Scylla logo
Get a quote

Anomaly Detection System

Get a quote

Proactive security requires the consistent monitoring of your environment to identify disruptive threats. Incidents can often be stopped before they occur with proper identification. Scylla Behavior Recognition allows security professional to identify disruptive threats and capture events as they occur. Increase the effectiveness of your security perimeter with Scylla.

Scylla Anomaly Detection & Behavior Recognition can effectively detect a range of anomalous events from your security cameras to make your environment proactive:

  • Fighting
  • Suspicious behavior that can result in shoplifting
  • Smoke & fire

Scylla uses state-of-the-art neural network architecture for behavior recognition collected from your security cameras. Our proprietary software uses machine learning (ML) to review collected data of frames and events to learn the standard security environment.

Scylla Behavior Recognition is optimized to analyze multiple video streams with the ability to analyze 24 hours of video within minutes. Scylla provides advanced real-time and forensic abilities to allow for real-time event and environment tracking and awareness.

Scylla Behavior Recognition can be utilized in public areas and environments with people and visitors. Scylla has been tested in public transportation, universities, casinos, shopping malls, and other locations to deter and provide awareness of threats. With un-paralleled accuracy and the ability to analyze frames within seconds, Scylla is the exact solution needed for security and surveillance professionals.

Get a quote


There are 3 submodules of Scylla Anomaly Detection and Behavior Recognition system each one of which is designed to detect a specific event, such as fight and violence detection module, smoke and fire detection module, and potential shoplifting detection module. All three Scylla Anomaly Detection and Behavior Recognition submodules analyse such events from CCTV video footages either in real time or forensically, from video databases. When a specific event is detected, the system initiates a corresponding alert and distributes it through predefined alerting pathways to the end user.
A sequence of frames is provided to the pre-trained neural network engine of Scylla Anomaly Detection and Behavior Recognition. Thus this 3D matrix (2D frames + time) is considered as an input. Scylla Anomaly Detection and Behavior Recognition AI then decides if the actions in this matrix correspond to the action sought after and triggers an alert if the probability is above the defined threshold.
The anomalies detected by Scylla Anomaly Detection and Behavior Recognition are based on the training set. In particular, the fight detection model is trained on a large real-world dataset of fights and assault cases recorded from CCTV cameras. Thus the anomaly it detects will include acts of violence including fighting and wrestling. Similarly, the smoke & fire detection submodule of Scylla Anomaly Detection and Behavior Recognition is triggered when these hazardous events are detected in the scenery.
The system accumulates a number of consecutive frames and analyses the whole batch as a unit. The duration of this batch is different for different submodules but on average the chunk duration is 3-5 seconds.
The maximal distance will depend on the camera characteristics. More specifically, on the lens. Typically, for proper illumination (400 lx and higher) the requirement is that the person height should take up 1/6th of the frame height. For most cameras this results in maximal detection distances of up to 15 meters. If optical or digital zoom is used or the camera has not standard aperture/focal range, the maximal distance can vary.
Absolutely. After all, what Scylla Anomaly Detection and Behavior Recognition needs is the video feed from CCTV which is provided by most CCTV networks and video management systems. Moreover, Scylla Anomaly Detection and Behavior Recognition can be two-way integrated with major VMS providers, such as Milestone, Genetec, NetworkOptix, and others.
The shoplifting detection model is trained on a large dataset of videos from surveillance cameras, where shoplifting events occur. The submodule is designed to detect an action of taking an item and trying to conceal it. The system is supposed to analyze the streams coming from the retail environment excluding the cashier area, since after paying for an item a customer could take actions that are identical to concealing goods. Please note that since shoplifting itself is a subtle event to accurately detect with a high probability, Scylla monitors suspicious behavior that can result in shoplifting.
Yes. The alert is triggered based on the actions in the camera view. Facial biometrics plays no role in the decision making process of AI.
Smoke & fire detection submodule is built on the largest dataset for the Anomaly Detection and Behavior Recognition models so far. As the title suggests, the system alerts in the event of either smoke or fire. The environment where the system could be installed can be both indoors and outdoors.
Scylla Anomaly Detection and Behavior Recognition is designed to help security units by supporting their daily operations, augmenting their capabilities and eliminating possible human-factor related flaws. Also, in case of a possible threat the alert that is sent out by Scylla is enriched with information crucial for quick and inclusive analysis of the threat on site and effective planning of dedicated counteractions.
An alert containing all the crucial information is compiled and delivered to end users responsible for security. There is a number of customizable alerting pathways: Scylla dashboard, Scylla mobile alerting application, access point relay boards, and VMS alerting API, to name a few.
Yes, Scylla Anomaly Detection and Behavior Recognition can work both on cloud and on premise.
Scylla Anomaly Detection and Behavior Recognition treats the whole frame and analyses all actions within the frame. Thus, regardless of the number of fighting instances in the frame - the alert will be triggered if there are any instances of violence.
The acts of fights and vandalism are detected by a dedicated model that was trained on a large amount of real-world video recordings from CCTV cameras. The model for shoplifting is separate and it is trained to detect possible shoplifting actions only.

Contact us

+1 747 271 4717
11801 Domain Blvd, Austin, TX 78758, United States
Armenia Office
+374 94 008316
118/1 Papazyan, 7th floor, Yerevan 0012, Armenia
Email Address
  • Product center

    Gun DetectionThermal ScreeningIntrusion Detection & Perimeter Protection SystemBehavior Recognition
  • Case Studies

  • Company

  • Legal

  • Integration with