Motivation

With the increase of security level at public infrastructures, surveillance camera networks having wide baseline setting (including the case of non overlapping fields of view) have received more attention. Driven by cost, surveillance cameras are in most cases distributed and placed in such a way that an observed area is maximised resulting in a very tiny or no overlap of camera fields of view. In some cases, the fields of view overlap but the camera viewpoints are such that the people are seen from totally different views. Many efforts have been done on the Computer Vision area to develop algorithms to track people, mainly using a single camera. However, aforementioned camera systems using Computer Vision algorithms still remain a challenge.
We are interested in going a step further and evaluate the performance of possible solutions for tracking people in such camera systems.

Tracking Network

Figure 1. The goal: People tracking in the camera network.
Left: four cameras (1‐4) with slightly overlapping fields of view.
Right: cameras are localised in a coordinate frame by tracking; top view is shown. Colours depict available observations in the particular cameras. This information can be used to discover the camera network topology.