An image-processing system for pedestrian data collection

The aim of this research project was to investigate the software techniques which can be used to obtain information on pedestrian movements automatically using image processing technology. To fulfil this aim, two separate algorithms have been developed. One is designed to give a measure of the pedestrian density within a crowd scene and the second to count and determine the walking direction of individual pedestrians. The problems presented by a scene of pedestrians include high levels of occlusion, variability in the shapes of pedestrians with time, and the unpredictable nature of changes in walking speed and direction, especially in crowded scenes. To overcome these problems a number of assumptions have been made about the scenes being analysed and the nature of pedestrian movement. The pedestrian monitoring functions investigated in this project illustrate the feasibility of automatic pedestrian data collection. Real-time performance has been achieved for the density measuring routine using a small network of three transputers operating in parallel