摘要
在计算机视觉与模式识别领域,人的视频身高肩宽范围检测与识别具有相当广泛的应用前景。本文提出一种新的方法从双目动态图像序列来获得人体身高肩宽范围。文中主要论述了该方法中对人体三维特征的重建部分,在无手工标注实现摄像机标定的基础上,应用人体特征三角形在双目摄像机投影下的几何意义,结合人体三维模型知识重建三维信息,决策身高肩宽范围。最后给出了实验结果及其在WMFVR(加权多特征视频识别)中的应用。
In the domain of computer vision and pattern recognition, there is a very wide application for the study of detection and recognition of video stature and shoulder-width. This paper proposes a new approach to obtain the stature and shoulder-width of human based on binocular image sequence. It focuses on the three dimensional human feature reconstructions. The stature and shoulder-width result is obtained by using geometry of feature-triangle binocular projection, three dimensional human model knowledge and camera auto calibration without being labeled. Experiment shows that the algorithm can allocate position of human face fast, which makes subsequent identity recognition robust and fast. The paper also gives its application on identity recognition based on Weight-Multi-Feature-Video-Recognition（WMFVR） model.
出处
《光电工程》
EI
CAS
CSCD
北大核心
2006年第7期105-109,共5页
Opto-Electronic Engineering
关键词
摄像机标定
三维重建
视频身高检测
加权多特征身份识别
Camera calibration
Three dimensional reconstruction
Detection of video stature
Weight-multi-feature-video-recognition.
