Security cameras are everywhere: building entrances, convenience stores, the streets of our towns and cities; always vigilant, always on the look-out for crime. No one can doubt that they play a significant role in making the streets safer, or that they bring a sense of security to our daily lives.
Yet every single camera generates huge amounts of video data—and as that builds up it becomes a problem for the people who have to work with it. The only way to keep up with the growing volume is with visual data processing technologies and systems that are themselves faster and more efficient. Having all the data in the world means nothing without technology that can fully utilize and analyze them with speed and accuracy.
But just how fast is fast? And with what kind of accuracy? In 2016, Toshiba introduced a data matching technology that is 50 times faster than other systems. So fast that it can find and identify the image of a single specified individual from among 10,000,000 images of people in a mere 0.008 seconds—and do so with an accuracy rate of over 98%. So it’s really fast, and it’s very accurate. And it also works with moving images, as this video demonstrates.
Using live footage taken on May 26, 2016, this 80-second video demonstrates how the technology can track the movements of a suspect and accomplice within seconds
As you can see, Toshiba’s technology is a real-world tool that can be used in criminal investigations in progress. Once a suspect is identified, the system instantaneously searches for similar facial features in video images from other security cameras in the vicinity. Almost immediately, the suspect’s current location and recent movements appear on the screen.
That’s not all. If it looks as if the suspect has an accomplice, the system can also visualize his or her movements. For example, it can find where they met if they started from separate locations, and follow both if they later separated. It’s essential information that helps investigators to frame their actions. Even in crowded places like stations or airports, criminals can no longer escape by blending into the crowd. The technology will find them, wherever they go.
AI × Deep Learning = ∞ (infinity)
Internationally, ultra-fast data matching using national biometric databases could assist in the detection of international criminals. Besides improving the efficiency of criminal investigations, Toshiba sees this technology to be applied in other situations as well. Toshiba is pleased to have developed a powerful crime fighting tool, but also sees the recognition technology as having many other roles to play in big data analytics: visualized sales data can be analyzed for demand trends; and fatal market losses avoided by monitoring and evaluating stock price fluctuations. Widespread application of ultra-fast data-matching technology to all kinds of data has the potential to contribute to better lives and to yield significant economic progress.
Fully aware of this promise, Toshiba now aims to expand utilization of the technology through the application of AI and deep learning. As Shinichiro Hamada, from Toshiba’s Software & AI Technology Center explains, “We are looking at ways to embed deep learning processes into the technology, which could yield further dramatic improvements in accuracy and speed. For instance, we believe that combining deep learning with data matching will make major contributions to the realization of city-wide intelligent information services based on data from sensors”
Shinichiro Hamada, data-matching technology developer at Toshiba
As the IoT makes further progress and enters into more and more spheres of life, the need for ultra-fast data-matching technology will expand significantly. In the age of big data, the blinding fast speed and the wide scope of Toshiba’s technology make it all set to become an indispensable part of life.
