Motion signals for provision of rapid discernment of pedestrians and pedestrian behaviour

Computationally efficient methods for identifying and analysing pedestrian motion in video are presented. All necessary data are accessed directly from the tracker´s motion detection stage. The methodology that has been developed uses three new invariants to rapidly discern pedestrians and pedestrian behaviour. Results show ~97% correct discrimination between cars and pedestrians, and ~ 90% correct classification of a number of categories of behaviour, including particularly walking and running.