Skip to Content
AI Analytics · Computer Vision · OpenVINO · Gaudi AI

260 Cameras. 7 Siloed Systems.
Zero Intelligence. Until Now.

Dubai alone operates 260+ AI surveillance cameras across 7 incompatible platforms. HCT's AI and Video Analytics layer — powered by Intel OpenVINO, Gaudi AI, and HCT's Intelligent Operations Platform — turns raw camera feeds into real-time operational intelligence.

50+Concurrent Stream Processing
<30msInference Latency per 4K Frame
OpenVINOIntel AI Runtime — Edge Native
99.2%+Detection Accuracy (COCO Benchmarks)
AI Model Portfolio

Production-Grade Models for Critical Operations

HCT does not offer generic AI. Every model is validated for operational environments — ATEX zones, crowded terminals, low-light surveillance — and optimised for Intel silicon inference.

Computer Vision
Object Detection (YOLOv8 + OpenVINO)
Person, vehicle, object detection in real-time across 4K feeds. Optimised with INT8 quantisation on Intel Xeon — achieves 15-20ms inference vs 80-120ms on GPU-only deployment. Used for: perimeter intrusion, crowd counting, PPE compliance.
mAP: 53.9% COCO · INT8 Quantised · 20ms @ Xeon E Core
Traffic Intelligence
ALPR / License Plate Recognition
Multi-angle, multi-condition license plate extraction with 98.5% accuracy in UAE Arabic/Latin dual plate format. Runs on-site edge servers — no WAN hop, no latency jitter. Feeds directly into traffic management and access control systems.
98.5% UAE Plate Accuracy · Dual Script · OpenVINO Pipeline
Predictive Maintenance
Anomaly Detection (Autoencoder + LSTM)
Time-series anomaly detection on vibration, temperature, and current waveform data from industrial assets. Trained on OEM baseline patterns, deployed as ONNX model on Xeon AMX tiles. Predicts failure 72-96 hours in advance.
72-96hr Failure Prediction · ONNX Runtime · OPC-UA Data Source
Safety Compliance
PPE and HSE Compliance Detection
Hard hat, safety vest, glove, and face shield detection per worker per zone. Generates compliance report per shift, per zone. Alert latency under 200ms. Integrates with access control to deny zone entry for non-compliant workers.
<200ms Alert · Zone-Level Reporting · CCTV Integration
Security Intelligence
Loitering and Crowd Density Analysis
Real-time crowd density heatmaps for evacuation planning and crowd surge early warning. Loitering detection with configurable zone/duration rules. Used at DXB and large event venues for proactive crowd management.
Configurable Zone Rules · Heatmap API · Alert to CAD System
Advanced Training
Custom Model Training on Intel Gaudi AI
HCT uses Intel Gaudi 2 accelerators for fine-tuning foundation models on customer-specific datasets — UAE traffic patterns, facility layouts, custom asset types. Gaudi 2 delivers 40% cost advantage over comparable GPU training runs.
Gaudi 2 · PyTorch · 40% Training Cost Reduction vs GPU
Inference Pipeline

From Camera Pixel to Actionable Alert in Under 100ms

HCT's OpenVINO-based inference pipeline processes, classifies, tracks, and alerts — all before the next camera frame arrives.

1

Ingest: Multi-Protocol Stream Capture

RTSP, ONVIF, MIPI, USB, GigE Vision camera streams ingested into OpenVINO DL Streamer pipeline. Hardware decode via Intel Quick Sync — offloads CPU, processes 4K at 30fps per stream with MediaSDK.

RTSP/ONVIF · Quick Sync · DL Streamer
↓
2

Pre-Process: Frame Normalization and ROI Extraction

Frame resize, color normalization, region-of-interest crop based on configured detection zones. GPU tile handles preprocessing — decouples pre-processing from inference thread pool.

Intel GPU Tile · ROI Config · GStreamer
↓
3

Inference: OpenVINO Runtime on Xeon AMX

INT8-quantised models execute on Xeon AMX matrix extensions. Throughput: 15-25ms per 1080p frame. Multi-model pipeline runs detection + classification + tracking in parallel using OpenVINO async inference API.

OpenVINO · INT8 · AMX Tiles · Async API
↓
4

Track and Correlate: Multi-Camera Object Tracking

DeepSORT tracking maintains persistent object IDs across frames and camera handoffs. Cross-camera correlation links the same individual or vehicle across 50+ feeds. Re-ID accuracy: 94% over 30-second gap.

DeepSORT · Re-ID Model · Multi-Cam Graph
↓
5

Alert and Integrate: Rules Engine to CAD/PSIM

Rule-based alert engine applies user-defined policies (loitering > 60s, crowd density > 80%, PPE absent). Alerts published via MQTT/WebSocket to HCT IoP dashboard, police CAD systems, or PSIM integrations in under 100ms total latency.

MQTT · WebSocket · CAD API · HCT IoP
Use Cases

Where HCT AI Analytics Deploys Today

👮
Police and Public Safety
Real-time CCTV monitoring for Dubai Police operations centers. Incident correlation from 260+ cameras, automatic alert routing to nearest patrol unit, ALPR for stolen vehicle tracking. Reduces response time by 3+ minutes vs manual monitoring.
ALPRCrowd AnalysisLoitering DetectionCAD Integration
⚡
Utility Substation Monitoring
Unauthorized access detection at DEWA substations, thermal anomaly on transformers from IR camera feeds, predictive maintenance on circuit breaker actuators from vibration signature analysis. All local — no data leaves the substation boundary.
Intrusion DetectionThermal AnomalyPredictive Maintenance
🚢
Port Terminal Operations
Container number recognition (ISO 6346), truck queue management, crane anti-collision, worker safety zone monitoring. Vision system integrated with TOS (Terminal Operating System) for automated gate in/out, reducing dwell time by 40%.
Container OCRPPE ComplianceCrane Anti-CollisionTOS Integration
🏭
Oil and Gas Facility HSE
ATEX-zone compliant camera feeds analysed for PPE compliance, gas cloud detection via thermal imaging, personnel tracking in restricted zones, fire/smoke detection. Alerts to HSE manager and emergency response team with zone/GPS coordinates.
PPE ComplianceGas Cloud DetectionFire/SmokePersonnel Tracking
See AI in Action

Turn Your Camera Network Into an Intelligence Engine

HCT deploys production-grade AI analytics on existing or new camera infrastructure — with no cloud dependency and full data sovereignty.

Request AI Analytics Demo Edge Compute Platform