1

I'm new in OpenCV or Image Processing and looking forward for any opinions from who has more experienced in this field.

So in my problem, my target is to recognized some specific object. Let's say it is a drink can.

Here are my approach for the beginning :

  1. I will use contour to recognize the drink can shape
  2. Add colour filter to add more specific option for choosing only a specific colour drink can.
  3. Add filter for recognizing brand (maybe using cascade classifier or feature detector and It's friend), so It could be more specific to only pay attention to a drink can with chosen colour and brand.

My questions are :

  1. The drink can contour is not regular shapes. Can we define our own shape as the filter ? What kind of approach I will need ?
  2. After the step one is a success. How to add another filter (in step 2) only in the region produced by the step 1 ? Or the filter could be used altogether simultaneously ?
  3. What is the best approach to be able to recognized the object from 360 degrees using the camera ? I mean the app should be able to recognized it from any point of view.

I think there will be a lot of things I will need to do. And this is quite a broad question.

I just need a pointer. Maybe some of you ever done almost or even similar things in the past.

I already tried to experiment with cascade classifier. The problem is It's need a training for each point of view and not independent from rotation. I also tried to experiment with Feature Detector (ORB), Descriptor Extractor (ORB), and Matcher (BruteForce). But, It's really slow.. And only recognized an object only if It's exactly the same with the references image I gave. But, I could misuse or make a mistake during the experimetn to make it's performance reduced.

Anyway, any suggestion or opinion are appreciated.

Thank you.

Lyn
  • 507
  • 3
  • 15
  • I'm not sure if it can add your own shape, but the OpenImaj API has an easy to use image recognition tool. I prefer OpenImaj than OpenCV. However, I'm not sure if can recognize the shape of the can, but it can match features on an object. Here's a link if you wanna check it out http://openimaj.org/tutorial/sift-and-feature-matching.html – bmarkham Jan 05 '17 at 02:24
  • I'm looking into it right now and will consider about it. Thank you. The reason I use OpenCV because It's widely known and I assume there will be many people already have experiences with it. – Lyn Jan 05 '17 at 02:41
  • Are you going to be using video or still images ? – bmarkham Jan 05 '17 at 02:43
  • It's video using Android Camera. It will be an application. – Lyn Jan 05 '17 at 02:47
  • Okay well then the video part will be a bit complicated then. I suggest lowering the resolution when processing then blowing it back up to view. Will increase the FPS significantly – bmarkham Jan 05 '17 at 02:52
  • I believe so, although I still have not do any experiment with the resolution. Anyway, Thanks for the suggestion. – Lyn Jan 05 '17 at 03:03

0 Answers0