- Computer vision harnesses AI to make discoveries and predictions about business operations using digital imagery sourced from cameras.
- The best use cases of the technology lie at the intersection of good source data, technical feasibility, good training data, the organisation’s capabilities and the ability to articulate business value.
- When considering implementing the technology, organisations must ensure they understand AI and have planning in place to address the associated risks and regulatory implications.
The human senses are our connection to the world around us. Of those, sight is a function of the reflection of light and the focus of our eyes, but vision is the interpretation of what we see. Informed by many external factors, vision is subject to prejudices and conditioning.
But computer vision — a form of machine learning, powered by artificial intelligence — is promising to unlock a world beyond these human limitations. With multiple channels of perception and the ability to process large amounts of data, machines can have far greater visual capabilities, offering the potential to uncover crucial information within an organisation’s physical presence.
Bringing business into view
Using cameras and incorporating other sources of data, computer vision collects intelligence about the most important aspects of a company’s operations: the people, products, assets, and documents that form the backbone of their processes. By collecting digital images and applying deep learning models and other computer vision algorithms, computers can accurately identify and classify objects or data presented in an unstructured format, and react to what they ‘see’. They can work with digital image data in the form of photographs, videos, views from multiple cameras, infrared or alternative spectrums, LIDAR (Light Detection and Ranging, which is a remote sensing method that uses light in the form of a pulsed laser bursts to measure distances) or multidimensional data from inputs such as medical scanners.
For business, it can have a multitude of applications, including:
A business that depends on its physical components, such as a manufacturer, can apply computer vision to help identify problems with equipment before they arise. For example, a software program called ZDT (Zero Down Time), developed by FANUC, collects images of equipment via cameras attached to robots and, with accompanying metadata, sends these to the cloud to process for faults. During an 18-month pilot, the solution was deployed to 7000 robots in 38 automotive factories, and in that time, detected and prevented 72 component failures.1 PwC’s Drone Powered solutions also uses computer vision to detect faults, performing automatic inspection of remote and difficult to access assets such as high power transmission lines.
The digitisation of physical documents in an organisation is a necessary, but often onerous, part of the digital transformation process. Much of it is unstructured and in forms such as invoices, receipts, handwritten notes or contracts. Where traditional object character recognition technology fails to extract specific content from image-based documents, computer vision can accurately detect and narrow down target objects and deliver key insights that may have otherwise been lost or take years to uncover. Researchers at the US Department of Energy’s Lawrence Berkeley National Laboratory last year created an AI algorithm that, with no previous training in materials science, trawled through old science papers and predicted now-known thermoelectric materials.2 It hints at the power of this technology to uncover knowledge that might have been overlooked by human experts.
Retail businesses, including supermarkets, are looking to create a seamless shopping experience for customers that removes the need for checkouts and, inevitably, waiting in line. In the US, Amazon Go employs computer vision as part of the technology suite enabling this new kind of shopping experience, using cameras that recognise individuals, track them through the store, know which account they are linked to, and record which products are placed in their bag with a high level of confidence.3
Multi-dimensional image intelligence
In the field of medicine, a viable use case for computer vision is in radiology. Current AI solutions generally involve aiding radiologists to diagnose diseases and conditions from X-rays, MRI or CT scans. MaxQ AI, for example, offers a solution that uses computer vision to help physicians identify rare anomalies in brain scans and quickly suggest treatment options.4 In the case of some illnesses, such as strokes, the faster the treatment time, the better the outcome for the patient, so technology such as this could be incredibly important. Microsoft’s InnerEye visually identifies and displays possible tumors and other anomalies in 3D images uploaded to the software.5
Getting the most out of computer vision
According to a survey conducted in 2019, 64 percent of global senior business purchase influencers said that computer vision will be very or extremely important to their firm in the coming year, while 58 percent said that they were implementing, or going to implement, or interested in implementing computer vision in the coming year. 6
However, while this technology applies to a host of valuable use cases, not all are destined to succeed. Those that will drive the most impact at an organisational level will lie at the intersection of deployment capability, data and the inherent value of the technology to business needs.
The four questions therefore to ask when getting started to help drive the benefits computer vision offers are:
- Is it valuable? A computer vision solution must be able to articulate substantial business value relative to existing solutions, in order to justify the investment.
- Can you deploy it responsibly? Ensuring the right infrastructure is in place is vital. Hardware constraints, computing power, cloud hosting, bandwidth and cost are all key considerations to enable accurate modelling. More importantly: is it appropriately governed? Has the technology passed all tests for performance (bias, fairness, explainability, robustness and security)? Is it implemented considering an appropriate ethical framework?
- Can you get good data? Quality and availability are key. Any variation in the source (ie. different types of cameras, different lighting, angles and distances) will make it harder to leverage computer vision effectively.
- Can you train it? The simpler the object, motion, or other input, the easier it will be to create a computer vision model to detect it. The greater the nuance and variation, the higher quality data will be required.
Manage the risks, reap the rewards
Organisations should assess the use cases for computer vision and base their decision making on the business value, the data they have available, and their ability to deploy the necessary solutions. However, before implementing a solution, it is imperative that boards and senior management increase their understanding of AI and have a plan in place to manage potential risks and regulatory implications, such as cyber security, compliance, the availability of key skills to manage it, data governance and critically, the ethics underpinning it. Computer vision is a powerful use of AI and as such must be approached through the lens of Responsible AI in order to be used safely.
With the appropriate processes in place, organisations can leverage computer vision to deliver innovative solutions to drive growth and operational efficiency for their business. Given its disruptive ‘seeing power,’ the potential impact computer vision will have for organisations and more broadly, society, promises to be significant.
Visit the Responsible AI website for further articles and information on PwC’s comprehensive suite of Responsible AI frameworks and toolkits.
Written with Sathesh Sriskandarajah.