Visual tracker fusion and outlier detection on thermal image sequences

Visual object tracking is a challenging task in computer vision, especially if there are no constraints to the scenario and the objects are arbitrary. The number of tracking algorithms is very large and all have different advantages and disadvantages. Often they are developed for a single task and they fail in other scenarios. Normally their failures in the tracking process occur at different moments in the sequence. So far, there is no tracker which can solve all scenarios robustly and accurately. One possible approach to this problem is using a collection of tracking algorithms and fusing them. There exist various strategies to fuse tracking algorithms. In some of them only the resulting outputs are fused. This means that new algorithms can be integrated with less effort. This fusion can be called ”high-level” because the tracking algorithms only interact through the last step in their procedure. Trackers in the collection which lost the object are called outliers. To ensure the robustness of the fusion methods these outliers should be detected and reinitialized or removed from the tracking process. Additionally three fusion methods are investigated. They are called Weighted mean fusion, MAD fusion and attraction field fusion. In order to evaluate the performance of the fusion methods and the outlier detection a collection of thermal image sequences has been investigated