Here I present my vision of the future for this area and its applications, taking the opportunity to make a critical reflection on its current state and pointing out possible ways to explore.
Artificial Vision (or Computer Vision) aims to provide the machine (computer) with the ability to see and interpret the world as humans do. To do this, it analyzes the images taken by the camera (s) and tries to interpret/identify/relate patterns to obtain information about the scene that is watching. This process is called Image Processing and has been the basis of Computer Vision for many years.
Nowadays, as a result of the evolution of machine learning techniques, there has been a transition to the application, almost exclusively of artificial intelligence (AI), techniques, specifically machine learning, in the area of Computer Vision.
CB Insights reports that Computer Vision news coverage has increased by over 500% since 2015, becoming a booming industry. This growing interest in this area results largely from the advances that AI has made possible, making it mainstream.
The global computer vision market is expected to surpass $ 48.3 billion by 2023 and AI in the computer vision market will reach $ 25.32 billion by the same year.*
*Source: MarketsandMarkets Analysis
Future challenge to overcome
I highlight here some of the future challenges that need to be addressed and which derive, to some extent, from the problems/limitations of current computer vision algorithms and techniques.
First, many of the approaches are not yet sufficiently accurate (or have robustness issues) for their real-world use. The real world is imperfect, diverse and unpredictable, which strongly influences the performance of algorithms. Computer vision techniques work very well in a controlled environment, but when we move into the real world everything gets complicated. It is necessary to improve the accuracy of algorithms and find ways to deal with unpredictability and diversity.
Most current techniques are highly data-dependent, that is, they work very well if the necessary data is provided, as they are based on machine learning techniques. These techniques work properly when there is a diverse database that covers all possibilities so that you can learn and answer any questions and situations that may occur.
On the other hand, most of these approaches are used blindly: we do not understand well what is going on in terms of algorithm, much more when deep learning techniques are used where features that are extracted from images are also learned by the techniques themselves. This can cause some problems, for example, misclassifying images, even in situations that for us humans is extremely simple.
Finally, we need to improve the performance of the algorithms. Although the use of machine learning techniques has made much progress in this regard, there is a need to invest in this field.
Some researchers, including myself, consider that there is an excessive (or not very careful) use in some cases of machine vision techniques in terms of computer vision, which has posed several problems.
We need to look back at the techniques of the past, which were based on knowledge, on what was real and merge with machine learning techniques – trying to figure out what is behind the image, what I am learning – try to recognize and use this information to create my models and not rely solely on data in a blind way. This should be the way forward in this type of techniques.
Hot topics to keep in mind
Some topics in vogue today that will continue to be very interesting in the future, and where there will be much research in this area are:
- Intelligent image enhancement (improve image quality through AI techniques).
- Semantic scene understanding (perception and understanding of the scene to its fullest extent, ie to detect and recognize objects and their relationship, taking into account the context).
- Human activity recognition (recognize the activity performed by humans, although there are already several approaches, the algorithms are not yet perfect, so there is still work to do).
- 3D imaging (we have a world to work on in terms of 3D images, we need to rethink and create specific techniques for working with 3D images, not just adaptations of 2D to 3D algorithms.).
- Data fusion (take advantage of the various data from different sensors and their relationship through data fusion techniques for better response).
Future applications of computer vision
Photos of the presentation Future Applications of Artificial Vision Techniques (One Vision – 2025)
⇒ It is estimated that autonomous cars will run on our roads between 2020 and 2040. There is currently a big bet on autonomous cars, and there will continue to be, where the role of Computer Vision is crucial, and the points highlighted here even more.
⇒ The security issue is very important not only of surveillance but also in video analysis. The techniques of facial recognition and recognition of actions taken by humans, for example, to detect illegal activities, are extremely important for this subject. However, there are issues related to the right to privacy that must be taken into consideration (the danger of “Big Brother”).
⇒ At the factory of the future level, the bet on the implementation of industry 4.0 will remain a reality. The idea is, for example, to have several cameras, which are monitoring the activity performed by human operators, assisting them when necessary, collecting data on what machines are doing and providing that information to a central system.
⇒ In our future workspace, I would like to look at a document and quickly get a highlight on what matters most, this means using computer vision techniques to know where the user is looking and recognizing the document, and, for example, augmented reality techniques for presenting the information. Another possible application will be the creation of collaborative and hybrid work environments, that is the bridge between the digital and the physical world, where visual information about the physical surroundings and the objects is correlated with the digital knowledge obtained from a company.
⇒ How are we going to shop in the future? Imagine a world where all visual content is instantly buyable, and available to try virtually before purchase. The idea is to look at a particular product you like and take a picture, get immediate information about the various stores where the product is available, try it virtually, and make the purchase.
⇒ In precision medicine, computer vision can help a doctor prescribe the best medicine for a specific patient given his or her entire history. Improving diagnostic support in terms of image analysis will continue to be a gamble and, why not, to predict disease from image analysis and fusion with other existing patient and disease data.
⇒ In terms of marketing, there will continue to be a focus on presenting the most relevant advertising by trying to understand who the user is, his or her history, and displaying relevant ads according to the search history, ie moving towards personalized marketing. Also stands out facial recognition and emotion for advertising feedback.
⇒ In agriculture, the use of drones and robots will be a reality in the near future, where computer vision will certainly play a relevant role. Computer vision is already and will continue to be, used in agriculture to monitor crops and detect disease.
But the areas of application of Computer Vision are many and diverse (eg, inspection activities, robotics, home helpers, drones, mobile devices, consumer electronics, entertainment) with the most exploitable potential being presented here.
About the author:
Luis Magalhães – scientific coordinator of the CVIG applied research domain of the CCG.
With a degree in Systems and Computer Engineering, Master in Computer Science and Doctorate in Computer Science. He is currently Assistant Professor with Aggregation at the University of Minho and integrated member of the ALGORITMI Center. He is the author or co-author of more than 90 scientific publications in international journals and conference proceedings. His research interests include Computer Vision and Computer Graphics. Participates and/or participated in various research projects related to Virtual Environment Modeling, Immersive 3D Environments, Virtual Environments for Education, High Dynamic Range Imaging, Mixed and Augmented Reality Systems for the culture and entertainment industries.