Work safety: The system for checking at the manufactures
Computer vision
Сhallenge
The more employees there are in manufacture, the more difficult it is to monitor, how they comply with industrial safety rules. For example, whether employees always wear helmets, glasses, uniforms. Do they follow the rules of working at the lathe, etc.
A real-time video incident recognition system and an interface for reporting and managing of violations.
Solution
Task
Automate the monitoring of industrial safety in production, monitor in real time and receive reports.
What can be checked?
01/
Do all employees use personal protective equipment: helmet, mask, glasses, uniform, gloves and etc.
The set depends on the specifics of the customer.
Does the employee comply with the regulations: does he handle the equipment correctly.
For whom is it useful?
For any production — in workshops and on the street — where strict compliance with safety regulations is necessary.
02/
How it works
03/
The system receives data from several cameras at once
Takes frames from the video stream with the certain frequency
Classifies objects in the selected zone. For example, the presence or absence of the helmet or mask.
Determines by the frames the staff member and the necessary zones, for example, the head.
Checks that the same person is in a series of pictures
If employee violates the rules of work, the incident is recorded
The incident card is created in the web interface
The responsible person can look at the information and decide what to do with it next
The visualization of decision-making by neural network
04/
The model builds a prediction
We take a specific grid layer
We are building a heat map based on the prediction on this layer.
What we have learned
05/
We have developed our own pipeline, which allows you to prepare markup faster.
Developed a modular architecture, taking into account that customers can be different.
We have built into our markup tool the ability to manage to complex markup: consider different types of personal protective equipment.
We have learned to monitor the work speed of the system in detail at each stage.
Interesting fact
We can do not only on the server, but also run in a more offline and mobile format, even if there is no unified network yet.
Project team
06/
Teamlead
Sergey Solovyov
Teamlead
Valery Shlyapnikov
Maxim Lukin
Teamlead
Tools
07/
PyTorch
Framework of the machine learning. Deepest learning:)
ORI MarkUp
Our internal data markup tool.
Contact us
PRODUCTS
SECTIONS
STUDIOS
Russia, Innopolis, st. Universitetskaya, 7
2023 © EORA. All right reserved