With the advancing development of technologies based on artificial intelligence, new areas of use are emerging. Computer vision is one such area. The technology can assist people in several ways, who can instead devote themselves to other tasks, and in some cases the technology is both faster and more reliable.
– “It may involve monitoring processes in industry, for example, where the system can detect an event early on in the preliminary stage and issue a warning before anything has happened,” says Olof Mogren, Senior Researcher in Machine Learning at RISE.
In healthcare today, a computer can read an X-ray just as well as a doctor can – and even better in some cases. But the technology has many potential uses, such as automatic quality inspections or safety monitoring. However, before a computer vision system replaces a person, very high safety requirements must be met. Autonomous vehicles are one example.
"In many cases, the system is very good at making the right decisions quickly and is sometimes better than a human being at driving a car,” says Aleksis Pirinen, Senior Researcher in Machine Learning and Computer Vision. “Yet we still we don’t feel that the technology is mature, even though it could already, to some extent, lead to fewer deaths on the roads.”
Various and high thresholds for use
For some computer vision applications, there are ready-to-use AI models, such as for X-ray images where a computer has learned to distinguish between an intact bone and one with a fracture. But for those seeking to develop something new and specific to their business, a number of steps must be implemented.
“A fundamental point is to have training data that your AI model should train on,” says Pirinen. “For example, if you want the system to recognise agricultural plant diseases, you need pictures of healthy plants as well as diseased plants.”
And once you have your training data, you need to tell the system what it depicts by means of so-called annotations, or information about what each image contains.
For those wanting to implement or develop computer vision technology in their organisation, RISE can be a partner for teaching, collaboration, or research.
“We have many research projects through which we can demonstrate what can be done with the data and to meet the needs of the partner,” says Mogren.
There are extremely wide areas of application that are only limited by our imaginations
Several ethical aspects
An ethical dilemma related to computer vision is bias, which is sometimes unconsciously built into the system by those who trained it. Among other problems, there have been instances where the system does not read dark-skinned faces as well as it does for light-skinned faces.
“This is of course extremely problematic, you have to work actively to minimise this kind of bias in your models,” says Mogren.
“One way could be ensuring greater diversity among those who develop the models,” says Pirinen.
Another ethical dilemma, often highlighted in the media, is the risk of ‘deepfakes’. Not only can computer vision learn to understand what it sees, but it can also create images.
“With the right skills, very realistic images and videos can be made,” cautions Pirinen. “This is something that will need to be tackled in the future.”
“As with all technology, it can be used for both good and evil,” adds Mogren. Positive uses for images generated by the systems include digital twins, which can be used in research to perform tests in a very realistic simulated environment, and images that show the potential effects of upcoming climate disasters.
“It’s a way to demonstrate how powerful these systems are, and how there are both risks and positive uses.”
Retinal imaging and weed control
Two computer vision applications that RISE helped to develop are retinal imaging as a form of drug testing, and new possibilities for weed control.
“We worked with data and algorithms to find weeds, then another operator developed a robot that seeks and destroys them using a small laser,” says Mogren.
Computer vision is an emerging technology, and we have not yet seen all the application areas for which the technology can be useful.
“Autonomous vehicles and all types of driver assistance systems are also arenas where this technology is incredibly useful. I would say that computer vision has an abundance of benefits. There are extremely wide areas of application that are only limited by our imaginations,” concludes Mogren.
