Depth-based person re-identification

Person re-identification aims to match people across non-overlapping camera views. For this purpose, most works exploit appearance cues, assuming that the color of clothes is discriminative in short term. However, when people appear in extreme illumination or change clothes, appearance-based methods tend to fail. Fortunately, depth images provide more invariant body shape and skeleton information regardless of illumination and color, but only a few depth-based methods have been developed so far. In this paper, we propose a covariance-based rotation invariant 3D descriptor called Eigen-depth to describe pedestrian body shape and the property of rotation invariance is proven in theory. It is also insensitive to slight shape change and invariant to color change and background. We combine our descriptor with skeleton-based feature to get a complete representation of human body. The effectiveness is validated on RGBD-ID and BIWIRGBD-ID datasets.