Executive Summary

Worker Safety: A Global Challenge

Ensuring worker safety is a critical but challenging task across all industries. Construction and related fields like Oil & Gas often present elevated risks due to the nature of the work and the environment. Here’s a look at the Enhanced Worker Safety with Video Analytics using data from authoritative sources:

• High Accident Rates: According to the International Labour Organization (ILO), a specialized agency of the United Nations focusing on work-related issues, in 2020, an estimated 2.3 million workers globally succumbed to work-related accidents or diseases. (Source: https://www.ilo.org/global/about-the-ilo/newsroom/news/WCMS_902220/lang–en/index.htm)
• Human Cost: These accidents cause immense human suffering for workers’ families, impacting livelihoods, physical and mental health. Beyond the immediate impact, these tragedies leave lasting emotional scars on loved ones.
• Economic Burden: The ILO estimates that work-related injuries and illnesses cost the global economy an estimated 3.3 trillion USD annually, translating to a significant portion of global Gross Domestic Product (GDP). (Source: https://www.ilo.org/global/about-the-ilo/newsroom/news/WCMS_902220/lang–en/index.htm) These costs encompass medical expenses, lost productivity, disability payments, and training for replacements.

The Need for Automated Video Analytics in Worker Safety

Traditional methods of worker safety monitoring,  reliant on manual inspections by safety officers, have limitations that can be overcome with the use of technology:

• Scalability: As construction projects grow in size and complexity, manual monitoring becomes increasingly difficult to maintain effectively across numerous sites. A single safety officer cannot be everywhere simultaneously.
• Limited Coverage: Safety officers can only observe specific areas at a time, leaving blind spots where violations might occur.
• Human Error: Manual observations are susceptible to fatigue and inattentiveness, potentially missing critical safety issues like workers not wearing Personal Protective Equipment (PPE).

Also read about- How LLMs change UI and UX in software engineering

This case study explores an implementation by Clarion Analytics for an multi billion dollar Oil & Gas Marketing company in India to leverage computer vision and deep learning for automated worker safety monitoring at its construction sites. 

The traditional approach, reliant on manual inspections by safety officers, faced limitations in scalability and real-time coverage. This implementation aimed to demonstrate an centralized video analytics solution to address these challenges and achieve the following objectives:

• Improve worker safety: Ensure adherence to Standard Operating Procedures (SOPs) regarding Personal Protective Equipment (PPE) usage 
  • Safety Helmets
  • Safety Gloves
  • Safety Boots
  • Safety Harness for working at heights 
  • Job Area Barricading
  • Scaffolding  
• Enhance scalability: Effectively monitor worker safety across numerous construction sites as the company expands its operations.
• Gain Deeper insights: Obtain data-driven insights into worker safety compliance patterns to inform targeted interventions and training programs.

Challenges and Traditional Approach

• Manual monitoring limitations: Manual inspections were time-consuming, prone to human error, and couldn’t provide real-time data for comprehensive analysis.
• Scalability concerns: As the company expands, deploying safety officers at every site (remote) becomes impractical and expensive.
• Limited data insights: Lack of real-time data on worker PPE compliance hindered proactive safety measures.

Also read about – OpenVINO 101

Technical Solution

• Network Infrastructure: The client had established a video surveillance system with the following specifications:
  • Cameras:
    • 7 x HIKVISION 4 MP 4mm cameras installed at 8 feet for worker safety gear monitoring.
    • 2 x HIKVISION 6 MP 4mm panoramic cameras installed at 20 feet for zone violation monitoring.
  • Network: High-speed broadband internet connection.
  • Storage: Network Video Recorder (NVR) on-site.
  • FTP Storage : IP Cameras configured to push images on Motion Alerts
• Computer Vision and Deep Learning: Object detection and image recognition techniques were used to identify workers and their PPE usage in video footage captured by the IP cameras.
• YOLO Deep Learning Model: A YOLO deep learning model, known for its speed and accuracy, was trained on a dataset of images labeled with workers wearing or not wearing the required PPE.
• AWS Cloud-Based Processing : While future deployments may explore edge or hybrid processing, this implementation leveraged the cloud for processing.
  • The trained deep learning model was deployed on a powerful AWS EC2 g4dn.xlarge instance with NVIDIA T4 GPU’s.
  • IP Cameras on detection of Motion (human , vehicle) pushed the images to the AWS EC2 FTP Server (t3.medium) . The FTP server is configured using vsftp daemon.
  • A custom script using cron job automated the transfer of images received at FTP server to a designated Amazon S3 storage bucket in the AWS cloud at regular intervals. This ensured efficient data transfer and scalability.
  • Celery, a distributed task queue, was used in conjunction with RabbitMQ, a messaging broker, to manage the workload of processing incoming images efficiently from multiple cameras.
  • The model was exposed as an API service using FastAPI, allowing Celery and RabbitMQ task queue to execute images against the PPE AI Model and update the MySQL DB for any Non Compliances .