Research into optical markerless human motion capture has attracted significant attention. However, the complexity of the human anatomy, ambiguities introduced by lacking full-perspective view data and noises involved in data processing make human motion capture a non-convex, multi-modal optimisation problem in high dimensional space. The proposed method incorporates an isotropic distance map calculated by Fast Marching Method with simulated annealing to efficiently explore the high dimensional space in the probabilistically important directions. This allows simulated annealing to relax the fitness test criteria and work in an incremental fashion, reducing the likelihood of being trapped in local minima. Moreover, instead of using a slow annealing schedule, the annealing progress is controlled by accounting for the shape of the importance distribution, thereby leading to an effective exploration and fast convergence. To improve robustness of tracking, Fast Marching Method is utilised to smooth noisy silhouette images. Finally, several experiments show the efficiency of the proposed method.