A PCA-based automated method for determination of human body orientation

Human body orientation is a fundamental preprocessing task in surface registration, motion analysis, and data driven animation for human body. Previous human body orientation determination methods suffer from pose dependence and non-robustness. In this study, we propose an automated pose-independent approach to human body orientation by combining geometric characteristics of human body based on principal component analysis (PCA). We first analyze the relations between the bones of lower limbs and the body orientation, and obtain a feature vector consisting of the angles between the bones and the body orientation. We then use PCA to analyze the training samples of the feature vector and construct a classifier based on the first principal component to determine the human body orientation. Our experimental results show that the linear separability of the feature vector is perfect, and the classifier trained by PCA can be used for orienting human body efficiently in a pose-independent manner. This method can also be easily integrated into existing approaches that extract skeleton from human body shape.