Spatial-temporal Perception Methods For Regional Crowd Status And Behavior

The crowd in public gathering places has the characteristics of high concentration and high mobility. This constitutes a dynamic and uncertain geographical scene. High density crowds often lead to stampede and other public emergencies, and caused massive casualties, huge economic losses and adverse social impacts. Emergency management is the core issue of ensuring public safety. So, fast perception and monitoring of public emergencies is a hot issue of public safety and security at home and abroad.With the development of society, the research and application of video surveillance technology has become a national strategy for governments around the world. And give strong support from the policy, legal, economic and other aspects. However, the current intelligent processing level of mass emergency perception and monitoring based on video monitoring system needs to be further improved. Such as intelligent video analysis, multi-sensor cooperative work, abnormal event detection, especially video analysis and behavior perception in geographical environment. It is imperative to develop an efficient and accurate perception technology with the integration of video analysis and geographical environment to meet the management of public emergency. Consequently, in order to break through the key scientific issues of perception and monitoring for the regional crowds status and behavior in geographical environment, we combined video data with GIS data and proposed the crowd status and behavior monitoring methods. The contributions are as follows.(1) A scene invariant method for crowd density estimation is proposed. Current crowd density estimation methods based on video analysis are scene-dependent. An estimation model acquired with video data taken by one particular camera cannot be adaptively applied to data taken by other cameras. This paper presented a scene invariant crowd density estimation method using GIS to monitor crowd size for large areas. The proposed method mapped crowd images to GIS. Then we can estimate crowd density for each camera in GIS using an estimation model obtained by one camera. It does not require additional training when deployed for crowd density estimation on a new camera. This greatly improves the efficiency of crowd density estimation modeling.(2) We can analyze the crowd behavior in geographical environment using the method we proposed. The traditional crowd motion analysis methods are based on image space. So we cannot get the true movement state in the real world. For real-time monitoring crowd behavior in geographical environment, this paper presents crowd behavior pattern analysis methods using GIS. The measured crowd motion vector field can be calculated by crowd images in GIS. Then we can get the crowd motion pattern, crowd motion trend and crowd motion velocity by analyzing the crowd motion vector field.(3) We proposed a crowd for abnormal behavior detection based on the geo-referenced crowd motion vector field. This method can be used to detect crowd abnormal behaviors, such as centers gathering, centers divergence, motion trend mutations, motion velocity mutations and reverse walking et al. And then we can further analyze the spatial-temporal hotspots in the monitoring area, and provide the basis for the prevention of emergencies.(4) Surveillance cameras installed in the monitoring areas are sparse, discrete and non-overlapping. So the crowd status information in the areas with no camera cannot be captured. In this paper, we create the crowd status inference model for the areas with no camera using Bayesian network. Then the spatial pattern of the regional crowd status can be predicted by the sparse crowd status data using the inference model.(5) We analyzed the spatial-temporal evolution process for the multi-time crowd status spatial pattern in experimental area. Then we can get the crowds spatial-temporal distribution modes, and we can further analyze the formation mechanisms of the crowd status spatial-temporal pattern. This can provide basis for security personnel deployment, facilities planning, crowd management, business strategy and so on.(6) We designed and developed a system for the regional crowd status and behavior monitoring based on the above research results. And this system was applied to crowd monitoring in Nanjing Confucius Temple Pedestrian Street.