Precision is an overriding theme for the field of Artificial Intelligence (AI). The scores of AI start-ups that have emerged during industry 4.0 promise greater clarity, insight, and transparency over whatever problem they’re trying to solve. In the main they deliver, usually by leveraging the mountains of data businesses create to paint clear and lucid pictures of fuzzy scenarios that befuddle businesses. In so doing, they help organisations to see.
Few technologies do so with the accuracy of Computer Vision (CV), however. The technology, powered by AI, interprets information from images and videos usually garnered from something as perfunctory as CCTV. It can, to an exacting standard, identify specific objects and scenarios and then make decisions based on its ability to interpret the environment. In a practical sense, it can be applied to almost any scenario and environment – and indeed has been to brilliant effect.
For the business world, that is proving a particular boon and goes some way to explaining why the industry is expected to be worth close to US$18bn by 2024. The demand for quality inspection and automation is rising and the need for highly specific CV systems is increasing. It’s a scenario backed up by the assertion of EmpiricAI’s CTO Zafar Khan. His organisation uses CV as the basis for solutions that monitor health & safety in business and industrial settings. `Right now’, he says ‘we’re seeing an increasing demand for what we are doing.’ So why the sudden explosion?
The demand for CV is down to advances is computing power. The use of deep learning and artificial neural networks allows CV models to learn a specific data set and then to recognise similar instances. To begin with the model is fed training data – a few hundred images for example - which are annotated according to the parameters it wants to detect.
Over time it’s fed more and more of these images until it can detect images that were outside of the original data set with a very high degree of accuracy. The beauty is that the model can be applied to any scenario as Khan explains. “No new code needs to be written to solve a new problem,” he says. “You just put new data in the directory and allow the AI to learn it. It’s a relatively straightforward process.”
At the start of the pandemic, EmpiricAI trained its framework to recognise COVID-19 safety measures such as mask wearing, social distancing and building occupancy levels. Having already developed similar tools that monitor health and safety in industrial settings, it’s a clear demonstration of how CV can be applied to new scenarios. Moreover, it’s brought clear and tangible benefits to businesses and solved a problem that is way too complex for humans to monitor alone.
In other industries, it’s a similar story. Agriculturalists in Australia have trained an AI system to distinguish between weeds and crops on its plantations. It means they no longer have to use chemicals to kill the weeds, resulting in higher quality yields, healthier plants and a reduction in labour and chemical costs. They’re training the same system to recognise early signs of damage to fruit grown on trees and expect similar results and an accompanying increase in profits.
That type of dynamic pattern recognition has already been shown to surpass the human eye in other fields. In the healthcare sector, a system trained to look for neurological disorders from a CT scan outperformed trained physicians. It can do the same with skin cancers. In industrial settings, similar systems - including EmpiricAI’s - can be trained to detect health and safety parameters, fire hazards and even instances of horse play among employees. For a technology in relative infancy, how does this evolve?
Today’s AI is considered narrow by the projected standards of the future. Those operating in the field are aware that applying trained models to new challenges requires a lot of new data, training and time. Though effective, there’s a genuine desire to make that process more fluid and more efficient. Nobody is resting on their laurels.
So, what does the future have in store? It is expected that Computer Vision AI will combine with other forms of AI to produce dynamic systems that can integrate different forms of knowledge, unpack causal relationships, and learn new things autonomously. It moves the needle from deep learning to something that resembles deep thinking. That’s a prospect that is altogether more human – and an exciting one.
It’s a future that is beginning to emerge. Separate studies into how CV can be used to diagnose mental health problems corroborate that. For a problem with wildly complex data sets, the results are promising. One study in Thailand was able to diagnose depression from facial cues with 65% accuracy, while another in Europe could do the same for children with ADHD with 96% accuracy.
The next iteration is to combine Computer Vision AI with Audio AI to generate a more complete picture. It is hoped that the technology could diagnose early warning sign of mental health problems and stress in schools and workplaces in the future. It’s a fascinating use of computing power. But should we be surprised?
Khan says of his organisation, “Good AI systems require large data sets, but in the industrial and corporate sectors there is a very limited amount of tangible data typically available. To overcome that, we’ve developed an innovative framework that works with and learns from small datasets.
“If there’s a problem that a human can solve, we are quite sure AI can solve it too. Often, it can do it quicker and more accurately. That’s now. In the future, as the technology improves and evolves, its capacity to identify and solve problems will only get better. In that respect, there are few limitations and vast potential for this to scale and to help every aspect of business and society.
“As we’ve seen over the last 12 months, unforeseen circumstances will always arise and will always need to be dealt with. Businesses will need and want to advance. Society needs to evolve. Having the tools to do so is invaluable. There are exciting times ahead and we’re very proud to be playing a part.”
Find out more about WorkSafe Analytics.