Chapter Localizing and Visualizing the Degree of People Crowding with an Omnidirectional Camera by Different Times

Abstract

The Corona Disaster increased the demand for information on the degree of human crowding, as it was essential to balance avoiding restricting behavior and reducing the risk of crowding. Although there are many technologies for detecting people using monitoring cameras, the number of cameras installed in a wide area is costly, and coverage is limited. In this study, we propose a method to qualitatively visualize the distribution of people by using images captured by a moving omnidirectional camera from the viewpoint of facility management during regular security patrols. Omnidirectional images are used for both 3D modeling of the target space based on SfM (structure from motion) and person detection/tracking by machine learning. The distribution of people is visualized qualitatively by obtaining the positions of the extracted people on the 3D model of the site and mapping them. The parallel software processing of visitor observation and mapping is expected to be highly cost-effective in terms of implementation and operation. On the other hand, although there are time deviations in the mapping depending on the location, the visualization and the updated time show their usefulness in understanding the distribution of congestion