Page 778 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Data • Availability: public and privately held data (proprietary datasets and
Availability operational camera feeds)
• Sources: Internal DP World data, including feeds from hundreds of
CCTV cameras. And public datasets
Metadata (Type of • Images and video streams
Data) • log data
• Analytical reports and alert notifications
Model Training Tanbeeh uses customized YOLOv8 models with approximately 750 million
and Fine-Tuning parameters. The system was trained on 100,000 images collected from
diverse environments (factories, ports, offices, etc.) with 500 hours of GPU
training, and is continuously updated using live data feedback to refine
detection accuracy.

Testbeds or • Deployed live at DP World since November 1, 2023
Pilot • Currently connected to 125 CCTV cameras with plans to scale up
Deployments • Actively used across various departments (HSE, Traffic &
• Security, Operations, Crisis Management)

Code repositories While the code remains proprietary, an overview and further technical
details can be accessed via public information at [1].

2 Use Case Description

2�1 Description

Tanbeeh is an AI-driven safety and security platform specifically developed for industrial
environments such as DP World’s port facilities. It leverages advanced computer vision, real-
time data analytics, and deep learning to automate:

• Fire & Smoke Detection: Reducing traditional detection times from 25–30 minutes to
mere seconds (currently 1 second)
• PPE Compliance Monitoring: Automatically verifying safety gear compliance, eliminating
labor-intensive manual checks, PPE compliance accuracy rose to 70% (from 50%) under
the AI-based SMART detection system, fully eliminating the error-prone manual checks
• Facial Recognition: Securing access control and reducing gate delays by rapidly
authenticating personnel, Facial recognition cuts gate-entry turnaround from 3 minutes
down to 15 s.
• Traffic Analysis: Monitoring and optimizing traffic flow to reduce congestion and
emissions.

Data Sources:

Tanbeeh connects to various cameras and pulls frames at FPS (Frames Per Second). Each
module might have a different FPS, A grabber process fetches live frames from cameras and
sends them to respective AI models for inference.

Since inference is centralized within a very fast LAN (5 Gbps), real-time analytics are possible,
and various dashboards show real-time information.

742