Aggressive Behavior Detection
Scylla Aggressive Behavior Detection swiftly detects movements associated with fights and vandalism, allowing for a rapid response to prevent violence.
Book a demoRelated materials
FAQ
The system analyzes 5-second chunks of video footage. It first detects individuals or groups of people who are close to each other. Then, each region where a person is detected is analyzed for anomalies. The system also uses an "effective area" to reduce false positives by focusing on signals that are within the frame and are perfectly visible.
For best results, the camera angle should be around 45 degrees to prevent people from obscuring each other. The camera should be elevated higher than a person's height, and people should occupy a portion of the screen. A minimum frame rate of 20 FPS and an aspect ratio of 16:9 are also recommended.
Yes, after operating in your environment for a while, the Scylla Fight Detection model can be further trained to better differentiate abnormalities.
The Aggressive Behavior Detection module identifies violent and hostile behaviors, such as fight, assault, arrest, and brawling.
No, the module requires direct physical contact lasting at least a few seconds to trigger detection. It does not detect aggressive walking or chasing behavior.
Yes it can, as long as there is direct physical contact.
No it can’t, direct physical contact is needed for the fight detection alarm to be triggered.
The system is capable of detecting both individual fights and large-scale group fights. It analyzes multiple interactions within the frame to identify violent behavior across different individuals.
The number is dependent on the configurable parameter, the larger is the number, more instances of fight per camera could be detected, but also the hardware consumption is rising as well, recommendation is to set this number to 3.
The vandalism detection module is a modified version of the aggressive behavior detection system. It flags vigorous actions such as striking an ATM with a tool or repeatedly punching an object, which may indicate property damage. However, it only detects cases where significant damage is inflicted—activities like graffiti drawing would not trigger an alert.
Scylla's anomaly detection modules are trained on footage from static cameras, meaning that for optimal accuracy, the camera should remain stationary. However, slow-moving PTZ cameras can still be used, though any movement — especially abrupt changes — may increase the risk of false alerts. Proper calibration and strategic camera positioning can help mitigate this issue.
The preferred aspect ratio is 16:9 for optimal performance. However, the module supports all aspect ratios and can function across different video formats.
Yes, Scylla AI’s video analytics can operate on drone-mounted cameras; however, accuracy may be reduced compared to static setups. For optimal performance, the drone should remain as close to the event as possible and minimize movement to reduce false detections.