基于MeanShift的自适应尺度变化跟踪算法研究.pdf

27 4 Vol. 27 No. 4 2010 4 M ICROEL ECT R ON ICS COM PUT ER April 2010 M ean Shift 何小诚, 黄 凯, 谭毅华, 田金文 ( , 4 30074) : 在实时跟踪系统中, 要求在跟踪过程中跟踪窗口的大小实时地适应运 目标的外观变化, 这对应着M ean- shift 的尺度变化. 针对跟踪这种尺度变化的问题, 在跟踪框内检测角点并进行匹配, 将所有得到的匹配角点对建立 仿射模型, 并采用最小二乘法求解得到尺度变化系数, 进而更新跟踪框尺度, 使得 Meanshift 算法可自适应地变化 尺度并跟踪到大小不断变化的目标. 实验结果表明, 提出的算法具有较好的准确性鲁棒性和实时性. : M ean Shift 目标跟踪 角点 仿射模型 : T P3 1 : A : 1000 - 7180 ( 2010) 04- 0069- 0 6 Adaptive Scale Object Tracking with Based on Mean Shift H E X iao- cheng, HUA N G k ai, T AN Y-i hua, T IA N Jin-w en ( T he State Key L aboratory f or M ult-i sp ectra Inf orm ation Processing T echnolog y, Huazhong U niv ersity of Science and T echnology, Wuh an 4 30074, China) Abstract: In r eal tim e tracking system , the size of tracking w indow should be updated w hen the scale of the moving object is being varied , w hich is corresponding to the scale chang e o f m ean shift . In order to solve the problem , this paper pro- posed a m ethod to catch the adaptive scale change. F irstly, the corner point s are detected in the tracking w indow . Sec- ondly , the detected point s are matched w ith the pr eceding fr am e, and t he aff ine model is con stru cted based on the m atch- ing result s. T hirdly, least squ are m ethod is taken to indu ce the scale coeff icient s. Lastly, the obtained scale is applied to t he m ean shift tracking fram ew ork . Exp er im ental result s sho w the impr oved algorithm could track obj ect s pr ecisely w hen scale of the obj ect changes prog ressively , as w ell as is robu st and operated real tim e. Key words: M ean Shift object s tracking corner