Intelligent Safety Video Surveillance for Static Electricity Grounding Points

Project Description

Overview

This project developed an AI-powered safety video surveillance system designed to enhance operational security in chemical and industrial environments, particularly around static electricity grounding points.
The system employs advanced computer vision and deep learning models to detect human presence, small body parts (hands, heads), and unsafe behaviors such as phone usage, ensuring real-time safety compliance in high-risk zones.
Although the core prototype has been completed and validated, the system is currently being maintained and deployed in scalable cloud-based architectures for industrial use.

Research Objectives

The primary objective was to create a fast, reliable, and intelligent detection system capable of monitoring critical safety points in hazardous facilities.
The project sought to:

  • Detect small body parts near grounding equipment to prevent unsafe proximity.
  • Identify phone usage or other distractions that may compromise safety compliance.
  • Ensure sub-second inference latency through optimized models and hardware acceleration.
  • Enable cloud deployment and scalability for remote industrial monitoring systems.

Technical Approach

The system integrates computer vision, edge AI, and cloud computing technologies:

  • Deep Neural Networks (DNNs) for precise detection of hands, heads, and motion patterns.
  • Action Recognition Models trained to identify phone-handling gestures and unsafe behavior.
  • Edge–Cloud Architecture enabling low-latency response with scalable deployment.
  • Alert and Response Mechanisms that trigger safety notifications within seconds.

Methodology Framework

  1. Data Acquisition – Collection and annotation of safety-critical visual datasets from chemical plant environments.
  2. Model Training – Development of lightweight, high-accuracy CNN-based and transformer-based models for small-object detection.
  3. System Integration – Deployment of inference modules on cloud servers and edge devices with real-time synchronization.
  4. Performance Evaluation – Validation on operational sites with metrics focused on accuracy, latency, and false-alarm reduction.

Expected Outcomes

  • Reliable detection of small body parts and unsafe human behaviors in restricted zones.
  • Cloud-based surveillance infrastructure adaptable to different plant configurations.
  • Reduced risk of electrostatic discharge incidents and improved operator safety compliance.
  • Transferable framework for safety monitoring in other industrial sectors such as refineries or power generation.

Research Areas Alignment

This project aligns with ongoing research in:

  • Computer Vision and AI Safety Systems – Intelligent video analytics for industrial risk mitigation.
  • Cyber-Physical Integration – Real-time decision systems bridging physical safety and digital monitoring.
  • Industrial IoT and Cloud Infrastructure – Scalable and distributed architectures for intelligent manufacturing environments.