Robust object tracking in crowded and cluttered dynamic scenes is a very difficult task in robotic vision due to complex and changeable environment and similar features between the background and foreground. In this paper, we present an improved mean-shift tracker which uses discriminative local saliency feature and a new spatial pattern preserved similarity metric method to overcome above difficulties in mean-shift based tracking approaches. The local saliency feature, which is composed of contrast color, texture and gradient around the target, is proposed to find the most distinguished features between the target and background, and it could enhance the tracking performance greatly in the cluttered and complex environment. Another important benefit of this feature is that the saliency map form could be easily embedded into the meanshift framework. The new similarity metric try to preserve the spatial pattern to reduce the similarity errors caused by different spatial structure. It is beneficial to the mean-shift tracker to face the targets and scenes which has identical texture and color feature and with different spatial patterns. Finally, the efficiency of the proposed improved mean-shift tracker is validated through the plenty experimental results and analysis.