Computer Vision in Industry

Computer vision is one of the components of artificial intelligence, specifically the ability of a computer to perceive and analyze visual images like a human. The use of such technology is widespread in many areas of human life, but in this article, we will pay special attention to the role of computer vision in industrial production. How is the application of computer vision in industry organized? What tasks does it solve? What are the known examples of implementing computer vision in industry and how has it helped productions? Read the answers to these questions below.

Computer vision in manufacturing is used to increase its productivity and profit. To achieve this goal, it is necessary to transfer all production data to digital format, starting from individual installations and ending with the entire production process. An important feature of using computer vision in production is that the result of use should be not a picture, but a number, which will characterize the quality and quantity of what is produced.

There are three clusters of problems that computer vision solves in industry. The first cluster is the safety problem: for example, computer vision monitors the perimeter so that nothing is taken from the enterprise, or controls movements in zones so that workers do not get into places where it may be dangerous, or monitors compliance with safety rules (wearing helmets, gloves, etc.). The second cluster is related to accounting problems, which are solved using printed barcodes on labels — this allows fighting theft or simply accounting for how much product has arrived at the production and left it. Finally, the third cluster of problems is quality control, and these tasks are divided into easier and more difficult: it is easiest to track matches by physical parameters (for example, the sizes of certain objects, which often turn out to be defective due to the peculiarities of production), more difficult — to control whether consumers are using the product correctly.

IT specialists have developed two approaches to solving the tasks described above. The first classic approach is to solve the tasks using image processing algorithms, for example, highlighting some elements in them by color or using contrast, removing noise, morphologically transforming to further work with these objects. The second, more modern approach is to use a neural network capable of solving tasks such as Recognition, Classification, Detection, Segmentation, Superresolution, Pose estimation, etc. Despite the fact that the neural network copes well with the above operations, sometimes there is not enough data for full use of this approach. There are a number of enterprises that use computer vision in production and have already improved their indicators thanks to this computer technology. For example, the Chelyabinsk Metallurgical Plant used computer vision to facilitate and speed up the process of steel defect control. The company introduced a video stream system, thanks to which it was possible to classify about 20 classes of defects, starting from scratches and ending with microcracks. This allowed to increase the defectoscopy process by 6 times. Another example is labor protection and compliance with safety rules in production. Thanks to image analysis technology and comparing them with existing data, Rosenergoatom, having introduced video surveillance, improved indicators for safety technique control: if before the introduction, safety technique dispatchers manually viewed recordings from cameras, often missing violations, now with an automated video analysis system, violations are detected exclusively by computer vision in 95-98% of violations.