An Adaptive Unscented Particle Filter Tracking Algorithm Based on Color Distribution and Wavelet Moment

Robust real-time tracking of non-rigid objects is a challenging task. The difficulty in visual tracking is how to estimate the object position quickly and reliably. Particle filter (PF) has proven successfully for nonlinear and non-Gaussian estimate problems, but its degeneracy problem is very serious. For alleviating the degeneracy problem of the PF, the choice of proposal distribution plays an important role. Therefore in the context, the unscented kalman filter (UKF) is utilized to generate the proposal distribution of the PF, which is the unscented particle filter (UPF). The UPF algorithm not only overcomes the degeneracy problem of the common PF algorithm, but the estimation precision is better. The article proposes the integration of color distribution and wavelet moment into the UPF for more improving the veracity and reducing the computation complexity of the tracking algorithm. Color distribution and wavelet moment are applied, as they are robust to partial occlusion, are rotation and scale invariant and computationally efficient. And wavelet moment has the multiresolution properties, so it is suitable for differentiating the details of the moving objects. Thus, an adaptive unscented particle filter tracking algorithm based on color distribution and wavelet moment (CW-UPF) is proposed in this paper. Experimental results show the availability of the proposed algorithm in this paper.