Online RFC-PRM method for infrared target tracking

A novel online tracking method is proposed based on the response of feature combination (RFC) and relative position map (RPM). First, large numbers of features around the target are achieved after some designed filtering, then a suboptimal feature combination obtained by a given feature selection criteria can be used for next frame. The distinction between the target and background is maximized in the result image. Based on this distinction the target can be segmented out easily. While around the target there always exists some similar regions and will be residue left after filtering. To depress these clutters, the position relationship of local maximal response points cased by the similar regions is recorded as RPM and compared to the last frame. At last, the new feature combination is updated based on the new background around the target. The experiment using small infrared target sequence with different background texture conditions demonstrates the robustness of the new method.