Welcome new user! You can search existing questions and answers without registering, but please register to post new questions and receive answers. Note that due to large amounts of spam attempts, your first three posts will be manually moderated, so please be patient.
Because of un-manageable amounts of spam despite our use of CAPTCHAs, email authorization, and other tools, we have discontinued this forum (see the 700k+ registered users with validated email addresses at right?). Please email us any questions or post bug reports and feature requests on GitHub at https://github.com/jevois -- The content below remains available for future reference.

Label recognition

0 votes
I want to recognize a set of items based their text  label ( label recognition) 1). Which module should I use ? 2) is there a way to pass a set of labels to the module so it can process them and recognize them ? 3) what would be the best approach to this problem?  Thanks.
asked Jun 8, 2017 in Programmer Questions by Tinob (200 points)
edited Jun 17, 2017 by Tinob

2 Answers

0 votes
I'm still exploring but I seem to remember that there was a digit recognition module used in one of the demos. I think it was the demo that did saliency, motion, faces, objects, gist and probably other things. Let me look in videomappings.cfg.  Oh yes,   YUYV 640 312 50.0 YUYV 320 240 50.0 JeVois DemoSalGistFaceObj

or maybe YUYV 320 252 30.0 YUYV 320 240 30.0 JeVois ObjectDetect

I guess it was the object matching that I remember as matching digits.
answered Jun 8, 2017 by Tcornall (700 points)
Thanks for your feedback !
0 votes
For labels the ObjectDetect module might work very well:

http://jevois.org/moddoc/ObjectDetect/modinfo.html

You just need to put pictures of the labels you want to recognize in the module's directory (see the doc at the link above), and they will be processed when the module starts up. We have done this successfully with, e.g., getting a logo like Priority Mail (shown in the module's doc) from the web (google images), saving the picture to that folder on the microSD, and then we can recognize it with ObjectDetect.

It would usually make sense to try to save the picture at a resolution that roughly matches what it will be when captured by JeVois (e.g., do not save a super high-res picture of the label if you will run JeVois at 320x240). Best results would actually likely be with first pointing JeVois to the label you want, making a screen capture, cropping that around the label, and then saving that crop to the microSD folder. This will maximize your chances that next time you see that same label with that same camera they will match.
answered Jun 13, 2017 by JeVois (46,580 points)
Fantastic. !!  I will give it a try as I am running in a hackathon in ny  !
Thanks again for the direction.  One more question. Is there a host based routine that takes a snapshot of the image to the specified size you recommended and then saves it in the miniSD directory you mentioned?  

Put differently which is  the best way to save and delete  images on the miniSD directory?  As I will be training the  camera ,  it would be great to have a facility that can speed up the image file add and deletion process
We created a couple of new things that should make it better for you:

- live access to the microSD while JeVois is running, see

http://jevois.org/doc/MicroSD.html

last section of the page.

- for this particular module, what you want is this: live grabs and training on the fly:

https://youtu.be/qwJOcsbkZLE

This one is not out yet, it is part of JeVois 1.2 to be released very soon (hopefully before the weekend)
...