A Multiple Instance Learning tracking method with covariance matrix

作者：Ning Jifeng^*; Han Wenjing; Yang Shuqin; Zhao Yaobo

来源：Journal of Computational Information Systems, 2012, 8(20): 8553-8560.

• Face sequences
• Feature values
• Haar-like features
• Harris response
• Location accuracy
• Multiple features
• Multiple instance learning
• Object Tracking
• Tracking method

A Multiple Instance Learning tracking method with covariance matrix is presented to improve the location accuracy of classical Multiple Instance Learning tracking method with Haar-like feature. Covariance matrix fuses the multiple features and is enough to discriminate it from other distributions and so it is selected to model the target. Then the difference between determinant and square of trace of the covariance matrix called Harris response is regarded as the feature value of the interested region. We employ the Harris response of covariance matrix instead of traditional Haar-like feature in Multiple Instance Learning tracking method to represent the target. Experimental results for two representative face sequences show that the proposed method gets better performance than classical Multiple Instance Learning tracking method.

• 出版日期2012