Beyond the hype: delivering value with computer vision #1 has been saved
Beyond the hype: delivering value with computer vision #1
Real world applications and challenges of computer vision
The ongoing hype around computer vision is here to stay due to the many success stories which showcase its strength. However, these stories often omit crucial challenges, which lead to organizations struggling to derive value from their computer vision investments. In this blog series we describe some of these main challenges and showcase how to circumvent them.
By Vincent Bolwerk and Pavel Bugneac
A brief journey into the Computer Vision landscape
As a new technology emerges, so does the hype around its potential value. With the initial hype around Artificial Intelligence (AI) in the 1960s, computer vision was at the forefront of the optimism shown by researchers1. This branch of AI allows machines to extract meaningful information from image and video data by using sophisticated machine learning models. Modern day examples of such models include object detection, pose estimation, image classification and face recognition.
Although implementing computer vision in practice proved to be more difficult and resource-intensive than initially thought, in the recent decades we have started to see examples where it has shown its full potential. This has been enabled by several technological advancements such as internet, (cloud) computing capabilities and the development of new (deep) neural networks. By now, computer vision has progressed through the Gartner Hype Cycle for Artificial Intelligence2 , and is one of the AI technologies closest to entering the “Plateau of Productivity”. In this phase, real-world benefits of the technology are demonstrated and accepted and tools become increasingly stable. Despite this, several implementation challenges remain, which can obscure some of its proven benefits. In this blog series, we will demonstrate how these challenges can be tackled in order to reap the tangible value from computer vision.
Hello world: computer vision in real-life applications
Computer vision has allowed for numerous applications in a vast range of fields and industries, such as automotive, retail, medical or manufacturing industries. We now encounter computer vision applications in our daily lives, making the benefits of its implementation more apparent. For instance, computer vision is the backbone of autonomous driving where complex neural network models are used to detect various objects on roads in real time to determine the next move of the car.
Similarly, it is used in the retail industry by enhancing self-checkouts without the use of barcodes in order to improve the customer experience or improving store layout by tracking customers in-store and generating heatmaps which allows for an increase in revenues.
Example of computer vision application in the autonomous driving industry. Deep neural networks are used to help autonomous vehicles detect other vehicles/objects on the road. Source.
Through the Connected Stores initiative Deloitte has generated heatmaps to establish person traffic inside of stores. Source
Another area where computer vision brings value is in the field of medical research. Within this field, a number of early adopters of computer vision have developed numerous applications that deliver life-saving outcomes for patients. Examples include early detection of cancer, tracking tumor development, speeding up MRI scan analyses, accelerating medical production by identifying and counting microscopic cells and many more. By using computer vision, these applications are able to process enormous amounts of data in a short time span while still maintaining a high degree of accuracy. This becomes crucial when the degree of curability of certain conditions is highly dependent on the timing of the diagnosis3.
Computer vision applied in medical research to identify and count different cell types at microscopic level. Source .
Why successful computer vision implementations are challenging
While these successful and groundbreaking examples of computer vision applications showcase its ability to transform and disrupt industries, the challenges faced during the implementation process can potentially become crippling. In fact, most companies that invest in computer vision struggle to bring proof of concepts into reality and derive tangible business value from their investments4 .
This raises the question about the ideal approach that should be considered by analytics leaders to preserve the enthusiasm around the value of computer vision and AI in general. A recent study5 supported by Nyenrode and Deloitte, finds that three factors emerge as cross-cutting for successful adoption of AI:
- Businesses should not engage with AI for the sake of AI but use it to solve a specific business problem.
- More data is not always better, but rather quality data is needed. For computer vision, in particular, this implies both quality footage, stressing the importance of hardware, as well as accurate and consistent labelling when training models.
- AI’s transformational character requires targeted investment in upskilling and reskilling throughout the organization.
Deloitte analytics leaders are making use of the above 3 factors to support the implementation of computer vision solutions capable of adding value to our clients’ organizations. Our approach entails starting from a clearly defined problem, making impact more measurable, as well collecting as much high quality data as possible by exploring different sources and modes of collection.
As an example, a scalable cloud solution was built for a client in the renewable energy sector. By automating drone flight routes around wind turbines, high quality and standardized image data was collected which was then used to identify potential inspection issues. The success of this implementation rests on the fact that there was a clear business problem expressed by the need to process increasingly larger amounts of data; as well as a highly specialized team that was able to develop the solution6.
Another great example is our work on preventing illegal deforestation7 , where we successfully improved and scaled existing solutions by integrating data pipelines and creating fully custom web applications. By using satellite images paired with other geographic data, the improved solutions are able to accurately predict future illegal deforestation which enables early action. In this example, both the availability of high quality data as well as the support of a highly skilled team of specialists were some of the key success factors for tackling the existing problem.
Overcoming implementation challenges to generate value with computer vision
In the previous section, we have briefly introduced a few examples where we managed to successfully incorporate computer vision into solutions and bring them into production. In subsequent blogs of this series, we will deep-dive into two of our active computer vision projects in which we face the challenges stressed earlier and show the approach we are taking to overcome them. These two computer vision solutions developed by Deloitte comprise:
- AI4Animals – Animal welfare application in which we use computer vision modelling paired with an intuitive dashboard to improve animal monitoring in slaughterhouses . Video fragments of potential animal welfare problems are collected in a dashboard for review so that corrective actions can be taken. This not only brings highly innovative solutions to our clients but also helps to improve animal welfare.
- Field Hockey Insights – Field hockey is known to be an innovative sport where new technologies are quickly adopted. With loads of video data available we are developing a player detection tool for the Dutch National Hockey team that augments already available event data, which opens up new possibilities to support tactical decision-making, engage fans and increase sponsorship value.
If you would like hear more about these solutions and the latest computer vision technologies we deploy, make sure to check out our upcoming blogs! Additionally, if this blog has inspired you to explore if computer vision can help tackle challenges within your organization, do not hesitate to contact us.
1 How Artificial Intelligence Revolutionized Computer Vision: A Brief History
2 Hype Cycle for Artificial Intelligence, 2020. (gartner.com)
3 Computer vision opportunities in Medical Imaging Explained
4 Why Is It So Hard to Become a Data-Driven Company? (hbr.org)
5 Navigating the path of AI adoption
6 Automated, Cognitive and Connected
7 Cognitive Deforestation Prevention