摘要： Mean-Shift算法无法自动跟踪目标，且对目标形状要求较苛刻。针对该问题，提出一种基于形状感应的运动目标跟踪算法，采用混合高斯分布对背景建模，协助Mean-Shift算法自动定位初始目标，增加描述形状的协方差参数，使跟踪能感受到目标形状的变化。实验结果表明，该算法基本解决了自动定位问题及形状变化问题，在保证实时性的前提下，跟踪准确度提高40%以上。

关键词: 目标跟踪, 高斯混合模型, Mean-Shift算法, 协方差, 形状

Abstract: A moving object tracking algorithm based on shape induction is proposed to solve problems of the Mean-Shift algorithm, which can not track the targets automatically and adapt well to the shape changes of objects. It obtains initial contour of objects automatically by building a background image with mixture Gaussian distribution. The parameter of covariance matrix is added to Mean-Shift algorithm to feel the shape changes of objects. Experimental results indicate this algorithm solves the problems concerned with automatic positioning and shape changes and the accuracy is increased by 40% at least with real-time guaranteed.

Key words: object tracking, Gaussian Mixture Model(GMM), Mean-Shift algorithm, covariance, shape

中图分类号: 

• TP311.52