Computer Vision: Opening Our Eyes to the Potential Of AI
Although many people have now heard of computer vision, they often still struggle to understand how it could affect their daily lives. Driverless vehicles, for instance, are still a long way from mass acceptance, while the use of facial recognition has slowed as technologists and lawmakers wrestle with implications on civil liberties.
Look a bit further, however, and it’s plain to see that computer vision has a role to play in nearly every walk of life imaginably. Below are six examples where technology is revolutionizing our world, from recycling to insurance, and even the way we shop for groceries.
Recycling: Studies show that more than three-quarters of plastic waste in the U.S. end up in landfills rather than being recycled. The problem is simple. Manual recycling is both tedious and prone to error. Using computer vision and machine learning, some recycling facilities are now using robots that can sift through some 80 items a minute, reducing errors and speeding up the process substantially.
This is similar to software found in self-driving cars that “navigates” the road and its surroundings. When recycling, the software “sees” incoming materials, determines whether they are waste or reusable, and decides what to do next. It might use a robotic arm to remove the item from a conveyor belt or leave it for further processing.
Agriculture: The United Nations predicts that the world will need to increase food output 50% by the middle of the century in order to address global hunger. This doesn’t necessarily mean that we need to grow more crops, but we do need to develop better ways of utilizing and redistributing our existing resources more effectively. Computer vision can be instrumental in this mission.
Farmers already use computer vision and satellite imagery to determine whether their crops have been attacked by pests and to respond before an infestation gets out of control. In another example, drones can send real-time video for analysis so that farmers can improve crop health and yields. Combined with sensors in the soil that measure nutrient levels, technology like this promises to transform the agriculture industry in the coming decade, a necessary step towards combating global hunger.
Retail: How can someone enter a store, pick items from the shelves and pay automatically without ever stopping at a cashier? The answer, once again, is computer vision. Cameras installed in the ceiling identify shoppers, follow their movements and track what they have purchased using object detection algorithms.
This information is then cross-referenced with weight sensors on the store shelves that identify how many items were taken. The cameras themselves are relatively simple. The algorithms themselves do the heavy lifting by crunching numbers for the next generation of retail experiences.
Finance: As anyone who has opened a bank account knows, it can be a frustrating journey to get things moving. While some traditional banks still require a branch visit to verify an applicant’s identity, the majority of challenger banks have introduced computer vision to facilitate this process. This means that new customers can set up their accounts from their smartphones in minutes. They simply join a video call with an advisor who can validate their identity documents using automated facial recognition and document checking solutions.
By making this step more convenient, banks and other financial institutions that leverage progressive technologies are much more likely to attract more and more customers. This is especially true in younger cohorts who have grown up expecting fluid digital experiences.
Car Insurance: If you have ever rented a car, you are well aware of the dreaded vehicle inspection process. The person at the front desk hands you a document with the outline of a car showing existing dents and other areas of wear and tear. You then spend several anxious minutes circling the vehicle matching this damage and looking out for any other issues. Get it wrong, and risk being charged an exorbitant fee upon return.
What if computer vision could improve this process by automatically scanning the car at pick up and drop off and comparing the before and after images? This technology already exists. Currently, it is largely used by insurance companies to make accurate damage assessments but car rental agencies are getting in on the opportunity. Rental giant Avis has tested the technology, using ordinary CCTV cameras to take 360-degree scans of vehicles as they enter and exit rental facilities. As well as speeding up the checkout and return of vehicles, the software was more successful than manual inspections, spotting 22% more damage incidents than employees.
Space: Geospatial imagery has been with us since photographs were first taken from very early aeroplanes. Now aerial photography, from drones to satellites, plays a critical role in many diverse industrial sectors. These range from agriculture (see above) to construction, to military and defense, to the natural sciences and wildlife conservation, where images are used to count populations of endangered species such as seals and elephants.
Unsurprisingly, however, geospatial imagery can be more difficult to work with than other types of visual content, presenting several barriers for such industry use cases from being accomplished efficiently. Many systems struggle to prioritize and detect accurate and relevant information from such detailed imagery, while others struggle to find solutions capable of operating in the low-latency, hardware-intensive environments where such content is generally captured.
In the past, developers have used complex workarounds to filter out unimportant parts of geospatial imagery, while passing on the more problematic images to human specialists for judgment. Today, however, the latest computer vision systems from companies like us are able to analyze this intricate data which a much higher degree of precision and accuracy via models that are lightweight and suitable for even the most hardware-intensive environments.
Such modern technologies employ a technique known as “Few-shot Learning,” which allows models for very specific concepts to be trained using small amounts of data. For instance, satellite manufacture could train a new detection model for waterborne shipping containers using fewer than 50 examples. This allows for highly technical problems to be solved in a manageable fashion, even where training data is very limited. Additionally, these models are designed from the ground up for efficient deployment onto edge devices.
These are just a few examples of how computer vision is set to transform different aspects of our lives. Even if your specific industry is not one of those listed in this article, you can be sure that this technology will have a role to play in the future of your business.
Put another way, vision has always been a fundamental aspect of human life, one without which few business sectors could operate. By combining vision with AI, new technology is opening our eyes to remarkable new commercial opportunities for whoever can see them.