Video-based pedestrian grouping model considering long-span space in a big hall

Abstract

Pedestrian group detection is a challenging but significant issue in pedestrian flow control and public safety management. To address the issue that most conventional pedestrian grouping models (PGMs) can only identify a pedestrian group at a limited distance of less than 2 m, this study extended the pedestrian distance constraint of conventional PGMs with a reconstruction of the normal group detection criterion and development of a novel group detection criterion suitable for long-span space. To measure the movement behavior similarity with normal distance, five necessary constraints: velocity difference, moving direction offset, distance limitation, distance fluctuation, and group-keeping duration were studied quantitatively to form the criterion to detect normal groups. Meanwhile, a long-span group detection criterion was proposed with extended distance and direction consistency constraints. Therefore, this study proposed an improved PGM that considers long-span spaces (PGMLS). In the PGMLS workflow, the MMTrack algorithm was used to obtain pedestrian trajectories. A difference measurement method based on sequential pattern analysis (SPA) was adopted to analyze the velocity similarity of pedestrians. To validate the proposed grouping model, experiments based on pedestrian movement videos in the exit hall of the Shanghai Hongqiao International Airport were conducted. The results indicate that the proposed model can detect both normal and widely separated pedestrian groups, with a long span range of 2–12 m.