AI Safety · 6 Use Cases · Real-time

The AI that watches when you can't.

AIVision turns ordinary CCTV cameras into an always-on intelligence layer — detecting fire, near-misses, vehicle hazards, and unsafe behaviour in milliseconds.

0% Detection Rate
0ms Response Time
0+ AI Use Cases
AIVision Live Dashboard
Live Dashboard Active
98% Accuracy
34ms Latency
6+ Use Cases
Computer Vision · Real-Time Detection · LLM Analytics · Cloud Ready · On-Premises · Smart Alerts · Edge AI · CCTV Intelligence · Computer Vision · Real-Time Detection · LLM Analytics · Cloud Ready · On-Premises · Smart Alerts · Edge AI · CCTV Intelligence ·
Use Cases

Six layers of
intelligent protection

See the AI working in real time — click any tab to switch between detection modules.

🔥 Fire & Smoke Detection Live Monitor
Live Feed Alert Log Analytics
LIVE CAM-03 · Zone C
W02
3:15
W05
1:28
FIRE — 87%
SMOKE — 74%
CAM-03 16:23:07
🔥 FIRE + SMOKE DETECTED
Computer Vision · 34ms · Streak 4f
FIRE SEVERITY: HIGH (87%)
Near-Miss Detection Live Monitor
Live Feed Clips Saved Analytics
LIVE CAM-01 · Zone A
! NEAR-MISS W12
HEAD DROP 42%
W03
3:58
W07
1:31
CAM-0116:23:07
⚡ NEAR-MISS DETECTED!
AI Vision Engine · 3-signal fusion · Clip saved
FALL SIGNAL + HEAD DROP 42% · SIGNAL STRENGTH HIGH
🚛 Vehicle Proximity Detection Live Monitor
Live Feed Zone Config Analytics
LIVE CAM-02 · Zone B
MACHINERY — 91%
W12
140px
DANGER
W08
280px
WARNING
W03
SAFE
CAM-0216:23:07
🚛 PROXIMITY DANGER
Computer Vision Model · 3-layer zones
W12 IN DANGER ZONE — 140px FROM MACHINERY
🚭 Smoking Detection Live Monitor
Live Feed Violation Log Reports
LIVE CAM-01 · Entry Gate
Person — W07
SMOKING
93%
FRAME CONFIDENCE
f.0421
93%
f.0438
87%
f.0455
81%
f.0472
78%
CAM-0116:23:07
🚭 SMOKING DETECTED
Computer Vision Model · High Accuracy · Frame 421
SMOKING CONFIRMED · HIGH CONFIDENCE (93%)
🦺 PPE Compliance Live Monitor
Live Feed Violations Compliance Report
LIVE CAM-04 · Entry Gate
NO HARDHAT — 95%
W04
✕ VIOLATION
COMPLIANT — 98%
W09
✓ COMPLIANT
SITE COMPLIANCE
60% COMPLIANT
OK Violation
CAM-0416:23:07
🦺 PPE VIOLATION — W04
Computer Vision Model · 11 classes
SITE COMPLIANCE: 60% (1/2 workers non-compliant)
RESTRICTED ZONE
INTRUDER — W14
UNAUTHORISED
CAM-0516:23:07
🚧 ZONE BREACH DETECTED
Computer Vision Model · Zone perimeter monitoring
📹
VIDEO FEED PENDING
Detection model ready · Awaiting footage
ZONE BREACH — UNAUTHORISED PERSONNEL DETECTED
CAM-03 · Fire Zone 🔥 ALERT ACTIVE
FIRE
87%
SMOKE
74%
SEVERITY HIGH
Use Case 01

Fire & Smoke Detection

YOLO11s fine-tuned on pyroelectric datasets. Per-class confidence thresholds and streak filtering eliminate single-frame flicker false positives.

  • Separate fire and smoke confidence thresholds
  • Streak filter — 3+ consecutive frames = real event
  • Heat glow overlay around detection zones
  • Real-time severity meter
0% Recall
0ms Response
0+ Frame streak
CAM-01 · Zone A ⚡ NEAR-MISS!
NEAR-MISS
DROP 42%
DROP SIGNAL · FALL DETECTED
Use Case 02

Near-Miss Detection

Three-signal fusion model — fall posture detection, sudden head-drop tracking, and optical-flow motion blast for falling objects.

  • Signal 1: Bounding box aspect ratio flip (fallen = horizontal)
  • Signal 2: Head keypoint drop > 35% of body height
  • Signal 3: Farneback optical flow motion blast
  • Auto-saves 2s pre-buffer + 3s post clip
0 Signals fused
0s Pre-buffer
0 Cooldown frames
CAM-02 · Zone B 🚛 PROXIMITY DANGER
EXCAVATOR
MACHINERY 91%
W12
140px
W08
DANGER ZONE BREACH
Use Case 03

Vehicle Proximity Detection

Three-layer proximity analysis creates tiered alert levels — from early warning to critical zone breach, with dashed danger zone visualisation.

  • Layer 1: WARNING — worker within 300px of vehicle
  • Layer 2: DANGER — worker inside 1.6× expanded bbox zone
  • Layer 3: CRITICAL — direct bounding box overlap
  • Pixel-distance line drawn between each pair
0 Alert layers
1.6× Danger expand
0px Warning radius
CAM-01 · Site Entrance 🚭 SMOKING ALERT
Person
SMOKING
93%
f0421
93%
f0438
87%
f0455
79%
SMOKING CONFIRMED · 3 FRAMES
Use Case 04

Smoking Detection

Custom model trained on overhead CCTV frames from the exact site environment — annotated and fine-tuned via Roboflow for maximum accuracy.

  • Custom dataset: 21 annotated frames, site-specific
  • Trained via Roboflow → 34 epochs → mAP 60.8%
  • Direct Roboflow inference API — no local model needed
  • Person detection overlay for context
60.8% mAP50
0 Training epochs
0% Peak confidence
CAM-04 · Entry Gate 🦺 PPE VIOLATION
NO HARDHAT
95%
COMPLIANT
98%
60% COMPLY
1/2 WORKERS COMPLIANT
Use Case 05

PPE Compliance

Automatic detection of hard hats, safety vests, masks, and PPE violations — flags non-compliance the moment workers enter frame.

  • Detects: Hardhat, No-Hardhat, Safety Vest, Mask
  • Worker-level compliance scoring
  • Real-time site-wide compliance percentage
  • Construction-Hazard-Detection YOLO11s model
0 Classes detected
0% Violation conf.
Real -time
Real-Time Insights

Live Analytics

AIVision delivers instant visual intelligence — every frame processed, every anomaly flagged, every metric tracked in real time across all your CCTV feeds.

Vehicle Proximity PPE Compliance Worker Tracking Fire & Smoke
AIVision Vehicle Proximity Analytics
Vehicle Proximity & Alert Detection
AIVision PPE Compliance Analytics
PPE Compliance & Worker Metrics
How It Works

Advanced CCTV Analytics with
Computer Vision & LLM Intelligence

AIVision delivers best-in-class vision capabilities using Cloud-based or On-Prem LLM models — giving your site the power of enterprise AI without changing a single camera.

84% Reduction in safety non-compliances
73% Reduction in recordable incident rate
50% Reduction in manual safety inspections
🔥

Fire & Smoke Detection

Real-time detection of fire and smoke across all camera feeds. The moment ignition starts, AIVision triggers an alert. Our Computer Vision Model analyses every frame with streak filtering to eliminate false alarms — delivering reliable detection even through partial obstruction or low visibility.

  • Simultaneous fire & smoke confidence scoring per frame
  • False-alarm elimination with streak-frame filtering
  • Instant alert dispatch in under 1 second

Near-Miss & Fall Detection

A 3-signal fusion approach powered by advanced pose analysis — detecting worker falls, sudden head drops, and abnormal motion before they escalate. Auto-clips save a 5-second window around every event, giving you evidence for every near-miss that occurs on site.

  • Pose skeleton tracking across 17 body keypoints
  • Fall posture + head-drop + motion-blast signal fusion
  • Auto-saves pre- and post-event clip for every incident
🚛

Vehicle Proximity Detection

Three-layer proximity zones around every vehicle — warning, danger, and critical. AIVision tracks every worker's distance from moving machinery in real time, triggering escalating alerts the moment anyone enters a hazardous zone, before an accident can happen.

  • Real-time pixel-distance measurement per worker
  • 3-tier alert escalation: Warning → Danger → Critical
  • Compatible with any heavy machinery or construction vehicle
🚭

Smoking Detection

Identifies smoking behaviour automatically from existing CCTV — no additional hardware required. Paired with LLM-powered reporting, it generates instant violation records and full audit trails, making it easy to enforce site policies and demonstrate compliance.

  • Site-specific model for accurate real-world performance
  • Works in both outdoor and indoor environments
  • Automated violation logging with timestamped evidence clips
🦺

PPE Compliance Monitoring

Continuously checks every worker entering the site for correct PPE — hard hats, safety vests, masks, and more. Violations are flagged instantly with worker-level detail, giving site managers a real-time compliance score and a complete digital audit trail every day.

  • Detects 11+ equipment classes simultaneously
  • Worker-level compliance scoring in real time
  • Daily compliance reports and trend analytics
🚧

Unauthorised Zone Access

Defines and enforces virtual perimeter boundaries across any zone of your site — no physical barriers needed. When anyone crosses into a restricted area, AIVision triggers an immediate alert with timestamp, zone ID, and a full evidence clip for accountability.

  • Fully configurable virtual perimeter zones
  • Breach detection with sub-second response time
  • Full audit trail with timestamped evidence
☁️

Cloud & On-Premises Deployment

AIVision supports both Cloud-based LLM models and fully On-Premises LLM deployment — giving enterprises complete data sovereignty alongside best-in-class Computer Vision capabilities.

☁️ Cloud Ready 🖥️ On-Premises 🔒 Data Sovereign ⚡ Real-Time
Start Today

Make your site
intelligent today.

No hardware changes required. Deploy AIVision on your existing CCTV infrastructure — choose Cloud-based or On-Premises LLM deployment — and go live in under an hour.