Options
2009
Conference Paper
Titel
People tracking and identification using laser features and colour distributions
Abstract
Tracking multiple crossing people is a great challenge, since common algorithms tend to loose some of the persons or to interchange their identities when they get close to each other and split up again. In several consecutive papers it was possible to develop an algorithm using data from laser range scanners which is able to track an arbitrary number of crossing people without any loss of track. In this paper we address the problem of rediscovering the identities of the persons after a crossing. Therefore, a system of two cameras is used. An infrared camera detects the people in the observation area and then a charge-coupled device (CCD) camera is used to extract the colour information about those people. The colour information is represented by colour histograms, which are computed within the HSV colour space. Before the crossing the system learns the parameters of a Dirichlet distribution for each person. After the crossing the system relocates the identities by comparing the actually measured colour distributions with the distributions, which have been learnt before the crossing. The most probably assignment of the identities is then found using Munkres' Hungarian algorithm. It is demonstrated using data from real world experiments that our approach can reliably reassign the identities of the tracked persons after a crossing.