PegaSARs

This system helps physical security teams monitor controlled-access areas by automatically reviewing video feeds and generating Suspicious Activity Reports (SARs) only when predefined access rules are violated. A concrete end user would be a campus security office, training facility, industrial site, or secure installation that manages vehicle and pedestrian access across multiple cameras and locations.

The operational value is reduced monitoring burden. Instead of assigning personnel to continuously watch several live or archived feeds, the system screens activity across cameras and flags only events that require review. This creates manpower-equivalent value: one analyst can review generated SARs and short video clips rather than manually scanning hours of footage. In practice, this can reduce labor costs, improve coverage across sites, and help standardize incident documentation.

A specific use case is after-hours vehicle monitoring at a training campus. Foot traffic may be approved, vehicles exiting campus may be approved, and vehicles using a designated pickup/drop-off loop may be approved. However, vehicles traveling inbound on a restricted road toward a protected building may trigger a SAR. Because each camera has a different orientation, the system uses camera-specific rules to define what counts as entering, exiting, or approved movement.

The system can support archived video review and could be extended toward near-real-time alerting if video ingestion and model latency are acceptable. Its main limitations are video quality, camera placement, model uncertainty, and the need for clear rules. It should not replace human judgment; rather, it acts as a screening layer that reduces workload and directs analysts to the events most likely to matter.

Built With

• amazon-web-services
• pegasus
• python
• s3
• sagemaker