An AI-based video analytics system analyzes video streams in real time: it recognizes fights, weapons, falls, abandoned objects and anomalous behavior (Re-ID), calculates a risk score and alerts the operator. Runs on your existing cameras.
One operator physically cannot watch all streams continuously at the same time.
Concentration during monotonous monitoring falls after just 20–30 minutes of a shift.
Incidents are usually found in the archive during review, when nothing can be done about them anymore.
All events look equal: without a risk score, the critical one drowns in a stream of routine alerts.
The system ingests video streams and events from multiple sources, analyzes the behavior of observed objects, detects deviations from the norm and assigns a risk level. The operator receives an incident with a recommendation — the history is stored for analytics and reporting.
Cameras, archive, sensors, access logs, external security systems.
Object detection and tracking, noise alert filtering.
Models determine what is happening in the frame.
Comparison with the normal behavior profile, risk calculation.
Operator alert, recommendation, log and reporting.
Figures are based on delivered projects; the actual effect depends on the site, video stream quality and the agreed recognition scenarios.
We deploy video analytics systems on operating sites — from schools and university campuses to public transport and industrial facilities. Every project starts with agreeing the target recognition scenarios and ends with a system running 24/7 on existing cameras. Below are real scenarios from our practice.
Fights, weapons, falls and climbing over the perimeter fence. An incident with a video clip goes to school security the moment it happens, not during archive review.
The system remembers and re-identifies a person across cameras, stations and routes: abnormally long presence, repeated appearances and movements of a group across the monitored area.
Person-down detection, presence in a restricted zone and movement against the vehicle route on the shop floor. A signal to the dispatcher and the occupational safety log — before the situation becomes an accident.
Abandoned objects, crowd gatherings, conflicts and unusual night-time activity across the campus and dormitories. A single incident queue for the duty shift.
Not quite. Deviant Detect is a platform we adapt to your site: target recognition scenarios, critical zones, rules and the risk score are agreed with the client before launch. These define the model set and the alerting logic.
No. The system does not replace the decision of a responsible security officer — it acts as a tool for early detection, event prioritization and operator decision support.
Yes. Deployment is possible on your existing video surveillance infrastructure — provided the video streams are of sufficient quality and the target recognition scenarios are agreed in advance.
Fights and hostile interaction, sharp movements and aggressive gesturing, weapons and dangerous objects, falls, stalking, abandoned objects, anomalous presence of a person or group (Re-ID), movement against the usual route, crowd gatherings and unusual activity for the time of day.
An aggregated indicator the system calculates for every deviation: behavior type, confirmation by multiple models, event duration and intensity, presence of a dangerous object, number of objects involved, zone criticality and time of day. Incidents in the operator's queue are prioritized by this score.
Before analysis, the stream goes through preprocessing: object tracking between frames and noise alert filtering. A deviation is confirmed by multiple models and compared with the normal behavior profile for that place, time and load — isolated noise signals do not become incidents.
Alerts are generated in real time: the system creates an incident, assigns a risk level, shows it to the operator with a recommendation and stores the video clip in the log for review and reporting.
The system is deployed inside the client's infrastructure and works with the video streams of that infrastructure. The scenario set (including Re-ID) is agreed before launch, taking into account legal requirements and internal security policies.
We'll review your cameras and zones, agree the target recognition scenarios and a pilot scheme.
I-SOL develops and deploys the intelligent video analytics system Deviant Detect, which turns existing video surveillance into an early threat detection tool. Real-time AI analysis of CCTV cameras: fight and conflict recognition, weapon detection on video, person-down detection, abandoned objects and suspicious behavior.
For every deviation the system calculates a risk score: behavior type, confirmation by multiple models, event duration, zone criticality and time of day. The incident, with a video clip and a recommendation, goes into the operator's queue, and the history is stored for analytics and reporting.
AI-powered smart video surveillance is used in schools and university campuses, at industrial enterprises, in public transport, shopping malls and offices with elevated security requirements. Cross-camera tracking and Re-ID are supported — re-identification of a person or group within the monitored area.
Video analytics can be deployed on existing CCTV cameras with sufficient video stream quality — no equipment replacement needed. The system does not replace the decision of a security officer — it speeds up detection, prioritization and response. Describe your site — we'll agree the scenarios and propose a pilot scheme.
