Computer vision, project ideas?

Question

Hi all

I am a undergraduate student and its my final year in this program. as in any computing degree i have to do a project(individual) regarding any topic which will come under computing.I want to something which will come under computer vision(object detecting or tracking to be exact).While searching about information on this topic i found out that there are already so many people who have done these types of projects.my question is this if i to do a project like tracking an object in video do i have to come up with my own algorithm or are there any algorithms available for me to do the coding.(i am familiar in java and started a bit of python)

Please guide me in picking up a topic and a small idea of how to start or from where to start

Answer 1

Here was my project: paper, demo Was good enough to send me to japan apparently ^.^

Anywho, we used C++ and OpenCV. I think OpenCV has bindings for other languages too (like Python). Matlab is also really good for prototyping computer vision stuffs. OpenCV has a lot of algos that will help you with computer vision, but CV is a tough field. There's no "works" or "doesn't work", and a lot of "it kinda works". So good luck!

Here are some other projects our school's vision lab has done, if you need some more ideas.

Answer 2

There are many computer vision and image analysis algorithms already in existence. I took a class on it in grad school a few years back that was interesting, so I suggest looking through your university library or bookstore for a text on the topic to get a good handle on what's available.

There're real-world applications for this technology. Optical Character Recognition (OCR) is one field that has seen some high-profile application in a big way.

You picked an interesting topic, have fun! :)

Answer 3

I'd done a basic course in computer vision during my grad studies and one of the first projects we did was to implement a system that stitches a series of images into a seamless 360 degree panorama.

It involved:

1. detecting discriminating features in the individual images (using SIFT feature extraction),
2. the best matching features in the other images (feature matching),
3. automatically aligning the pictures (homography estimation),
4. determining their overlap and the relative positions of cameras (camera pose estimation),
5. project the images into a cylindrical coordinate system (image warping),
6. and then, finally, blend the resulting photos into a single seamless panorama(image blending).

The challenge with this project is to make the code efficient enough to allow for fast image stitching.

You can find a lot of resources on the internet to help you out with the project.

Answer 4

If you want to do something really cool and interesting, try developing an activity detection algorithm in video. For example, "man leaving car", "people entering building" etc. It is not a trivial task and challenging enough for an UG thesis. You could use a toolkit such as OpenCV to do the ground-work in video processing and object detection etc, while you focus on the algorithm.

Answer 5

Hi,

a very hot topic is blurring faces in images like Google streetview.

There are dozens of websites that need automated face detection and blurring. My open source project for example.

Interested in doing an open source project?

A starting point with python and openCV:
http://creatingwithcode.com/howto/face-detection-in-static-images-with-python

Jan

Contact me for details: janmartin AT diy-streetview DOT org