Virtual Boundary Crossing Detection without Explicit Object Tracking

Virtual boundary (tripwire) crossing detection is an essential component in almost all modern digital visual surveillance systems. In this paper, we address the problem of achieving reliable tripwire crossing results in crowded scenarios and propose a new technique for the replacement of conventional tracking based tripwire techniques. We introduce the concept of "ground patches" which are a set of sub-regions sampled around the defined tripwire. Each ground patch is declared active when it is occupied by one or more foreground objects. Temporal and appearance features are extracted from each of the patches, and they are correlated across the virtual boundary to determine if a crossing occurs. The proposed method has been applied in situations with different traffic loads, and promising results are obtained.