The ‘classical’ video analytics like detecting loitering and unattended baggage are not a good fit for a crowded scene. Masses of people standing and moving at one point near a race track and large numbers of bags in the scene will create too many alarms to make this solution practical.
Role of Video Analytics
The ‘classical' video analytics like detecting loitering and unattended baggage are not a good fit for a crowded scene. Masses of people standing and moving at one point near a race track and large numbers of bags in the scene will create too many alarms to make this solution practical.
Other analytics, such as face capturing and face recognition are also impractical in an outdoor event where there is no way to control the environment and make sure the cameras have a good view of people's faces.
“Video analytics tools are typically used for smaller crowds and are used in such instances as in prison yards, bars and other situations where there are a limited number of objects/people to be identified and analyzed. In such scenarios, the video and analytics are expecting to capture something specific – a certain face, a license plate image, a certain traffic pattern in the crowd or to see if a crowd is dispersing. Larger crowds/more objects make facial recognition ineffective because the faces are too small and/or occluded” explained Ian Westmaccot, Computer Engineering Manager, Tyco Security Products.
However, some video analytics can solve the need for detecting overcrowding and detecting aggressive behavior, like a fight that takes place in a crowded environment. Such an event might escalate and cause a stampede.
“In some instances you know that there is a very high likelihood of a crowd, such as in and around a football stadium on a match day. Other times you will have unexpected crowd build up, such a protest march, or disruption at an airport terminal. It is important to be able to plan for both eventualities” said Jamie Wilson, Security Marketing Manager for EMEA at NICE systems. “Video analytics is an important part of this process in being able to detect and flag early that there is an ‘unusual' congregation of people. Being able to display to the operator a live-feed of what is occurring and also the ability to roll-back and review what has triggered the incident is key to helping them to contain and control the situation” he added.
“Our new video analytics are now able to perform reliable detection and tracking under crowded urban environment. The system could perform real-time crowd size estimation; identify dominant patterns and statistics of the crowd, pick-up suspicious unknown and other hidden information” added Karianto Leman, Head, Situational Awareness Analytics Program, A*Star's Institute for Infocomm Research (I2R) in Singapore. Apart from crowd analytics, new video analytics are now able to very reliably detect suspicious human behavior such as violence, attacks, battering, breaking and entering etc. “I2R's aggression detection system had achieved beyond 90% accuracy in live trial at Singapore safe city project” stated Leman. Despite recent advances, video analytics are still very limited in a crowd-scenario. Crowd scenes are by nature detail-rich and this affects the reliability of video-analytics. According to Leman, a detection accuracy rate of 70% or more is a realistic expectation in this context.
Alternative surveillance cameras: Drones and Cell phones
Traditional pole-mounted cameras are still the most prevalent method of surveillance. However, other platforms have potential to become relevant in coming years. “Aerial surveillance using cameras on drones is gaining momentum in tandem with vast capability improvements and cost reduction of commercial drones” said Leman. “The challenge is in video transmission from drones to the ground. It could be costly over LTE and unreliable over WIFI. The solution is to enable edge computing on drone for video analytics. However, video analytics on video from drone is more challenging. It may require video stabilization and detection using different approaches from ground's video analytics” he explained.
Another potential camera, found in almost every pocket is the cell phone camera. “Mobile phones, in theory, have enough resolution to be useful in large crowd scenes. These mobile devices might play a larger role in surveillance once more cloud-based services become available and video management systems (VMS) are able to be used on mobile devices” added Steve Gorski, Chief Sales Officer at Scallop Imaging.
>>> Crowd control: securing outdoor events