View Full Version : What is the best way to implement Object Recognition

12-22-2008, 02:36 PM
In college, we have to submit something called a mini project at the end of our 6th semester. I have some experience in autonomous robotics, using basic IR sensors, and I am seriously considering doing something in the area of Object Recognition(OR) to go up to the next level.
Unfortunately, I haven't really seen any projects in this area myself, and similar projects haven't been done in the recent past in our college(as far as I know).
What I want it to do is (for starters) to locate a specific object, say a ping-pong ball, against a plain surface. Eventually, in my final year, I would like to guide a basic robotic platform using this OR system.

Firstly, please tell me if this(only the OR part) is feasible - I'll have about 1 - 1.5 months to do it.
Also, where can I find information about how to implement an OR system?
I know this much:
1. We need a CCD camera
2. We need to grab frames from the output of this cam at approx 25 frames/second. - (Problem #1: How do I do this? What H/W or S/W?)
3. Accessing this info from a computer.
4. We need to run a matlab/labview program on it. (Problem#2: can the data from the frame grabber be accessed by the program?)

I am particularly interested in building this system on a Linux machine.

Can we use any other language, with a c/c++ syntax?

Please add any other info, especially about interfacing the cam with the comp.
I've been looking for info at this site and others, but I can't find anything specific.


12-24-2008, 02:53 AM
That's a pretty neat way of putting things on the post.

So, here you go:

Since this will be your first project using Computer Vision, I suggest you stick to the basics and then as you said carry the idea forward for your final year project. By that I mean, don't get distracted and try doing a LOT of things in one go - you seem to understand that, but I still reiterate.

So, a good idea would be to start off with something like this -

The second link esp. has good info. about the necessary background and nice links at the end for further reading.

Reading the stuff on the above two links, you should be able to answer all 4 of your questions.

You can do all of this on a Linux machine using C. Read about OpenCV. It's a pretty neat and well-documented library for Image Processing/Computer Vision. You shouldn't find it too difficult to find tutorials for setting up and using OpenCV on a Linux machine to do the above tasks.

My suggestion would be to use your hardware experience to build a good robotic platform for you to experiment with the vision part. Something, that you will be able to utilise for your final year project as well. For some ideas/inspirations look at -

All the best!

12-26-2008, 05:17 PM
Thank you very much for replying. The links seem very useful.