Abstract: To prevent pedestrian safety accidents and optimize and help crowd security management in public places, the collection region distributions of crowd are studied at three common multidirectional flow intersections of 90°, 120° and single corridor two-way pedestrian flow using the density-based spatial clustering of applications with noise (DBSCN) algorithms at different times based on a dataset from scenario experiments. Moreover, the clustering effect and parameter difference of the two algorithms are compared. Study results show the collection region distribution patterns and characteristic changes of the crowd. The clusters cannot remain in a stable clustering state but vary dynamically in 3 subject scenarios. The proposed method can be used to identify potential locations of crowd, observe hidden danger areas, provide evacuation facilities in advance in such areas, and protect the entire section to improve the efficiency and safety of the evacuation process.

Key words: traffic safety, multidirectional pedestrian flow, local collection region distribution, clustering, density peak algorithm, DBSCAN algorithm