Camera calibration parameters for oriented person re-identification

Person re-identification is a challenging task when there exist strong appearance changes for different viewpoints of the person captured by a distributed camera network. To better solve this issue a multi-view oriented model of the person has been proposed. In this paper, we analyze the camera calibration parameters required to be used in the oriented people re-identification and propose a method to retrieve those values to be used in the capture of people perspectives with different known orientations respect to the camera. Usually, individual camera calibration parameters on a large distributed camera network are not available. A self-calibration method through the usage of short-term trackers of multiple persons is proposed. Only two extrinsic camera calibration parameters are required. Experimental results based on the processing of different public datasets demonstrate the effectiveness of our approach.