基于改进C-V模型的步态轮廓提取方法  

Gait Contour Extraction Approach Based on Improved C-V Model

作　　者：徐中宇[1] 姜洪霖[1] 张忠波[2] 

机构地区：[1]长春工业大学计算机科学与工程学院,长春130012 [2]吉林大学数学学院,长春130012

出　　处：《计算机工程》2010年第17期182-183,186,共3页Computer Engineering

基　　金：吉林省科技发展计划基金资助项目(20070562)

摘　　要：为从步态图像视频序列中提取运动人体轮廓,提出一种新的分割算法。结合背景减除法和对称帧差法初步提取运动人体区域,采用形态学方法消除噪声和阴影,提取较完整的人体轮廓,将此轮廓作为改进C-V模型的初始零水平集,通过少量迭代达到细化人体轮廓的目的。在NLPR步态数据库上的实验表明,该方法能提取完整、精确的运动人体轮廓,消除了大量阴影,具有良好抗噪性且时间复杂度低、识别率高。In order to extract human contours from moving gait image sequences, this paper proposes anew division approach. It preliminarily extracts the area of the moving human-body by combining the background subtraction method and the symmetric frame difference method. Morphology algorithm is used to stuff the holes, banish the noise and shadow, and extract the mostly entire contours of human-body. The contours serve as the primary zero level set of improved C-V model to get the aim of gaining the accurate contours of human-body by handful iteration. The experiments in NLPR database show that the approach can extract the entire contours of moving human-body accurately, banish a lot of shadow and has good performances including noise robustness, low time complexity and high recognition rate.

关 键 词：步态识别 背景减除 对称帧差 C—V模型 水平集 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]