The developers of Novosibirsk State Technical University (NETI) are working on the creation of computer vision technology that will help identify violations in production. The system is already able to detect non-compliance with standards related to the use of personal protective equipment (workwear).
"Computer vision is a technology for automatically extracting information from images. A human can handle this task, but due to the huge amount of information and the human factor, it cannot work as efficiently as a trained algorithm. In this case, machine learning technologies come to the rescue. Today, computer vision technologies are widely used, and the necessary data sets and even trained models are publicly available. The main engineering task is the effective use of existing tools with refinement to solve a specific practical problem," said Ivan Tomilov, Head of the Department of Automated Control Systems at NSTU-NETI.
As a rule, computer vision uses neural network algorithms created in the image and likeness of the human brain. Training a neural network requires careful work, similar to human training: it is necessary to repeatedly present certain images of the necessary objects in order to establish strong "associations".
The hardware and software complex includes a camera that receives a real-time video stream and transmits it to a specialized computer that analyzes data for violations of the use of personal protective equipment. Information about detected violations can be transmitted to a special control room for rapid response.
The developers noted that a prototype of a system for monitoring the use of personal protective equipment has already been implemented. The next stage is the development of an additional feature that will allow you to create a customized set of workwear for different work sites. This will allow for a flexible approach to ensuring occupational safety. For example, in one workshop only rubber boots and gloves may be required, and in another — a helmet and a protective vest.
The approach developed by NSTU-NETI specialists has its own characteristics. The algorithm does not just recognize the necessary elements in the entire image, but precisely determines the area in which this element should be located. For example, to check for gloves, you must first determine the position of a person's hand.
The introduction of industrial video analytics will improve occupational safety during work at hazardous industries or facilities; reduce the level of occupational injuries, including the number of deaths and severe injuries; reduce costs in investigating cases of violations of safety regulations and occupational injuries.