Hierarchical Grid-Based People Tracking with Multi-camera Setup

Abstract

We present a hierarchical grid-based tracking methodology for multiple people tracking in a multi-camera setup. In this system, frame-by-frame detection is performed by means of hierarchical likelihood grids, by matching shape templates through an oriented distance transform over foreground intensity edges, followed by clustering in pose-space. Subsequently, multi-target tracking is achieved by means of global nearest neighbor data association, with a fully automatic initialization, maintainance and termination strategy. We demonstrate our system through experiments in indoor sequences, using a four-camera calibrated setup. Moreover, in the paper we present the improvements obtained by means of a fast algorithm for computing the oriented DT, as well as using multi-part shape templates in place of a simple cylinder model, for a more precise localization. © Springer-Verlag Berlin Heidelberg 2013.