Human vision makes it possible to interpret countless aspects of the world. These perceptions are diverse, for example, when looking at a tree in the street, you can notice different aspects of the plant, such as colors, textures and shapes. You can also recognize characteristics of images, such as emotions on the faces of people in a photo. 

Computer vision is one of the branches of artificial intelligence that studies the processing of real-world images by a computer. In other words, this area investigates ways of giving machines the ability to visually interpret information, in other words, to see. In this article, developed by the team at Kot Engenharia, you'll find out how this technology works and how the company can work in this area.

Figure 1: Human and computer vision. [1]

Basic concepts 

Firstly, for a better understanding of computer vision, it is necessary to define the concepts of Image and Image Processing.

The concept of an image can be somewhat abstract. With this in mind, it is defined, from a computational point of view, as a set of data. This information is related by means of a scalar function of two unknowns [2].

This data usually contains a lot of noise from factors such as where the image was captured, the equipment used, the objects in focus, and so on [3]. Image processing aims to make the necessary adjustments to attenuate these variations. To this end, as the image is seen by the computer as a function, there are numerous mathematical techniques to model the data as required.

Deepening understanding

Using the concepts presented above as a basis, computer vision is responsible for designing a complete automated system that captures images, processes them, analyzes the visual information and triggers a command to initiate the appropriate subsequent process.

Stages for the realization of Computer Vision [4]

Initially, it is necessary to establish that there is no single system for applying computer vision. The most common steps for implementing this technology are listed below:

• Acquisition: stage aimed at capturing the images;
• Image processing: the aim of this stage is to adapt and optimize the visual data acquired. To do this, techniques such as removing noise, rotating the image, applying filters, etc. can be applied;
• Image analysis: this is when the images are made unique from the computer's point of view. Each image is assigned a unique function of two independent unknowns, which can be viewed more objectively by the machines;
• Pattern recognition: at this point the images are classified according to their similar characteristics.

By recognizing patterns, actions can be initiated by the system automatically, giving the necessary sequence to the process.

Advantages and Limitations of Computer Vision

The advantages of computer vision include the possibility of implementing a system that reduces the company's costs. It is also possible to increase the quality of the products the company produces.

As it is a specific system for each application, each case must be analyzed, planned and executed individually. This means that the investment required for development is high, which is the main limitation of computer vision.

Industry applications

Computer vision has a number of applications in industrial automation:

• Non-invasive check

 In this case, it is possible to check that the contents of the packaging are complete, excluding the need to break the product's wrapping, for example. Figure 2 shows the schematic used for this application.

Figure 2: Schematic of the application of non-invasive verification of package integrity. [5]

• Inspection of machinery and structures

Computer vision can also be used to carry out inspections and analysis on machines and structures [6]. Check out in this article a methodology applied by Kot.

• Accident prevention

Another plausible application for computer vision is in preventing machine collisions. Figure 3 illustrates the situation.

Figure 3: Distance monitoring and collision avoidance scheme. [6]

• Pattern recognition

The applicability of computer vision also extends to recognizing patterns that initiate automated processes, such as identifying printed labels, codes, letters and numbers. Figure 4 exemplifies the use of computer vision for stock management.

Figure 4: Computer vision box stock management scheme. [6]

Conclusion

From reading the article, it can be concluded that computer vision has many possible applications for solving problems and optimizing processes. Kot Engenharia has the necessary knowledge to apply it, being able to evaluate different operating contexts and contribute to the results. Contact our team for more information!

Follow our pages on LinkedIn, Facebook e Instagram to keep up with our content.

References:

[1] Blake, Scott (2018). Disponível em: <https://unsplash.com/photos/K6JzHiV4aq8>.

[2] G. Kovasznay, L., & Joseph, H. (1955). Image Processing. Proceedings of the IRE, 43(5), 560-570. doi:10.1109/jrproc.1955.278100.

[3] Santiago, Gaubert (s.d). Method Based on Computer Vision for Digit Recognition Aimed at Reading Consumption in Hydrometers with Analog and Digital Indication.

[4] Backes, A and Sá Júnior, J (2019). Introduction to Computer Vision Using MATLAB.

[5] Szeliski, Richard (2010).Computer Vision: Algorithms and Applications .

[6] IFM Collection.