Robots are taking jobs away from people. According to Bank of America Merrill Lynch, by 2035, humanity will lose 800 million jobs due to the growth of industrial automation. In his report at the conference “Data & Science: the world through the eyes of robots”, Alexander Belugin from the Tsifra company said that full automation can increase the profitability of an enterprise by 10-15%. In many ways, this significant figure determines the global trend for the mass introduction of robots, the design of which rarely goes without machine vision systems.
Machine vision in manufacturing
The term “machine vision” is not synonymous with “computer vision”. Computer vision includes the entire set of theories and technologies for creating devices that can detect, track and classify objects. Machine vision is a narrower term, it considers the application of computer vision to industry and manufacturing.
In production, machine vision is used at almost all technological stages. Most often, we are talking about its use by robotic manipulators during assembly operations, as well as control systems to minimize scrap. On a traditional production line, a person decides how well a product is made. But the automated line is equipped for this with conventional or “smart” cameras, as well as a system for processing incoming images.
The composition of such a quality control system is rather complicated. It includes:
a set of digital or analog video cameras;
image pre-processing processor;
Image pre-processing software;
machine vision software;
light sources and actuators for product sorting;
a communication channel for transmitting the results obtained.
In recent years, more and more often the above elements can be found in a single device, known as a “smart camera” (do not confuse them with home video surveillance systems). These, for example, are the Genie Nano GigE models from Teledyne DALSA or Baumer CX from the company of the same name.
How machine vision works
How to make the machine vision system perform the task correctly? There are two main approaches here. First, we describe the algorithm for recognizing certain features in an image using mathematical formulas.
This is how the OpenCV open source algorithm library, which has been developed by Intel since 2006, works. This method is good when you need to recognize homogeneous graphic elements. For example, when parts lying in the same position move along the conveyor, on which light emanating from the same point falls.
The second way is training an ultra-precise neural network. This method is more time consuming, because the neural network needs to be trained on a huge number of examples, helping it recognize marriage with the desired probability (we have already described how neural networks learned to recognize hieroglyphs). But the neural network will work in a much larger range of possible conditions. It can be, for example, randomly scattered objects with glare from lamps on them. Another example of the use of such machine vision is the sorting and packaging of vegetables, which, although similar to each other, are still not calibrated, and can also have many different defects.
An example of the application of machine vision on assembly lines is visual servo control. The robot manipulator operates according to the program, but due to the wear of its parts, vibrations and other factors, its movements may not be accurate enough. In such cases, the camera is fixed next to or directly on the manipulator, and in real time calculates a correction that compensates for small errors.
Not only vegetables
Vegetables are a good example, but certainly not the only one. You can also sort vehicles in the flow of urban traffic. Such a machine vision system was developed by the company VisionLabs. The solution allows you to determine the types of vehicles, types of public transport, car brands. All this is already in demand when building Smart Cities systems.
In addition to “Smart Cities”, “smart factories” are also developing. There is a corresponding project in our country, in which the ROBODEM company takes part. She is engaged in video analytics of production. At one of the large factories, she deployed a production control system based on machine vision technology. By the way, all kinds of cameras for automatic reading of barcodes and QR codes in warehouse automation and logistics systems also belong to machine vision systems.
The large retailer Dixy has chosen the solution of the domestic startup GoodsScan for automatic monitoring of product balances. Using a machine vision system and cameras installed on forklifts, the system allows you to track stocks, not only decoding barcodes, but also building a depth map of objects and determining their sizes.
Today News Blogs is the first internationally affiliated newspaper in the United States. Partnered with The International New York Times the global edition of The New York Times the paper caters to the modern face of the United States.