Panoramic Appearance Maps for Human Body Representation and Analysis

Multiple overlapping cameras help to provide robust detection as well as 3-D localization of people in their field of view [2], cover their features from all directions, and handles the problems of occlusion. We propose a novel representation called Panoramic Appearance Map (PAM) for combining appearance information from multiple cameras tracking the same person. The map is a discretized 2-D array with horizontal axis representing the azimuth angle around the body center and the vertical axis representing the height above the ground. Each bin in the map contains appearance information such as color or texture from the cameras which can observe it, as well as a weighting factor that denotes the reliability of the information. The human body is modeled as a convex generalized cylinder shown in Figure 1. A point p on its surface is parameterized by the azimuth angle θ and height h as: