Real-Time CCTV Safety Compliance Analytics

Overview

Built a real-time, multi-camera safety compliance analytics system for industrial facilities, streaming over RTSP and covering four core detection features: obstruction, intrusion, PPE (personal protective equipment), and fire/smoke.

Pipeline

graph TD
    classDef trigger fill:#e1f5fe,stroke:#01579b,stroke-width:2px,color:#000000;
    classDef process fill:#f3e5f5,stroke:#4a148c,stroke-width:2px,color:#000000;
    classDef gate fill:#fff3e0,stroke:#e65100,stroke-width:2px,color:#000000;
    classDef result fill:#e8f5e9,stroke:#1b5e20,stroke-width:2px,color:#000000;

    A(["RTSP Camera Streams"]) --> B["Per-Feature Workers<br/>(poll latest frame)"]
    B --> C["Centralized GPU Scheduler<br/>(drops stale requests)"]
    C --> D["YOLO / YOLOE Inference"]
    D --> E{"Detection Confirmed?"}
    E -->|Yes| F["Webhook Alert<br/>(Teams / Power Automate)"]
    E -->|No| G["Discard"]
    F --> H["Human Review<br/>(Correct / Wrong / Uncertain)"]
    H --> I["Dataset Export<br/>(YOLO format)"]

    class A trigger;
    class B,C,D process;
    class E gate;
    class F,G,H,I result;

Technical Approach

• Independent Per-Feature Pipelines: Each detection feature (obstruction, intrusion, PPE, fire/smoke) runs as its own worker polling the latest available frame at its own rate, so a slow detector never stalls a fast one
• Centralized GPU Scheduling: A shared priority queue arbitrates inference requests across all cameras and features, dropping stale requests in favor of newer frames to stay real-time under load
• Zone-Aware Detection: Intrusion and obstruction use per-camera configurable zone masks, with background-change pre-filtering to avoid running inference on static, uneventful frames
• Evidence-Based Alerting: PPE non-compliance accumulates evidence over a short window rather than alerting on a single frame, cutting false positives from momentary detection noise

Human-in-the-Loop MLOps

• Review Workflow: Reviewers confirm, reject, or mark each alert uncertain from a dashboard, optionally correcting the detected class
• Dataset Curation Loop: Reviewed feedback is exported into YOLO-format training datasets for offline model fine-tuning — every reviewed alert becomes potential training data

Key Features

• Multi-Camera, Multi-Feature: Each camera is independently configured for which detection features it runs
• Automated Alerting: Confirmed detections trigger webhook notifications (Microsoft Teams / Power Automate) with de-duplication to prevent alert spam
• Resilient Streaming: Automatic reconnect handling and live camera-status tracking

Impact

Encouraged stricter adherence to safety procedures on the factory floor and enabled proactive incident response by surfacing issues before they escalated.

Technologies Used