Computer Vision Is More Than Just Image Recognition
Source: Ken Weiner
Computer vision and image recognition are terms that are often used synonymously, but the former encompasses more than just analyzing pictures. That’s because, even for humans, “seeing” also involves perception on many other fronts, along with a lot of analysis. Humans use about two-thirds of their brains for visual processing, so it’s no surprise that computers would need to use more than just image recognition to get their sight right.
Of course, image recognition itself — the pixel and pattern analysis of images that computers undertake to see what’s there — is an integral part of the machine vision process that involves everything from object and character recognition to text and sentiment analysis. But as Cornell Tech computer scientist Serge Belongie pointed out at the recent LDV Vision Summit, at its core, today’s image recognition still mostly just identifies basic objects such as “a banana or a bicycle in a picture.” Even toddlers can do that, but computer vision’s potential is superhuman: to be able to see clearly in the dark, through walls, over long distances, and process all of that intake quickly and in massive volume.
Already, computer vision in its fullest sense is being used across daily life and business to conduct all kinds of functions, including warning drivers of animals in the road, pinpointing medical maladies in x-rays, identifying products and where to buy them, serving contextual ads inside of editorial images, among others. At my company, we use computer vision to scan social media platforms to find relevant images that can’t be discovered through traditional searches. The technology is complex, and just like all of the aforementioned tasks, it requires more than just image recognition, but also semantic analysis and big data.
So just what goes into computer vision besides image recognition, and what else is it being used for? Here are some examples and the technologies that help out.
Thermal Imaging
Humans can’t “see” heat or gas. In many cases — especially where fires, predators in the wild, or gas leaks are concerned — these are the types of dangers humans would want to see before they feel or smell them. Advances in thermal imaging mean that this capability is already built not only into portable cameras for industrial and consumer uses, but also into smartphones, as demonstrated by the Cat S60. Eventually, this capability will be integrated into every cell phone. But dangers aren’t the only things thermal imaging can help with. They can help keep sports honest, too, as evidenced by the thermal imaging cameras used to detect mechanical doping at this year’s Tour de France.
Sensors
Sensors that detect temperature, light, air quality, gas and motion are just a handful of the ones that computer vision uses to identify exactly what’s out there. For example, some of today’s smartest buildings use sensors built into the lighting and temperature systems to detect the movement of people so it can optimize light and energy levels, getting smarter over time. As well, home monitoring systems not only use motion sensors to allow built-in cameras to track the movement of your dog, but combine those with temperature and air quality sensors to get the full picture of what’s going on while you’re away from home. Meanwhile, in-store sensors and beacons, in combination with cameras, track shopper movements, cross-referencing them with “big” behavioral data in the cloud. The goal is ultimately to help retailers not only optimize store layouts and pricing, but also serve coupons to customers in real time.
| }
|