The path of crowd dominated-motion detection based on spatial autocorrelation
Wei-Lieh Hsu, Yucheng Wang, Kun-Feng Lin · 2016
Due to the ever increasing affordability of video surveillance equipment, surveillance cameras have become widely used to great effect at various large scale events, to reduce the risk of crowd-related incidents and enhance public safety. If a large amount of existing video data can be used to capture motion features and detect the distribution of a crowd within a monitored area over a long duration, and then be analyzed to identify the path of crowd dominated-motion as well as the spatial hot spot, the effects of the crowd flow can be more efficiently managed and congestion more effectively reduced. In this paper a grid model is constructed to describe crowd distribution within a measured area. The measured area is divided into several unit areas, and each unit area is considered as a simple cell in a grid model. The crowd motion within a measured area can be represented as a 2D status matrix. The status matrices of all the images are then accumulated and normalized over a period of long duration in order to detect the path of the crowd dominant motion. All element values in the cumulative status matrix are limited to between 0 and 1 after normalization. If the value approaches 1, it indicates that the crowd remained longer in the corresponding unit area. On the other hand, a lower status value indicates that people spent less time in that corresponding unit area. In order to determine the aggregate area occupied by the crowd, a global spatial autocorrelation in the spatial statistics is adopted to determine whether the crowd distribution in an area is one of dispersion or aggregation distribution. If the distribution is dispersed, then the path of crowd dominant motion should be distributed throughout the measured space. If the crowd distribution is aggregated, then the local spatial autocorrelation of each observation unit must be calculated based on the observation value and the LISA (Local Indicators of Spatial Association) of each observation unit in the cumulative status matrix. This will obtain the degree of occupation by the crowd of each observation unit and its surrounding units. Based on the observation value and the local spatial autocorrelation of each observation unit, the path of crowd dominant motion can be efficiently determined. This proposed method supplies valuable spatial information for space planning and crowd management, and is applicable to public spaces for safety surveillance and the layout planning of exhibitions.