System automatically detects threat levels originating from hundreds of video streams (CCTV cameras, dash cam / drone footage etc). Algorithm weighs threat levels according to object & scene classification, quantity, velocity and other factors. User interface automatically alerts human operator and zooms into streams requiring immediate attention. System increases situational awareness, reduces reaction time, reduces human labour and hardware requirements in both real-time surveillance and historical footage analysis.

Our system enables better Situational Awareness. This is achieved through (1) artificial experts who monitor multiple live video feeds in real-time and (2) automatic triage system ranking and weighing all incoming signals. This allows one human operator to handle hundreds of live imaging feeds, reducing cognitive burden and improving reaction time.

Figure 1: Schematic representation of Computer Vision / AI triage system architecture
  • Web-based monitoring dashboard – works on any laptop or desktop PC
  • Compatibility – system transcodes multiple video codecs on the fly
  • Triage logic – operator receives the most important signals
  • Audio and visual alerts. Auto zoom and re-focus
  • Pattern of life detector – monitors changes in Fluidity and Occupancy levels
  • Object detector – small arms and assault weapons, explosions etc
  • Scene detector – scene anomalies: fires, explosions, fast crowd movement etc

Figure 2: Web interface of Computer Vision / AI triage system

Demo shows an example with 4 video feeds in different phases:

  1. Normal phase – system does not detect any threats
  2. Threat detected at CAM2 (anomaly in pedestrian and traffic behaviour due to earthquake) – system issues visual and audio alert and zooms in automatically to CAM2.
  3. While CAM2 is in “zoom in” mode system continues to automatically analyse and triage potential threats in all available video streams.
  4. CAM4 detects a higher threat (objects of ‘weapon’ classification detected) and system automatically zooms into CAM4.
  5. After threat factors are removed, system automatically goes back into surveillance phase zooming out or switching between different views.

The above mechanism works with virtually any number (hundreds) of real-time video streams from CCTV, hand-held, wearable, UAV / UxV surveillance cameras. System will analyse signals behind the scenes and automatically alert operator, zooming in to the channel of highest importance.

Video demo:

Video 1: Video demonstration of Computer Vision / AI threats triage system

There are two types of AI “expert agents” implemented in the system currently:

  • deep learning (artificial neural network based) – used mainly for object detection
  • statistical – used for scene detection and anomaly detection

Example: statistical analysis takes into account deviations in Occupancy and Fluidity values and the difference can be clearly seen in diagrams when comparing “normal” scene of road traffic flowing and pedestrians walking with “anomaly” situation when earthquake has begun, impacting the fluidity and occupancy levels.

Figure 3: illustration of anomaly scene detection using statistical analysis of Fluidity and Occupancy levels

To get in touch with us regarding this product and relevant technologies please use our contact form.