AI Sentry · AI Real-Time Protection Gear Check for Construction Sites
"AI Sentry" is the edge AI occupational safety video recognition solution developed in-house by Microsys Information. It reuses existing on-site cameras, and an edge AI host analyzes the streams continuously to recognize whether safety helmets, reflective vests and safety harnesses are worn; once a violation is confirmed repeatedly, an alert is pushed and evidence images are retained automatically. Each violator and each missing item is marked on the image, with manual review and report export. Events can be integrated into SI, VMS and the 4S360 smart site platform through Webhook, MQTT or REST API.
Real-life photo; detection boxes are illustrative.
Use case: workers wearing helmets and reflective vests on siteIllustrative scene generated with AI; not footage of the actual product or a customer site.
Key Features
AI SentinelFour Key Features
The MemryX AI accelerator chip infers on site, handling multiple streams in parallel; video does not have to go to the cloud, saving bandwidth and protecting privacy.
Supports RTSP / ONVIF and connects directly to existing IP cameras and NVRs, with continuous stream analysis and simultaneous multi-channel detection.
Marks each violator and missing item on the image; ROI and minimum violation time can be set, with manual review and report export.
Events are pushed in real time (Webhook / MQTT / Server) with person boxes and missing items, plus an event query API.
Detection and Alert Flow
Existing IP cameras / NVRs continuously send video to the edge AI host over RTSP / ONVIF.
Recognizes people and safety helmets, reflective vests and safety harnesses in the image, judging by ROI, person box size and minimum violation time; a violation is established only after repeated confirmation.
Events include each person's box, missing items and worn items, and evidence images are retained automatically.
Events are pushed in real time to Webhook, MQTT or Server; managers can manually review, query history and export reports.
One PPE AI, Three Major Scenarios
4S360 Platform IntegrationMany contractors, little management manpower: helmets and vests are checked instantly at work-zone entry, violation photos are tallied by contractor and trade, and it shares one platform with real-name check-in, electronic fence and environmental monitoring, with LINE group push to the site manager in real time.
Many plant employees and visitors, many restricted areas: PPE is checked automatically at restricted zone and production line entrances, safety harness detection strengthens work-at-height management, and violations feed into safety inspection and audit reports.
Tunnel environments are enclosed: CCTV at shaft entrances plus wear checks, with PPE events sharing the same event rules as UWB positioning, SOS, heart rate and gas alerts, and violation footage available for playback and query.
Integration Information for SIs, Platform Vendors and Security Hardware Makers
| Event Format | phase (start / update / end), one box per person (bboxes), missing items missingItems, worn items presentItems, and whether evaluated. |
| Image Size | Events carry the original image size (imageWidth / imageHeight) so the integrating side can convert box positions on the image correctly. |
| VMS Mapping | Cameras can be assigned an external device ID, and events carry externalDeviceId to map directly to the camera on the VMS side. |
| Push Content | Choose to push only violating persons, or all detected persons. |
| Partnership | We invite system integrators (standard event JSON and API documentation), platform vendors / VMS, and security hardware manufacturers (cameras and edge hosts with built-in AI, shipped jointly). |
Technical Specifications
| Recognition Items | Safety helmet, reflective vest, safety harness (person + 3 protective equipment items, 4-class model) |
| Video Source | Existing IP cameras and NVRs, RTSP / ONVIF streams, simultaneous multi-channel detection |
| Computing Method | Edge AI host, MemryX AI accelerator chip on-site inference, video does not have to go to the cloud |
| Monitoring Settings | Per camera: ROI monitoring area, minimum / maximum person box, minimum violation time and recognition threshold |
| Violation Retention | Violation evidence images mark each violator and missing item; manual review and historical report export |
| Event Push | Webhook, Server IP Endpoint and MQTT, all using the same event JSON |
| API | REST API (event push and event query), Bearer Token access |
| Account Permissions | Three roles: administrator, safety supervisor, read-only |
| Local Database | SQLite (stores settings, events and report data locally) |
| Deployment | Ubuntu 24.04 package installation, managed through a web interface |
Want to know how AI Sentry is deployed on site?
We provide site surveys, camera and edge host configuration advice, and integration with SI, VMS and the 4S360 platform, with a quote.