System Wide Tracking of Individuals

Abstract

Tracking the movements of people around surveillance systems is becoming increasingly important in the current security conscious environment. This chapter discusses recent research to assist with tracking the movement of individuals throughout the full surveillance area, even when those views do not overlap and may be separated by significant distances. This is due to advances in tracking within a camera view, which allows feature models to be built of an individual. The feature models can include a range of shape, appearance, and transition features appropriate for use with surveillance footage. These may also utilise spatial information about the objects from both the real world and image planes to increase individual discriminitivity. These feature models can be combined and compared within a statistical framework to determine the probability of any two tracks being made by the same individual. It is this combination of such tracks that allows for system wide tracking of individuals to be performed. Techniques are also presented to improve the accuracy of the feature information. These include the application of spatial or temporal smoothing, the identifcation and removal of signifcant feature errors, as well as the mitigation of other potential error sources like illumination. The results of both the individual features, and the combined system wide tracks are presented based upon an analysis of individuals observed in realistic surveillance footage. These show that current camera technology will not produce a fully automated system, but can provide signifcant information to assist security operators.