一种新的步态图像序列分割算法  

New segmentation algorithm on gait image sequences

作　　者：郭军[1] 文玉梅[1] 李平[1] 叶波[1] 李潇[1] 

机构地区：[1]重庆大学光电工程学院

出　　处：《计算机应用》2007年第8期2047-2050,共4页journal of Computer Applications

基　　金：重庆市自然科学基金资助项目(CSTC2006BB2155)

摘　　要：在运动目标步态识别中,从步态图像序列中提取出完整的人体运动轮廓对特征提取、目标分类和目标识别等有着非常重要的意义。提出了一种新的运动目标分割算法:首先应用改进的块匹配算法进行运动估计;然后运用分水岭算法把当前帧图像分割成许多封闭而不重叠的小区域;最后运用仿射参数模型进行运动块区域合并。在CMU步态数据库中采用基准算法进行的实验表明,运用所提出的算法能够提取出完整的人体轮廓,进一步提高步态识别的识别率。In gait recognition, the effective extraction of human contours from gait image sequences is of great concern to subsequent processes including feature extraction, object classification and object recognition. A new moving object segmentation algorithm was proposed. A new block-matching motion estimation algorithm was adopted at first. Then the watershed algorithm was used to divide current image into closed and non-overlapping regions. Finally, affine parameters model was utilized to merge the separate regions. The experiments on the CMU Mobo data set prove that this algorithm can effectively extract human contours and eventually improve the performance of gait recognition.

关 键 词：步态识别 块匹配算法 分水岭算法 仿射参数模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]