Abstract:Robust real-time visual tracking is a challenging problem due to the presence of noise, occlusion and variation of illumination. Compressed sensing based tracker, aka l1 minimization tracking, could overcome these difficulties to a remarkable extent. Nonetheless, the heavy computational complexity of linear programming causes a very poor efficiency of compressed sensing based trackers. We propose a modified compressed sensing tracker, which is termed Real-Time Compressed Sensing Tracking (RTCST), to strike the balance between efficiency and accuracy. Hash kernels yield a hash-matrix that is employed to reduce the dimensionality of observation. The optimization problem is solved via Orthogonal Matching Pursuit rather than standard interior-point approaches. With these two modifications, the proposed algorithm could achieve a real-time speed (which is up to 1000 times faster than that of the l1 …

Robust real-time visual tracking with compressed sensing