摘要
阐述了2种简单有效的基于步态的身份识别方法——基于模型的方法和非基于模型的方法.基于模型的方法利用人体的骨骼化模型,首先对输入的图像序列自动进行背景初始化;然后分割图像中运动人体的侧面影像,并进一步细化为人体的骨骼化模型;接着从模型中提取人体的静态参数以及动态参数作为特征.非基于模型的方法计算图像间的光流场,从光流场中进一步提取可识别特征.将2种方法应用于室内拍摄的视频,实验结果表明,通过提取可靠的步态特征,降低了数据处理的代价,而且得到了较好的识别性能.
Two simple and effective gait recognition methods-model-based and model-free are discribed. The model-based method are sorts to a skeletal model of the body. The background of the gait sequence is in itialized automatically. The body silhouette is segmented from the image and is converted into a skeletal model afterwards. And then we extract body's static and dynamic parameters such as height, the position of the joint and the angle of the body etc. The model-free method calculates optical flow among images and gait features are extracted from optical flow. The utility of the proposed method is illustrated using indoor video sequences in our experiments and a good identification performance is gained.
出处
《北京工业大学学报》
CAS
CSCD
北大核心
2005年第4期388-393,共6页
Journal of Beijing University of Technology
基金
北京市自然科学基金资助项目(4031004)北京市教育委员会科技发展基金资助项目(km200310005006).
关键词
步态识别
骨骼化模型
光流场
主元分析
gait recognition
skeletal model
optical flow
principal component analysis (PCA)