This paper proposes an improved target tracking method. It combines the mutual corner matching in normalisation coefficient with Mean-shift. When part of the characteristics of target appears in the background or the similarity of the objectives and the background is quite similar, the tracking performance of Mean-shift algorithm will reduce. To solve this problem, in this paper we propose the fusion of colour features and corner points features, by using the algorithm of normalised mutual corner matching can effectively reduce the false matching rate, improve the matching accuracy. In some frames, due to the interference of noise and occlusion, etc. , zero matching on corner points occurs, then the Mean-shift algorithm is used as a temporary replacement to track the target, and the target template is updated as well to adapt to the rotational motion of the target. When the corner points come back to matching again, the corner matching is renewed for target tracking.%提出一种改进的目标跟踪方法.将归一化系数的角点互匹配与Mean-shift相结合.当目标的一部分特征出现在背景中或目标与背景相似度较高时,Mean-shift算法的跟踪性能将会下降.针对这一问题,提出采用颜色特征和角点特征相融合,用归一化的角点互匹配算法,能有效降低误匹配率,提高匹配精度.在某些帧中,由于噪声、遮挡等干扰时,发生角点0匹配,这时采用Mean-shift算法作为临时替代跟踪器,并更新目标模版,以适应目标的旋转运动,当有角点恢复匹配时,重新进行角点匹配跟踪.
展开▼
