Thread Networks

Blog

Video Analytics

Simply put, a typical video surveillance system is a number of cameras deployed throughout an area with a network that aggregates the camera feeds to a command center.  The feeds can be monitored in real time from the command center and mobile devices.  Video surveillance systems provide situational awareness over large areas.  Traditional video surveillance systems also leave the burden of watching video, detecting threats, and locating persons to the human operator.  The process of manually watching multiple feeds is known to be monotonous, ineffective [1], and expensive.

Intelligent video surveillance systems actively “watch” live video and provide alerts and content-based search capabilities.  The software algorithms that analyze the video and provide alerts are commonly referred to as video analytics.  It’s not a new idea.  In fact, it’s what makes consumer cameras like Ring and Nest so popular.  Consumer camera analytics detect objects, record movement, and alert the user.

Commercial video surveillance analytics are much more sophisticated.  A security team can create parameters (or rules) for each camera.  When an object in a camera’s view violates a rule, an alert is generated.  They act as intelligent filters directing the watcher’s attention to cameras observing events that are of interest as well as tracking subjects through a large structure [2].  Analytics today can detect fire, a person loitering in a defined area, someone crossing a boundary or breaching a perimeter, identify abandoned objects, and identify when an object has been removed (theft).  Analytics can analyze stored video and tag each frame with recognized objects (red shirt, white car, blue truck, green hat) aiding in post hoc investigations.

Practical examples:

·        Cameras in main entry ignore usual foot traffic but alert when someone crosses a boundary leading to a seldom-used stairwell

·        All interior cameras will generate an alert if an object is left abandoned for a set period

·        Cameras on perimeter of building ignore people entering and leaving during expected times but alerts to an individual loitering outside an entrance for 10 minutes after hours

·        Cameras observing playground area alert when a person enters or leaves the perimeter

·        After a suspicious person has been identified, security can quickly search through days or weeks of video to identify where the individual went and if they’ve been on-site previously to perform reconnaissance [3].

Analytics aren’t just for tracking bad actors.  How about a lost child?  Instead of manually backtracking through video to the last place they were seen and then following them through each camera view, analytics can be shown the child once, then find them on every video feed it has available [4].

Analytics may conjure visions of Big Brother and The Terminator but will ultimately prove necessary to provide near real-time response to events and enable security teams to effectively monitor an increasing landscape of cameras.

 

References:

[1] M.W. Green, “The appropriate and effective use of security in schools,” Us Dept. of Justice, Rep. NJC178265, Sept. 1999.

[2] Adams, Andrew A., and James M. Ferryman. "The future of video analytics for surveillance and its ethical implications." Security Journal 28.3 (2015): 272-289.

[3] Findlen, Brian. “Turning the Tables: Intelligent Video Analytics in 21st Century Policing.” Policechiefmagazine.org, California Commission on Peace Officer Standards and Training, www.policechiefmagazine.org/turning-the-tables/.

[4] Dees, Tim. “How Deep Learning Is Transforming Police Investigations.” PoliceOne, 6 July 2018, www.policeone.com/police-products/police-technology/software/video-analysis/articles/476485006-How-deep-learning-is-transforming-police-investigations/.

Doug Potter