An improved approach to video target tracking based on Mean Shift

This paper proposes an algorithm based on Mean Shift, which improves on the kernel function, alterative weights, combing with Kalman filter and neighborhood searching. These improvements not only enhance the capacity of target tracking, but also reduce the computations to satisfy the need of the real-time job. Furthermore, experimental results illuminate that the proposed algorithm can cope with clutter, target partial occlusions, scale variations and fast moving in the real-time video target tracking.