2

I have just started a school assignment where they want me to detect and recognize objects in a webcam stream, in an webgl application. It will be added to an already existing javascript plugin. It is important that this will be done in real time since the objects will change stuff in the application.

Example, if a user wears a yellow shirt with a specific icon on it will change the layout of the application.

I have researched this a few days now and found some intresting articles.

This seems like an intresting approach: http://research.ijcaonline.org/volume83/number3/pxc3892575.pdf

And ofcourse the SURF algorithm seems to be a legimate approach: http://www.vision.ee.ethz.ch/~surf/eccv06.pdf

So my question is "what algortihms might be best to implement?". And also if possible, which might be easiest to implement? I have quite limited time and this is only one of the objectives for this project.

I appreciate all the help and answers I can get.

edit. Surf is not acceptable because of patents.

Jonas Gerling
  • 95
  • 1
  • 10
  • don't know much about that, but when I had to use a motion detect app, I used motion http://www.lavrsen.dk/foswiki/bin/view/Motion/WebHome (but don't know anything about the algorithms used). Good luck ;) – arnaudbey Aug 31 '15 at 07:58
  • Why did you tag this as javascript if this is a java plugin? Javascript != Java. Make sure you are referencing and tagging the right technology so you are talking to the right community of people. – jfriend00 Aug 31 '15 at 08:00
  • Oh, I am sorry. Didn't realize I wrote java. It is a javascript plugin. – Jonas Gerling Aug 31 '15 at 08:03
  • 1
    Does the course you are taking gives no hint on how to do this? I would imagine that object detection relies on some sort of flood fill algorithm in essence that lumps similar colors together. Do you know beforehand the shapes and colors of objects you are expected to detect? – WacławJasper Aug 31 '15 at 16:53
  • Well I did not recieve any hints on how to to it, it is for my bachelor degree so I only get a vauge project description. The shapes and colors can be anything. I think I will create my own image database and keep it simple with squares, triangles and circles. – Jonas Gerling Sep 01 '15 at 09:54

1 Answers1

0

Please refer to the HAAR cascade or SURF based classifiers featured in this answer (replace "face" with "any object").

However, you would probably need to train your own classifiers which is not possible within a limited timeframe.

Alternatively, simplify your application and go with simple color tracking.

Community
  • 1
  • 1
le_m
  • 19,302
  • 9
  • 64
  • 74