Small target tracking in forward looking infrared imagery based on kernel density estimation

To deal with difficulties inherit when tracking small targets in a forward looking infrared (FLIR) image, the authors proposed an approach based on kernel density estimation. The intensity and locally weighted intensity entropy were fused to model targets. Tracking was performed by computing the mean shift vector that minimizes the distance between the kernel distribution for the target cadidate area and the model. The target might change slowly or it can alter drastically if the illumination changes or the target is obscured by other objects during the course of tracking. A strategy was proposed to update the model based on the Bhattacharyya coefficient, thus overcoming the problem of tracking failures caused by the model being under-updated or over-updated. Experiments verified that the algorithm is robust in tracking small targets in FLIR image sequences.