Body Orientation Estimation Using Template Matching Based on Shape and Skin Color Information

In real world, body orientation is a problem of person identification. The variation of orientation makes features hard to be extracted in a uniform form. If the orientation is estimated, we can select suitable features based on orientation. It may improve the performance of person identification. For the reason, we propose a body orientation estimation method in this paper. First, body region in images is extracted by background subtraction technology. Template matching is adopted to estimate the body orientation. Because the head and trunk may face the different orientations, we construct the templates with the images of head and shoulder region. In addition to contours, shapes and skin color areas are reserved to provide more orientation information for templates. In the experiments, the average accuracy for detecting frontal, left, right, and back orientation is higher than 97.8%. It explains that shape and skin color information is credible to describe the orientations. Besides, the body orientation estimation scheme works in real-time.