Using our powerful video analytics algorithm on CCTV footage, aggression in crowds and large spaces can automatically be detected with high accuracy. Alerts are automatically generated as soon as aggression is detected, allowing for a direct response to prevent the situation from escalating. Around 99% of all video that is recorded by CCTV cameras is currently not being watched. Human operators can only monitor 3-4 streams at the same time and will lose focus after 30-40 minutes. Because of this, incidents like fights are not immediately noticed and can only be acted on after a bystander brings the situation to attention with the relevant parties (e.g. security/police/medical staff/etc.). Using our algorithm, an operator (in a control room or in the field) will only have to react on specific alerts and video clips that the system has classified as aggression. This way, one person can virtually ‘monitor’ dozens or even hundreds of CCTV cameras. Acts of aggression will be immediately detected and a swift course of action can be effected. After the situation has resolved, video footage of the event will be easily available and no longer requires intricate forensic searching of old recordings.
The technology consists of a software algorithm thatcan work with any IP(Internet Protocol)-based camera (other cameras after some adjustments). It has been tested on video footage of 12 CCTV cameras in a Dutch city, providing an extensive dataset including 22 acts of aggression and 150 normal situations. The algorithm was able to detect 91% of all aggression acts with only 4 false positives. Furthermore, only a limited amount of processing power is required, making it possible to have these analytics running 24/7 with limited investments in computational resources. The algorithm can be deployed on new and existing CCTV systems at a large scale.
This algorithm can be embedded in new and existing CCTV systems, operated by: Police Security firms Event organizers (Mental) healthcare sector Malls Schools Sport venues Etc. Alerts are generated immediately upon detection of aggression. These alerts, together with a brief video clip, are then instantly presented to an operator (e.g. in a control room, or on the mobile device of a nurse, etc.) for him/her to assess the situation and react accordingly.
There are no commercially available aggression detection algorithms yet, so detection of aggression currently relies either on expensive human operators in a control centre, or more commonly on witnesses calling it in. In both cases, the time of detection might be delayed, which could lead to unnecessary escalation. There is also a large probability of the event passing entirely undetected. This algorithm provides an accurate, real-time detection at low operational cost, enabling the transition from reactive surveillance to pro-active surveillance.