How Do Artificial Intelligence and Computer Vision Control HSE at Industrial Facilities?

Artificial Intelligence and Computer Vision in Occupational Safety

The implementation of artificial intelligence (AI) and computer vision technologies is becoming a natural stage in the development of occupational health and safety (HSE) systems. Vasily Dolgov, co-founder and CEO of VizorLabs, discusses the practical aspects of applying these technologies at industrial facilities. Using real-world cases, the speaker demonstrates how video analytics helps monitor compliance with safety requirements, track technological processes, and prevent incidents.

From Video Surveillance to Video Analytics

Traditional video surveillance systems, originally designed for physical security, require constant operator attention and often generate thousands of false alarms. The presentation details the transition to proactive video analytics systems. Modern solutions integrate with existing camera infrastructure and use neural network models to automatically detect violations: from the absence of PPE to presence in hazardous areas. This reduces the workload on HSE specialists and allows for prompt responses to real threats.

Control of PPE and Hazardous Areas

One of the key sections of the webinar is dedicated to automating the control over the use of personal protective equipment (PPE) and the presence of personnel in hazardous areas. The system can recognize not only the presence of a hard hat or workwear but also the correctness of their use (for example, whether a jacket is zipped or a face shield is lowered). The speaker explains how dynamic hazardous areas are configured, which are activated when equipment is turned on or during loading and unloading operations, and how the system records violations, such as the lack of three points of contact when climbing stairs.

Integration with Wearable Devices and Multimodal LLM

Additional technologies are used to improve the accuracy of identifying violators and analyzing context. The presentation shows the use of BLE tags sewn into workwear to determine the location of employees without the need for facial recognition. Special attention is paid to the use of Multimodal LLM (large language models working with images). These systems can analyze frames, identify non-obvious violations, and enrich event cards with links to regulations, which significantly simplifies incident investigation.

What you will learn from this webinar:

• How to distinguish real video analytics from video surveillance systems with basic detectors?
• What are the requirements for cameras and infrastructure to implement computer vision?
• How does the system recognize the correct wearing of PPE and adapt to the specifics of the enterprise?
• How is the control of employees' presence in hazardous areas and workplaces automated?
• How do Multimodal LLM technologies help analyze the context of violations and reduce the number of false alarms?