STRCFD: Small Maneuvering Object Tracking via Improved STRCF and Redetection in Near Infrared Videos

Tracking small objects in infrared videos is challenging due to a complex background, weak information and target mobility. To deal with these difficulties, an infrared small target tracking algorithm is proposed, which utilizes the Spatial-Temporal Regularized Correlation Filter (STRCF) as the backbone. First, the local image patch that refers to the target and its neighborhood background is given in the STRCF. Then, the guided local contrast mechanism is designed to eliminate noise and distinguish the target from the background. Furthermore, the robustness of the tracking is improved by using the detection model with an adaptive factor of search range, helping to alleviate the problem of tracking migration. Experimental results on entire public near-infrared videos show the superior performance of the proposed algorithm (named STRCFD), compared to several related algorithms in visual effects and objective evaluation. It should be mentioned that the proposed STRCFD achieves an overall precision of 81.1%.