The problems posed by crowds arise from their collective properties: congestion, excitement, fighting, rioting and mass panic. In his classic research on social psychology, Gustav Le Bon wrote: “an agglomeration of men presents new characteristics very different from those of the individuals composing it.” Contagion theory describes the infectious nature of human emotions, which can result in collective behaviour that poses significant threats to safety and security.
Closed circuit television (CCTV) provides a means for security personnel to monitor crowds for safety and operational purposes. Unfortunately, as CCTV becomes ubiquitous, it becomes impossible for humans to monitor all available viewpoints. It is estimated that there are between 1.85 million and 4.2 million security cameras installed in the United Kingdom alone. Furthermore, crowd monitoring may be limited by the “short attention span and lack of adequate training and experience of human operators”. In most cases, security footage is used to investigate events after they occur, rather than to generate real-time alerts during an evolving situation.
In recent years, researchers have turned to computer vision technologies to monitor crowds automatically from CCTV footage. This is an active field of research due to the large number
of problems which remain unsolved. The challenges include occlusion among a large number of individuals, as well as environmental changes, such as lighting fluctuations and changes in
context over time.
The aims and objectives of this project include:
Crowd counting can be used to estimate the size of a crowd, which is the most common indicator of abnormality. This project aims to estimate the crowd size by utilising local features in an image. The number of people within each group is calculated, as well as the entire scene. Our system has been tested on crowds of more than 200 people, as seen in the videos below.
Crowd flow can be used to determine the direction of crowd motion for operational purposes. This project aims to estimate the number of pedestrians passing through a virtual gate or turnstile using computer vision. This can be combined with crowd counting to monitor queue properties, such as throughput rate and estimated wait time, in airports and malls.
Anomaly Detection is used to detect abnormal patterns of motion. For example, cyclists or vehicles on a walkway or pedestrians entering restricted areas should be flagged to system operators. This can be used to detect an abnormal event as it occurs.
Demonstrations
The outcomes of this research project are highlighted in the following video:
http://www.youtube.com/watch?v=lTkfcOReg4U
Crowd counting can also be employed on a massive scale. The following sequence is taken from New York’s Grand Central Station, and features more than 200 pedestrians:
PhD student
