Mean Shift 是一種密度梯度的無參數(shù)估計方法，應(yīng)用于目標(biāo)跟蹤領(lǐng)域有較好的性能。
然而Mean Shift 算法是一種半自動跟蹤方法。為此，提出邊緣檢測與Mean Shift 相結(jié)合的方法。利用結(jié)合小波的Canny 邊緣檢測方法對初始幀進(jìn)行目標(biāo)檢測，通過力矩在圖像處理中的計算方法得出目標(biāo)的形心位置，為Mean Shift 方法提供初始信息。在目標(biāo)丟失時，可以利用形心方法修正Mean Shift 的迭代跟蹤過程。實(shí)驗(yàn)結(jié)果表明，本方法可以實(shí)現(xiàn)對目標(biāo)的自動跟蹤，同時有效修正了對紅外目標(biāo)的跟蹤偏移。
關(guān)鍵詞:紅外目標(biāo)；形心定位；自動跟蹤；Mean Shift

Infrared target auto-tracking method based on Mean Shift algorithm Lian Jie, Han Chuanjiu (Dept. of Information and Communication, Guilin Univ. of Electronic Technology, Guilin 541004) Abstract: Mean Shift is a nonparametric estimator of density gradient. Aside from the good application result in target tracking, it is a semiautomatic tracking algorithm. Therefore, a
combination method of Mean Shift and edge detection is proposed. Canny edge detection based on wavelet is used to detect target in the first frame. Through moment knowledge used in image processing, the target center is computed for Mean Shift to initialize. When occlusion makes the target lose, that can modify Mean Shift tracking. Experimental results show that the proposed algorithm can track automatically and correct the excursion during tracking.
Key words: infrared target；center locating；auto-tracking；Mean Shift