An Improved Dynamic Pedestrian Grouping Model in Public Transport Space

Dynamic pedestrian grouping is a significant issue in the field of crowd flow control, also a research hotspot in the field of computer vision and artificial intelligence. While it is difficult for a conventional pedestrian grouping model to identify and group the pedestrians with a large span large than 2 m besides depending on the pedestrian velocity modulus and direction. As the pedestrians between large-span possibly have the same movement goals and behavioural motivations, it is necessary to solve the grouping problems. This paper analyses the characteristics of pedestrian velocity vector, large span in a public transport space, then introduces the non-recursive motion similarity clustering (NMSC) algorithm. An improved dynamic pedestrian grouping model is proposed. The connectivity weight graph is designed in the pedestrian clustering. Measurement of differences based on a hierarchical pattern is implemented to analyse the velocity similarity of pedestrians. In addition, five grouping conditions are used to determine the dynamic weight of the pedestrians. Experimental results validate that the improved dynamic grouping model has a good effect on the recognition of regular pedestrian groups and the special groups considering the large-span space.