Citation

Tange M, Imanishi M, Sano T, Ohmiya Y. 日本建築学会計画系論文集 :: Journal of Architecture and Planning (Transactions of AIJ) 2016; 81(730): 2645-2652.

Vernacular Title

複数色マーカーを用いた多人数歩行者群の画像解析と群集避難実験への応用

Copyright

(Copyright © 2016)

DOI

10.3130/aija.81.2645

PMID

unavailable

Abstract

Reasonable guidelines based on actual evacuee's behavior lead to fine estimations of evacuation time and advanced designs of evacuation passages. Actual evacuee's behavior, however, depends on various factors such as passage width and crowd density and it is difficult to predict the crowd characteristics in real situations. Recent studies have described each pedestrian's reaction to his circumstance as a mathematical model and simulated the crowd evacuations using these models to reproduce the real evacuation in an actual space with complicated configurations. The validation and refinement of these models require detailed information of real evacuee's behavior. Image processing technique of video image of evacuation experiments is an effective way to extract pedestrian movement. Some of the early works, however, include the error of position measurement due to body height differences between pedestrians and require frequent manual operations to recover the loss of pedestrians. This study conducted a large-scale crowd evacuation experiment with 96 pedestrians walking in several fundamental passages or configurations: a straight path, an O-shaped path, an L-shaped corridor, a straight path with a bottleneck, an opening in a wide space, and merging to a path. The each layout emulates a part of real evacuation situation to examine the evacuee behavior. This paper proposes a novel image processing method to extract pedestrian movements in a crowd from video images of the experiment seen from above.   Each pedestrian wore a cap with two different color labels (color markers) indicating the pedestrian's body height. These multiple markers also have redundancy to find the pedestrian position. This method detects color markers and makes trajectories of pedestrians by clustering of neighboring markers and tracking them through the sequential video frames. The color combinations of clustered markers indicate the information of the body height of each pedestrian to obtain more accurate coordinate transform.  This method successfully tracked the pedestrians to make their trajectories. The positioning error was estimated as under 40 mm and the preliminary experiment supported that. This paper shows one of the results of the merging experiment. Over 99 percent of pedestrian's position data was automatically calculated without manual operation thanks to the redundancy of multiple markers and the marker clustering. Ordinary statistical quantities such as flow rate can be calculated from the data. In the merging experiment, the flow rate through an opening decreased due to turning after the opening and merging with pedestrians in the path. Detailed pedestrian behavior revealed that the walking direction and the walking speed were strongly linked to the interaction between other pedestrians such as distance and the relative velocity between pedestrians. The analysis of each experiment will be reported in future articles.  This method has an advantage in robustness against the pedestrian position loss. This method may have applicability to actual situations such as experiments of evacuation from theaters and stadiums.

Language: ja

Keywords

Image processing; Pedestrians; Pedestrian movement; Pedestrian flow; Trajectory; Evacuation; Color; Object markers