Computer Vision Applications Examples Across Different Industries

Allowing computers and machines to see the world in the same way humans do was one of the leading tasks of computer science in general from the moment it appeared as a discipline. So it’s no wonder when Computer Vision, the technology capable of analyzing regular images from any camera and transforming them into meanings and decisions, announced itself, it started changing the surrounding industries almost immediately.

By now, Computer Vision is a multi-billion dollar field, with Computer Vision applications triumphantly finding their way into an incredibly diverse range of companies and projects and establishing themselves as a part of our daily routine. This article will overview the present popular, rising, or innovative Computer Vision applications that already are or would shape the field in 2021 and beyond.

Afterward, we will overview some promising Computer Vision project ideas that can give a creative foundation for the field’s interest. Then, we will also try to sketch how it will develop in the future.

Table of contents

1. What is Computer Vision now?
1.1 Example of Computer Vision usage: pose detection
2. Manufacturing quality control
2.1 Steelworking
2.2 Food packaging
2.3 Breakdown prediction and detection
3. Computer Vision applications in Security
3.1 Real-time security
3.2 Authentication with Computer Vision
4. Computer Vision applications in Healthcare
4.1 Disease diagnosing
4.2 Blood loss
4.3 COVID-19
5. Computer Vision applications in Transportation
5.1 Self-driving cars
6. What is the future of Computer Vision? New Applications to Come
7. Conclusions
8. FAQ

What is Computer Vision now?

Current Computer Vision has an extensive application area, one of the biggest in the AI domain. In the recent few years, Computer Vision has developed into a field with very specific tasks, deeply oriented on solving distinct classes of problems, or even the particular problem, with incredible accuracy.

The majority of the development’s driving force continues to be the Convolutional Neural Networks (CNNs), which were the main factor of the field’s rapid progress. New findings on their technology create state-of-the-art performance, and this process is very likely to continue doing so for a prolonged period.

We can generally place Computer Vision tasks in four categories: image classification, segmentation, detection and tracking, and information retrieval from the visual data.

According to the report published in September 2020 by the Grand New Research, Computer Vision has reached a value of US$19 billion in 2019 and is expected to grow further. Right now, the biggest volume of Computer Vision usage is in the manufacturing quality assurance and control area, which in 2019 accounted for more than a quarter of the market. Healthcare, automotive industry, agriculture, and personal electronics are also among the front-runners in the industry implementation. Most of them we will outline in our coming sections.

Picture 1. Example of pedestrian detection with Computer Vision // ll0zz, Indif; Wikimedia Commons

Manufacturing quality control

Computer Vision applications in manufacturing quality assurance became the booming use case and are directly connected to the new wave of process automation. Availability of cameras and the ability to install them on the most machinery enables to automate the quality control (which can, according to some reports, take up to the 10-15% operation cost) on most of the production and even more importantly, in some places that are usually unfit for human access.

Computer Vision control does not have a margin of human error and doesn’t get bored of repetitive routine tasks, making it an ideal choice for the constant production monitoring and reducing resulting product defects share to the possible minimum.

Steelworking

One of the most prominent applications developed in quality control is the example of Computer Vision usage in the steelmaking industry. Traditional quality checking here required many hours of manual work of highly-qualified employees, which ineffectively distributed the workload.

This was one of the first industries to turn attention to the Computer Vision applications, and it already counts decades of improvement. Right now, all of the world’s biggest steelmaking companies are in the various stages of automatic defect detection implementation.

Steel surface defect inspection is one of the most crucial quality control procedures, but the full manual inspection is impossible due to enormous production. The previous solutions were selective inspections or real-time operator review, which was ineffective on a high-speed line.

There are several classes of surface defects, and the most common problems arising are defect classification and sorting. First, the possible defect locations are determined by segmentation. Then, the classifiers (usually Neural Networks) are used on these locations and are proven to reach the almost-perfect result in filtering defects. Such technologies give all implementing companies a significant competitive edge. Sometimes such models are relatively easy to implement.

Picture 2. Different classes of defects on the steel surface // NEU surface defect database

Food packaging

Developed Computer Vision solutions are accepted much faster when equipped to use the regular or already existing cameras and/or hardware, making the producers invest only in software and decision-making rather than into sometimes more expensive physical novations.

One such example is the packaging date label verification system APRIL Eye developed by OAL Connected with the support of the University of Lincoln, which is directed at food manufacturing and has the goal to reduce food waste as a result of misdating. System scans the product for the needed dated information and verifies it while using the ordinary cameras, making it extremely cost-effective for the manufacturer.

Breakdown prediction and detection

Another great Computer Vision application in manufacturing is the maintenance control, which enables to significantly lower the observation cost and make the malfunction detection easier and closer to instantaneous. As is the case for quality control, cameras can replace humans in dangerous situations or even in the monitoring of the critical infrastructure.

Such a model would detect a sign of abnormality on the camera input and then notify about the malfunction happening, or either initiate an automatic reaction, for example, redistributing the production load until the technical team resolves the detected issue.

Computer Vision applications in Security

The security area is one of the most promising areas of Computer Vision applications right now. This area generally divides into two groups: real-time security observation and access control.

Real-time security

Computer Vision can be more accurate than humans in spotting potentially hazardous objects, such as guns, as all it needs is just to spot a needed pattern, and so it can be used as a surveillance assistant. With a current amount of security cameras, up to hundreds in a single institution, there is sometimes just too much visual information to analyze effectively.

Computer Vision is challenging this and already has numerous advances in the field, no longer susceptible to the human errors mostly caused by hours of continuous observation. Live-feed video detection can help to filter out unneeded footage and focus on some situations that are potentially requiring attention.

Authentication with Computer Vision

Authentication systems also benefit primarily from the Computer Vision-driven solutions. Some technologies already made it to the mainstream market, with the most famous example of Face ID that is now widely used in everyday life.

Picture 3. Example of facial recognition with the OpenCV library // Beatrice Murch, Sylenius; Wikimedia Commons

However, possible applications of Computer Vision in identification do not stop at face recognition. Identifications by body parameters such as height, width, limb length, and even stride type are developing methods that can be used in complex to ensure identification error to be minimal.

Computer Vision applications in security often get intertwined with others. A great example of this was set on a steelworking plant in Latin America, where the automatic surveillance system with Computer Vision was implemented. The system almost eliminated the need for manual surveillance review, but apart from that, the system was also built to detect mechanical dangers and malfunctions. The number of hazardous situations and injuries on the plant significantly decreased after implementation.

Computer Vision applications in Healthcare

While the manufacturing and security applications are enormous in labor optimization, medicine is the area where Computer Vision has possibly the most lifestyle-changing applications.

Disease diagnosing

The most popular and influential Computer Vision application in healthcare so far is disease diagnosing. Due to high performance in pattern recognition, Computer Vision has reached the point where it started to outperform human pediatricians in finding some pathologies. Such accuracy can mean that in the future, doctors will spend more time consulting the patient than on diagnosing technicalities.

With many diagnostics requiring highly-skilled doctors that are sometimes not available in some places, highly accurate diagnostics powered by Computer Vision can substitute some of these functions for a lesser cost.

As another significant benefit, Computer Vision systems can be used for the effective early diagnosis of illnesses. For example, there are a lot of companies that use it to detect lung and breast cancer early using MRI scans. Computer Vision can also be used in real-time with the patient examination to create an interactive interface with spotting features to examine.

Blood loss

One of the areas where Computer Vision is already sure to outperform human observation is a measurement of blood loss during childbirth. Such situations sometimes resulted in deaths due to unaccounted blood. Still, with a new Computer Vision software easy enough to run on an iPad, which is already implemented in some hospitals, there is a way to assume where the help is needed correctly.

COVID-19

Another example of how adaptive Computer Vision approaches are revealed itself in the number of applications aimed to fight COVID-19. Apart from the diagnostics models which can differentiate COVID-19 pneumonia with more than 90% sensitivity, there were a lot of supportive models developed, for example, social distancing control system, which identifies people who are closer than 2 meters apart, or detection of people who are or are not wearing the face mask.

Picture 4. Social distancing control system using Computer vision // Basile Roth

Computer Vision applications in Transportation

While talking about Computer Vision, it is impossible to avoid the transportation category as it is the field where its applications became the most famous.

Self-driving cars

Self-driving cars technology is possibly the field where the transition of decision-making to AI-powered technology became the most apparent for the entire world. Decisions made by cars are primarily based on data gathered from many cameras installed in different positions. A number of other technologies, such as accident prevention, also mainly relate to visual data.

Other popular examples of using Computer Vision in transportation include licence plate recognition, real-time traffic violation analysis, and smart traffic systems that are in a large development.

The road applications became so developed that recently the multi-camera systems which detect the dangerous driving style or the driver’s fatigue and issue the respective warnings started to be installed in the corporate delivery trucks by companies such as Amazon.

Promising Computer Vision project ideas

Despite the possible impression that Computer Vision has found its way into every industry possible, there are still many applications that are waiting for optimal Computer Vision solutions. Here are some of the Computer Vision project ideas that are largely discussed but still have a lot of unexplored potential that can be used for boosting the project ideas:

  • Handwriting recognition – slightly more challenging than the character recognition and becomes more challenging with different writing styles;
  • Human gesture recogniser – despite a lot of research in the area, there are still many unexplored cultural situations. Sign language recognition is one of the most classic and yet challenging applications.
  • Image captioning – the borderline area with the Natural Language Processing, which allows finding the best explanations for what is pictured on the image;
  • Sports tactics recogniser – for some of the sports, such as American football, some existing models that can determine the tactical build from the footage, but the area still has a lot to explore.

What is the future of Computer Vision? New Applications to Come

Computer Vision is a rapidly developing subject with new surprising applications arising almost every month, and it is for sure that it has a lot more to offer. According to the mentioned earlier report published by the Grand New Research, the Computer Vision market is expected to grow with an annual rate of 7.6%, passing the US$19 billion mark by 2027, while other reports predict reaching US$17.4 billion in 2024.

In any case, with further development of more accessible hardware options and cloud computing, the possibility of using Computer Vision applications is expected to be available for more businesses and much more data, fueling its spread.

Computer Vision technologies can become more integrated with other fields, for example, with the rising Augmented Reality technologies, which will create more realistic perception. Connection of Computer Vision and Natural Language Processing can also create new borderline technologies, which in general will try to translate the world as visible to humans into the textual format, creating a wide range of possible applications.

Another possible future Computer Vision application can be the predictive area, which may gather popularity with new applications in smart city management, traffic regulation, or crime prevention. Possible tasks include determining suspicious activities faster than humans and potentially preventing crimes before they happen.

Conclusions

Computer Vision is one of the most challenging areas of the rising Artificial Intelligence. Computer Vision technologies became part of the wave that is now pushing our technology ahead.

And despite rapid development, there are still countless applications that are not discovered yet or are waiting for an appropriate business problem to announce itself. Therefore, as a field who already passed its initial forming stage and by now has realized how much it can theoretically do, Computer Vision promises a bright future to the already great present.

FAQ

What companies use Computer Vision?

A large number of companies use Computer Vision today. This list starts from such technical giants like Google, Amazon, Facebook, and Tesla, to some specific tasks in thousands of smaller companies or startups worldwide who work on the Computer Vision as their main direction, sometimes very unique. The most curious examples include NAUTO, which work on commercial driving safety with accident prediction, Verkada, which allow operating the intelligent security cameras without hardware improvements remotely, Hawk-Eye Innovations, which work on Computer Vision in sports technologies, and even the artificial sight technology OrCam which helps customers with disabilities of visual apparatus to recognise text, currency, or even people.

What is the difference between Machine Vision and Computer Vision?

Machine Vision is a technology developed specifically for enabling the machines used in industry to have recording hardware and software, but it does not require interpreting the results in itself. On the other hand, computer vision is used for understanding and interpreting the results and appeared later than Machine Vision. So, Computer Vision often works on the data provided by Machine Vision systems.

What are Computer Vision techniques?

The main techniques that Computer Vision operates with are:

  • Image Classification, where images are labeled with different classes;
  • Semantic Segmentation, where areas on the image are labeled;
  • Instance Segmentation, where the areas belonging to each defined label are found and determined with overlapping;
  • Object Detection, where objects on the image are localized precisely;
  • Object Tracking, where we not only localize the object but then follow its location on the image in subsequent frames.

Why is Computer Vision so hard?

This mainly follows from the differences between how we and machines see the world. All data captured by cameras is stored in a multi-channel digital way. Therefore all the work, all pattern recognition by any Computer Vision algorithm is based on comparing natural tendencies in numbers. And while it still can find many patterns, the algorithm has no empirical idea what that could mean. It compares numbers until it finds suitable ones without additional context. So from this, it is the hardest for algorithms to grasp image context.

How do I start a computer vision career?

To start your Computer Vision career, it will be beneficial to acquainting yourself with some basic calculus. It is in large the foundation on which the Computer Vision algorithms are built. In any case, you should become familiar with Python or C++ languages next as the most developed Computer Vision libraries are usually either present in or made for these two languages.

Learning the Computer Vision itself right now is remarkably easy and accessible. You just need to subscribe for one of the numerous Computer Vision courses, for example, the famous Deep Learning specialization by Andrew Ng, if the goal is to become familiar with Neural Networks. During this time, it is a perfect moment to start making some starting Computer Vision projects yourself to try to be more used to its specifics and methods.