-of-the-art methods. The presented method achieved results comparable and mostly exceeding the existing methods, which proves its capability for real life applications like e.g. vision-based control of UAVs.