- Home /
Augmented Reality : Recognize hand written number by the user?
Hello everyone,
Its been a few days I am trying to find a solution to my topic.
Basically I want the app to recognize a written number by the user.
The app will tell the user to write down for example number 24 on a paper and move the camera over the written number to see the 3d object.
This might be used for saving a Birthday, a wedding date .. etc
For accuracy, the app instructions will show the user a preview to tell please write the number 24 similar to this..
Although each hand writing will differ, but at least we do not get curly "2"-s or "4" with an open edge ..etc
So here we need AR to recognize the number, or be able to read the number according to approximation.
And the first question is: Is such a behavior doable or anyone familiar with a similar concept?
After searching similar apps, I found "Ink Hunter" apps for tatoo preview-s, although these apps use symbols not number, but we can think of a number as a symbol as well.
Also as this video: https://www.youtube.com/watch?v=9rXJcIE2Fcs shows, each user draws the symbol in a different way and still they get it working.
using Vuforia, they offer the free samples on the website, and there is one named "Text Recognition" , and here's the tutorial link: https://www.youtube.com/watch?v=W3MK6nC5FWE
But unfortunately couldn't make it work.
If someone has developed such a functionality using these sample projects from vuforia or have any ulternative method please I need you help :)
thanks in advance moghes
This is a fairly advance software topic called Optical Character Recognition. $$anonymous$$aking the computer read typewritten text can be difficult, reading handwriting is very-much harder. Some of the clues OCR uses to recognize character are the number of "enclosed" area's ( a "B" has two closed area's, a lower case "e" has one, an uppercase "E" has none. They also count number of straight lines (might not be helpful with handwriting) versus curved lines. The fact that a lot of the security on sites uses OCR to confirm you are a HU$$anonymous$$AN, shows just how difficult this can be to code, especially if any "noise" is introduced.
Were you able to complete this project as i am doing the similar project right now so kindly help me with this
Answer by Dave-Carlile · Jun 09, 2016 at 11:51 AM
This is something that requires some very sophisticated technology and you're going to need to find a product to help you with it. Google has their project to digitize all the things, so on the assumption that they maybe surfaced an API to the public I, uh, Googled it...
https://cloud.google.com/vision/
It does optical character recognition, but I don't know if it handles handwriting. But it should at least point you in the right direction for further research.
Thanks for the hint :)
Currently I am researching about OCR and the various tools used.
As for hand writing, the user will be offered an image to look up to and to write in a "similar" style.
$$anonymous$$y hand writing would be difficult for a computer to decipher even so. :p
Hi @moghes I am working single character recognition same as what you have done. Can you give me some tutorial or links that will help me to continue my work. I got stuck in between. Thanks.
Answer by Colorofjustice · Jun 10, 2016 at 12:09 PM
This is a great research question. I can see it requiring some type of very sophisticated machine learning algorithm. The mathematics behind would be quite beautiful in my opinion. As the others have suggested, it'll be best to find some already made technology by Google or Microsoft or some other well respected company and try incorporating into your game. That itself is might be a long task.
So a better question would be does your game/idea require this system? Is there a work around?