Parallel search strategy in kernel feature space to track FLIR target

Robust tracking in the forward looking infrared (FLIR) sequences is still a challenging problem in the field of computer vision. Because images acquired by the infrared sensors are characterized by low signal-to-clutter ratios (SCR) and targets of interest may exhibit profound appearance variations due to ego-motion of the sensor platform and complex maneuvers. Though many efforts have been delivered, there are still some issues to be addressed. First, intensity features are not enough to deal with complex appearance variations for the challenging sequence. Second, to obtain satisfying estimation of target state, a plenty of particles have to be employed to approximate its probability density function (pdf). To deal with the two problems, a parallel search strategy based on kernel sparse representation (KL1PS tracker) is proposed to perform the tracking task in the FLIR sequences. With the ability of capturing the nonlinear features, kernel method is introduced to deal with complex appearance variations. After the kernel function is constructed based on the histogram features, both the target templates and candidates are mapped into the kernel feature space. Then efficient state particles are selected based on sparse representation which can be used to estimate the target state. The proposed method is tested on the AMCOM database, and the experimental results demonstrate its excellent performance in tracking accuracy compared with some state-of-the-art trackers.