Analysis of crowd behaviour by various methods of image processing

The protection and rescue of people attending big events, e.g. sporting events, concerts, fun fairs, or at least crowded pedestrian areas, requires early recognition of a developing incident. This implies the analysis of ""normal"" behaviour and the response of persons on special situations like encountering other persons or barriers, increasing densities or passing bottlenecks. The derived results should enable simulations to emulate the behaviour of pedestrians in various situations. The analysis uses varied techniques of image processing like correlation methods, difference techniques and image compression similar to those used in engineering application of fluid dynamics or material research. These techniques were applied to real videos of several events like football matches, fun fairs and crossroads. The analysis focuses on situations specified above. Under normal conditions persons move with constant speed and slow unwillingly down, only if forced by strongly increased densities or appearing obstacles. People try to avoid direct face to face encounters at an early stage. Quantitative results are described and deviations from mean values discussed.