After approaching my bathroom, I lifted my mobile phone into the air. Shake it a little bit, and then something wonderful happened. "Toilet," said a female robot voice. "Shower shower," she said soon. My mobile phone helps me see the world, which is really incredible. This feeling is no less than the first time I know that Dragon Dictionary software can recognize what I say, or that Facebook can accurately find my face in the crowd.
The free Android software I am using is called BlindTool, and its developer is Joseph Paul Cohen, a doctoral student at the University of Massachusetts in Boston. As long as you point your mobile phone at an object, BlindTool can recognize it in a few seconds, just like music radar software that can recognize the world around you.
"A few years ago, I worked for a project to care for the blind, and I began to pay attention to the needs of those who are visually impaired," Cohen said. "I had hoped to use technology to help them see the world, but at that time, there was no existing technology to do this."
Today, our computers are perfect in picture recognition technology. Through the neural network tools developed by Google or Microsoft, and the millions of pictures of daily necessities such as photocopiers and trash cans, they have the ability to identify items. At this point, they can do better than humans. However, most systems with similar achievements need powerful computers as support, and some even need the development of cloud technology. They are not very practical for people’s daily use.
However, BlindTool is a completely independent application for smart phones. How does it do this? In fact, there are still some flaws in its design. The most advanced picture recognition neural network system can test more than 37,000 kinds of pictures, while BlindTool can only test about 1,000 kinds (however, it is unfair to be too harsh because it still stores as many as 150GB of files).
As Cohen explained, the compromise made in this respect makes BlindTool run faster. What he tried was more perfect and applicable to the neural network of mobile phones. Under the framework of mobile phone operation, it can take up to fifteen seconds to identify anything. "Even if it takes five seconds, I will be dissatisfied with this application," Cohen said. At this point, the BlindTool takes only one second. Depending on the time the user points it at the object and listens to the voice, this time will fluctuate slightly during the specific experience.
BloodTool also makes mistakes. Besides, I often make mistakes.
However, as the price of rapid recognition, the recognition of BlindTool will make mistakes, and the number of mistakes is not small. In the process of walking around my apartment, it called my Christmas tree a long scarf for women, recognized an ornament on it as a bubble, and called the door a wardrobe. Sometimes its recognition result is very close to the correct answer, and sometimes it is far from it. This is because the neural network it uses is based on what Cohen calls "random selection" of pictures. This is an all-inclusive system with open sources, which is not entirely suitable for you or other people’s most common things at home, in the community or at work.
"It collects a lot of specific kinds of things, such as dogs, but the Christmas tree is not within the scope of the pictures it collects. So what it does is actually guessing what this item is, "Cohen said." The photo frame may look like a microwave oven, and the microwave oven looks a bit like a dishwasher. It will say that a coffee cup is a soup bowl, and the same cup may also be recognized as a powder cake for makeup. "
For these inevitable mistakes, Cohen’s current solution is to design this application software to evaluate the degree of confirmation. BlindTool vibrates when it can confirm the object it is aiming at. On the screen of the mobile phone (of course, the user can’t see it), the application will display a list showing what this item may be. Under its evaluation framework, the lowest possibility may be only 0.02%. Only when the possibility is greater than 30%, the software will broadcast what this item may be, and only when the confirmation accuracy reaches 90%, the mobile phone will give feedback in the form of vibration. After a few minutes of using BlindTool, you can intuitively feel its server mode, which is completely designed.
Although BlindTool identified my Christmas tree as a scarf again and again, when I really began to think about this phenomenon, I felt that the thorns on the branches of pine trees were indeed similar to the fluff of scarves to some extent. Round decorations on Christmas branches? It is basically the shape of a bubble; My front door? It is indeed similar in shape to a wardrobe. Even the wrong judgment made by BlindTool has some enlightenment for us.
"If someone can look around and think about what this object is in essence, they will certainly broaden their horizons and increase their understanding of the world. Maybe (technically) BlindTool will make mistakes, but it is right in understanding the world, "Cohen said." I still feel that it gives people a sense of independence, which is also an important goal of my research on it. "
A similar smartphone software called "Be My Eyes" allows volunteers with sound vision to help visually impaired people identify objects. This is a good system, but it can’t meet the needs of people with visual impairment all the time. At the same time, when people in need only need to find their mobile phones from their pockets, BlindTool also shows us the shadow of technological accessibility, which is the development direction of the future world.
(Translation: Wang Huinan)
· · · · · · · · · · ·
If you want to get more interesting content, please go to the homepage of the interface website (http://www.jiemian.com/) and interact with us on the Weibo, flirting with the cute bacteria → [crooked building -Viral] (please poke here).
You can also pay attention to the WeChat WeChat official account of Fun Channel: esay1414.
关于作者