Braille is a method of physical writing used to allow humans to read by touch — most commonly used as a substitute for printed text by those who may be visually impaired. Both displaying Braille and reading it is difficult to do with machines, but there has been a development in the latter area. A research team has trained a robot to read Braille at a speed far exceeding humans.
The robot was developed by a team at the University of Cambridge. Rather than trying to read Braille by touch, it instead uses a camera and an image recognition algorithm to do the job. Their solution is a bit ironic in a way, given the purpose Braille was created for. The robot can quickly sweep across a Braille display, working at a rate of up to 315 words per minute at 87% accuracy. That’s roughly twice as fast as a human reading Braille, with a similar level of accuracy. Some nifty de-blurring algorithms were needed to achieve this speed from the camera’s video feed.
We’ve also seen some impressive development on the other side of all those little bumps, with two Braille devices taking home awards during the final Hackaday Prize in 2023.
I bet I could beat them with just a web cam and open cv
With those black and white dots, sure. But matte black on black dots with harsh lighting might be tricky.
four quadrants lights around the webcam flashing alternatively to cast shadows should be enough for most cases. Bonus points if you sync the lights with the framerate.
yeah thats what I was kinda thinking, sudo depth data simply off of occlusion from known vectors, pump that into a couple of grayscales and take the average of the frames, bing bong got ourselves a valid sample
Yeah, but your paper wouldn’t include the right buzzwords.
On the sort of paper that is normal for braille where there is NO contrast? Just whatever cardstock it’s stamped into?
I rate your boast as at best optimistic. Moreso when you’re scanning braille that might be worn down from years of use.
Perhaps if a group was trying to scan a braille book for archival purposes, they could use a roller to darkened to tops of the bumps to prep the page for the scanner.
Thinking back to those toy drawing tablets, that used a soft fluorescent sheet which could be pressed against a hard plastic surface to produce bright lines (where it interrupted the total internal reflection), I wonder if a similar sheet could be gently rolled onto the page to prevent damage. That said, the force might be more damaging than the ink.
I don’t have experiance with that sort of machinery, but i do with braille. The idea of ‘here let’s ROLL something over the dots to paint them’ to me sounds like ‘here let’s just smash the dots down’
My negativity to one side? I’m glad these thigns are being worked on because accessibility ANYTHING is bullshit. expensive.
I wonder how good something like a leap motion or even a mobile phones’ facial recognition camera, might be at reading brail. They have a very high refresh rate and structured light density. I suspect they would be able to discern the surface imperfections with enough resolution to interpret brail from a reasonable distance.
There is actually a good project called Angelina reader which is able to digitize braille pages by a webcam and some ML.
https://angelina-reader.ru/?language=EN
We tried a specific setup for scanning old braille books at UAS Technikum Wien wothnoir students:
https://github.com/ic21b031/obr-scanner
This will make robots a protected group.
How so? (Or are you just venting your frustration that some groups receive protections that the rest of the population doesn’t need?)
While the image processing and recognition is impressive, potential use needs to take into account the degradation of these braille cells. Many devices such as Orbit reader on video use piezoelectric braille cells which tend to get clogged with dead skin and many kinds of human juice, making those plastic points get stuck in the upper position. Reading such “used” braille display with a finger still works well, since the user is pushing on those extended pins, correcting for any stuck points.
Could be useful in QA testing, though the market for these kind of devices isn’t that big.
I see. So humans with new sensors are less fast at reading Braille in comparison to the robot? Just kidding. 😁
Well, it could be because people do not train to read Braille by sight… Otherwise the results might be quite different.
Amazing robot I was really hoping it sensed the bumps. I’m struck that 87% accuracy is shared by humans..is it a similar figure for printed text?
> Braille is a method of physical writing used to allow humans to read by touch — most commonly used as a substitute for printed text by those who may be visually impaired.
This is bizarrely vague. Did AI write it? “Those who may be visually impaired” especially stands out. Surely they would know if they were visually impaired. It’s not like people learn braille and a week later realise they were never blind very often.
>”Some nifty de-blurring algorithms were needed to achieve this speed from the camera’s video feed.”
Nah, no self-respecting AI would say “nifty”.
“Did AI write it?”
No.
Aside from the somewhat sad irony… The whole point of this project is… What, exactly ? I don’t quite see how this helps anyone except perhaps the sighted that cannot decipher it….
My favourite piece of braille text is on a public toilet in the train station. It says “Door locked when red light is lit”.
At that point you’re just rubbing it in.
I’ve always found it kind of amusing that there’s braille on the drive up ATM’s.
That’s so you can keep your eyes on your phone.
“rubbing it in”? lol
Another issue with embossed braille is that it may be “printed” on both sides of each sheet, yielding a combination of bumps and dents.