Automatic human head location for pedestrian counting

Accurate, real time data relating to the flow of pedestrians in open spaces is useful for the effective management of those same open spaces. Currently, most measurement algorithms are based on motion analysis. Although motion analysis can be practically simple, it is not suitable in many complicated situations, such as counting people on an escalator or a moving walkway, something which is demanded by many end users. The work here presents a new method for this application of counting pedestrians against a dynamic background using image processing techniques. The work presented here represents a significant advancement over work previously published. A model-specified directional filter (MDF) is used to detect object candidate locations followed by a novel matching process to identify the pedestrian head positions in an image even with complicated contents.