Tracking objects across uncalibrated, arbitrary topology camera networks
Richard Bowden, Andrew Gilbert, Pakorn Kaewtrakulpong · Institution of Engineering and Technology eBooks · 2006
In this chapter, an approach for recognising targets after occlusion is proposed. It is based on salient reappearance periods discovered from long-term data. By detecting and relating main paths from different regions and using a robust estimate of noise, salient reappearance periods can be detected with high signal-to-noise ratios. Off-line recognition is performed to demonstrate the use of this extracted salient reappearance period and the appearance model to associate and track targets between spatially separated regions. The demonstration is extended to regions between spatially separated views with minimal modifications. As the underlying process of reappearance is not the salient reappearance time but the average distance between paths, the performance of this recognising process is degraded if the average distance between paths is increased. These issues need further investigation.