Artificial Intelligence has been bringing in new waves of innovation since the recent past, and the latest advancement in AI getting attention from technology companies is visual AI. Simply put, this is making machine intelligence make sense of the images and visual data much like the human eye, which not only sees but perceives the meaning behind the images.
Earlier, developers were working with numeric data to create algorithms that machines could “learn” from and execute routine tasks. With progress in AI, now training computers to capture and “learn” tasks using visual data is increasingly being pursued today.
How Visual AI Has Grown
It was in 2001 that image recognition was first explored by AI scientists, when Paul Viola and Michael Jones used a face detection algorithm where machines could record facial features much like a human eye. In 2005, Navneet Dalal and Bill Triggs explored the possibility of machines scanning features of pedestrians for security purposes. Today, Google and Amazon are combining AI-based image recognition with IoT to bring unprecedented human-machine interfaces for ease of every area of human life.
Leading visual recognition AI companies like Vinsa and Visua presently work with use cases adopted for specific situations unlike other AI solutions that are generically developed for large-scale use. For instance, in the COVID-19 scenario, visual AI can ensure that PPE gears have no damages that may put the healthcare staff in danger. In unsafe or hazardous zones, visual AI can inspect and detect leakages in pipelines using drones and robots with cameras and capture visual data to be analyzed in a smart way. In the retail industry, visual AI can be used for gauging customer crowding at supermarkets and deploy attendants for serving their needs. In the hospitality industry, especially when hosting large-scale conferences, visual AI can help with parking situations, finding vehicles and matching them with owners to streamline crowds. Another common use case for visual AI has been detecting logo copyright infringements in print-on-demand marketplaces where brand names are compromised. While technology found a way to capture rich imagery several decades ago, computers are only now being trained to use such visual data to extract patterning and derive human-like intelligence.
A lot of visual artificial intelligence use cases depend on IoT, edge computing to arrive at the results they want. Interestingly, IoT traditionally uses structured numeric data to function. But with visual AI this is gradually changing. Now, the IoT devices capturing temperature are being trained to capture visual data like condensation to provide the same data inputs and derive intelligence. In that sense, visual AI can be thought to be an extension of IoT and driven to do similar tasks but with a more evolved human-like understanding. In this context, visual AI can help expand our problem-solving capabilities for situations that may have needed human intervention earlier.
Daniel Bruce, founder and CEO at Vinsa, which provides visual AI solutions to manufacturing, construction and retail industry shares an interesting example in these COVID-19 times in a podcast. He says entertainment venues, which need crowd analytics and to operationalize social distancing, can easily use visual AI to not only configure how long the queues take to move, but also if the requisite space is being maintained by groups.
Besides this, Daniel counts critical industrial inspections, especially of legacy equipment that maybe more expensive to replace, as having an important use case for visual AI, due to its cheaper, more efficient and safe features.
In Daniel’s experience, it’s necessary to have a buy-in from companies who are shopping for visual AI solutions. Making them understand the complexities of customizing visual AI solutions to specific use cases and explaining the benefits in terms of better ROI is mandatory, as this is an experimental approach at present. For cases, where this is not cost effective, he admits it’s only reasonable to set the right expectations for those who approach them.
Benefits of Visual AI Solutions for Companies
Commonly, safety, quality management and savings in outlays in terms of man-hours, extending lifespan of expensive equipment are the reasons most companies choose visual AI automated solutions. Visual AI has shown great success in reducing human error within industrial situations that are data-sensitive and need insights deeper than just numeric inputs andwhere traditional IoT can be expensive.
Visual AI Increases ROI through Automation
Futuristic scenarios according to Daniel for visual AI will ensure increased ROI through automation. He cites ideas like checking the breading on chicken before frying in burger joints, or ensuring pepperoni is spread evenly on pizzas as use cases for retail eateries, which can help deliver standardized quality at reduced rates and no human intervention involved.
Visual AI can Aid Admin Activities
For companies working on return-to-office scenarios, the Admin teams can be helped immensely with visual AI COVID-19 screening models with visual thermal and temperature scanning, social distancing being maintained in public places and more. Medical diagnosis for infected employees, contactless employee assistance, industrial inspections and instances where automation using images, visuals and videos as inputs can be worked out can best be served by visual AI.
Increased Efficiency in Routine Tasks
With visual AI’s capabilities to categorize datasets of visual inputs, task efficiencies are increased, as algorithms can be coded to align with outcomes and these looped back to improve further “learning” of the solution. For the construction industry, a real-world understanding of the terrains they are working with can improve the safety levels of on-site workers. Visual AI-driven deep learning can help rework architectural plans and drive data-driven decision-making, thereby improving the project’s execution and profitability.
Visual AI can Better Robots and Autonomous Machines
When visual AI is incorporated in robots that are now being extensively deployed in manufacturing, hospitality, retail, energy and several other sectors, the autonomous operating machines achieve highly improved output capabilities.Vinsa and Boston Dynamics are partnering to use their quadruped robot Spot for critical inspections, reading gauges to preempt anomalies with high accuracy.
Visual AI can provide extensively sharp datasets for deep learning to become more focused with better results. In the upcoming decade, businesses that ignore visual AI may get left behind in the competition and be rendered insignificant.
Visual AI technologies are expected to grow in the immediate future with more companies searching for niche solutions that adapt automation for their unique requirements. Many enterprises already have a lot of visual data like photographs and videos that they need to make sense of. Like Daniel says, “IoT deals with structured data, typically numerical readings. Once you’re able to bridge the gap between the types of data and really see visual data sources as first-class citizens right alongside that structured data, it really broadens the whole perspective of what we have at our fingertips.” While the challenges of reducing the costs involved in incorporating visual AI into IoT solutions are still nowhere to being resolved, the benefits of operational efficiencies for the modern enterpriseand the gain in ROI far outweighs this.
With this in mind, the best way forward is choosing the right scenario for use of visual AI tech mapped closely to the ROI. Daniel says, “Once the use case has been selected, it’s important to set reasonable expectations. These technologies are amazing, but they’re not perfect. Understanding that the technology will get things wrong, but having a plan in place for when it does, is critical to the success of visual AI combined IoT deployment.”
So, is your enterprise ready for robots to watch over your processes?
