People reidentification techniques for multiple viewpoints camera networks

We present a novel approach for performing people re-identification using feature descriptors from joint locations of the human body without utilizing RGB-D sensors. To obtain local keypoint coordinates we implemented a skeleton tracker built on top of a state-of-the-art pose detector and augmented its output to work with multiple input cameras. A signature is then extracted from each detection by gathering descriptors on keypoints, and is matched with a database to recognize a target person