Computer vision is one of the most exciting applications of artificial intelligence. Algorithms that can understand images—both still pictures and moving video—are a key technological foundation behind various innovations from autonomous, self-driving vehicles to smart industrial machinery and the filters on mobile phones that make the pictures uploaded to Instagram look pretty.
Along with language processing abilities, called natural language processing (NLP), it’s fundamental to AMCS’ efforts to build machines that are capable of understanding and learning about the world. It generally involves applications powered by deep learning and neural networks trained on thousands, millions or billions of images until they become experts at classifying what they can see.
The value of the market in computer vision technology is predicted to hit $48 billion by the end of 2022 and is likely to be a source of ongoing innovation and breakthroughs throughout the year. So, let’s take a look at some of the key trends involving this fascinating technology.
Data-centric computer vision
Data-centric artificial intelligence (AI) is based on the idea that equal focus should be given to the quality of data used to train algorithms as is given to developing the models and algorithms themselves. Championed by Andrew Ng, a renowned pioneer of deep learning, this newly emerging paradigm is relevant across AI disciplines, particularly in the field of computer vision.
Some of the first deep learning-based image recognition models were developed by Ng at Google for the purpose of training computers to recognize pictures of cats, and the models are particularly dependent on the quality of the data they are fed, rather than just the quantity.
This focus on iteratively improving the quality of labeling and using automated techniques of extracting and labeling data will enable developers to apply computer vision technology to problems where far less data is available, potentially lowering costs and opening new potential use cases.
Health and safety
A key use case for computer vision is spotting dangers and raising alarms when something goes wrong. Methods have been developed for allowing computers to detect unsafe behavior on construction sites, such as workers without hard hats or safety harnesses, and monitor environments where heavy machinery, such as forklift trucks, are working in proximity to humans, enabling vehicles to be automatically shut down if someone steps into their path. About 2.7 million injuries are caused due to workplace accidents every year, according to the US Bureau of Labor Statistics. This is an area where businesses are increasingly investing to reduce the human and financial costs caused by oversight or inattentiveness.
Of course, preventing the spread of illness caused by viruses also is an important cause these days, and computer vision technologies are increasingly being deployed to monitor compliance with social distancing requirements and mask mandates.
Computer vision algorithms also have been developed during the COVID-19 pandemic to assist with diagnosing infection by looking at chest X-ray images for evidence of infection and damage to the lungs.
Shopping and retail are other aspects of life where we are sure to notice the increasing prevalence of computer vision technology during 2022. Amazon has pioneered the concept of cashier-less stores with its Go grocery stores, which are equipped with cameras that recognize which items customers are taking from the shelves. More branches will open throughout 2022, and other retailers will jump on the bandwagon, including Tesco, which will open the United Kingdom's first checkout-free supermarket.
In addition to relieving humans of the responsibility of scanning purchases, computer vision has several other uses in retail, including inventory management where cameras are used to check stock levels on shelves and in warehouses and automatically order more inventory when necessary. It's also been used to monitor and understand the movement patterns of customers around stores to optimize the positioning of goods and in security systems to deter shoplifters.
Another increasingly popular use allows customers to get information on products by scanning barcodes using their mobile phones. In fashion retail, one application of computer vision is the “virtual fitting room,” which enables shoppers to try on items without touching them. Cameras in the mirror superimpose images of the clothing on the mirror’s reflection and can identify products customers are trying on and suggest matching accessories to go with them.
Connected and autonomous cars
Computer vision is an integral element of the connected systems in modern cars. Although our first thoughts might be of the upcoming autonomous vehicles, it has several other uses in the existing range of connected cars that already are on the roads. Systems have been developed that use cameras to track facial expressions to look for warning signs that we may be getting tired and risking falling asleep at the wheel. Since fatigue is said to be a factor in up to 25 percent of fatal and serious car crashes, it’s clear that measures like this could easily save lives.
This technology already is in use in commercial vehicles such as freight trucks, and in 2022 we could see it start to make its way into personal cars, too. Other proposed uses for computer vision in cars that could make it from drawing board to reality include monitoring whether seatbelts are being worn and even whether passengers are leaving keys and phones behind as they leave taxis and ride-sharing vehicles.
Of course, computer vision will play a big part in self-driving vehicles, too. Current thinking is that it will be the most important onboard element of autonomous navigation. Tesla announced this year that its cars will rely primarily on computer vision rather than lidar and radar, which use laser and radio waves, respectively, to build a model of the car’s environment.
At the edge
Edge computing describes systems where computation is carried out as close as possible to the data source. It’s a term used in contrast to the paradigm of cloud computing where data is collected via sensors and sent to centralized servers for storage and processing. In the domain of computer vision, it is an increasingly useful concept, as computer vision systems often do jobs where action needs to be taken immediately and there isn't time for data to be sent to the cloud.
Edge computing in relation to computer vision has important implications for security, an important factor to consider as businesses face tighter scrutiny and regulation over the way video data is captured and used. With edge devices like computer vision-equipped security cameras, data can be analyzed on the fly and discarded if there is no reason for it to be kept, for example, if no suspicious activity is detected.
Bernard Marris a technology consultant and contributing writer for AMCS and will be a guest speaker for the AMCS Inspire Series Executive webinar "How to unlock the advantage of AI to transform your organization with automation and actionable insights" June 14. Register to join the webinar: http://ow.ly/ZM6J50JlPR7.