A system was designed to automatically identify a person from a front-view angle in a video sequence, including the modules of Adaboost pedestrian detection, Adaboost face detection, complexion verification, gait preprocessing, period detection, feature extraction, and decision-making level amalgamation and identification. The face detection module and gait period detection module can be activated automatically by the pedestrian detection module. The experimental results show that the swinging arm region can be detected for obtaining the front-view gait period accurately with minimal computation, which is suitable for real-time gait recognition. Applying gait features assisted by face features to the decision-making level amalgamation method to solve human identification in a video sequence is a new idea. Even in gait recognition with a single sample per person, this proposed scheme can achieve an improvement in the correct recognition rate when face and gait information are integrated as opposed to using gait features alone.%为了能够实现视频下正面人体身份的自动识别,设计的系统包括Adaboost行人检测、Adaboost人脸检测、肤色验证、步态预处理、周期检测、特征提取以及决策级融合识别等模块.通过行人检测模块可以自动开启人脸检测模块和步态周期检测模块.实验结果表明,提出的根据下臂摇摆区域确定步态周期的方法对正面步态周期检测准确,计算量小,适用于实时的步态识别.采用人脸特征辅助步态特征在决策级的融合方法是解决视频下身份识别的新思路,在单样本的步态识别中,融合人脸特征可以提高识别精度.
展开▼
