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

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

One smart city already runs 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.

Request AI Analytics DemoIntelligent Operations Platform
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 on Arabic/Latin dual-script plates. Runs on-site edge servers — no WAN hop, no latency jitter. Feeds directly into traffic management and access control systems.

98.5% 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 major hub airports 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 — local 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 police operations centers (national police group reference). 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 utility 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

Frequently Asked Questions

What can AI video analytics detect?

Real-time threat and intrusion detection, crowd and traffic flow analysis, PPE and safety compliance, vehicle and licence plate recognition, anomaly detection and forensic search across recorded footage.

Does the analytics run on-premises or in the cloud?

HCT Group deploys AI analytics on-premises or at the edge so video is processed locally. This keeps latency low and keeps footage inside the customer's environment and the country, rather than streaming it to an external cloud.

Can it work with existing cameras?

Yes. HCT Group's analytics layer is designed to unify feeds from existing camera estates, including installations spread across multiple separate systems, rather than requiring a full hardware replacement.

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 DemoEdge Compute Platform