Fusion of trackers on thermal image sequences

A set of trackers operating on thermal video data is investigated. The contribution focuses on marine scenarios with ships that may perform maneuvers such as turning or occluding each other. The tracker methods are not original, and they are not improved here, only their parameters were adjusted in order to fit the scenario. This contribution focuses on the fusion of the trackers. Their failures occur at different times on the same video. Thus, a committee solution performing a weighted mean is calculated and outlier-trackers are removed appears appropriate. The outliers are then newly locked on to the majority estimate. Thus, later on, the fusion may still benefit from their contribution. Experiments are performed on a thermal video taken at a Singapore waterfront displaying boats performing turning maneuvers and occluding each other.