2

In current project we've to do some image recognition in order to verify national ID Document. Nothing very complicated:

  • first find 4 crossing lines to find an contur of the ID card
  • then match 3 or 4 patterns
  • then find face
  • and run in portrait mode

What we use at the moment is openCV, but the issue is that we run this in background after taking the photo. Our requirement is to run it realtime, so the users sees the results on the top of camere preview.

Beside openCV we're now testing BoofCV, which has a little better performance, but it's still not very good. Image preview should be smooth, which it isn't on openCV nor boofCV.

We've also tested CraftAR, but well... it recognized 1 (with score 20) of 4 patterns when supplying the same picture the patterns where extracted from.

Do you maybe know anything else we can test? It can be paid or free, doesn't matter.

Phantômaxx
  • 37,901
  • 21
  • 84
  • 115
esh0.pl
  • 61
  • 1
  • 6
  • The CraftAR image recognition is intended to recognise images, drawings, objects, etc.(https://support.catchoom.com/customer/portal/articles/1894599) ID cards have a common structure and design plus some text (which is not very helpful for the image recognition). You could use it to recognise the photo in the ID card but not to verify the text. You'd need to use OCR for that. – martin Nov 23 '17 at 14:09
  • make sure you turn off debug! It will slow down your code by about 2x. – lessthanoptimal Aug 16 '18 at 00:20

1 Answers1

0

You ca also test this on the CoreImage and Vision frameworks of Swift, on iOS devices. If you have a high-end device, it might be possible to get at a reasonable frame rate that works on camera preview. Face and face landmark detection can run at around 15fps, if you get the image size and other parameters right.

Totoro
  • 3,398
  • 1
  • 24
  • 39