GSOC with VIGRA

[Paul Muresan]

Hello,
I would like to improve vigra. First of all I would like to propose an idea and then if you will not like this idea I would like to ask for sugestions regarding the image feature extraction project.
The proposed idea is an obstacle detection based on elevation map:

Obstacle detection is a an important issue to all vision systems. In the automotive field it can be used to detect road obstacles that would further need to be classified, reducing the number of hypothesis and improving the speed of the alorithm. In the industial inspection field, object can be detected on the manufacturing line.

The method of obstacle detection is based on elevation map. After computing the BM ( I will also implement a robust BM algorithm), the elevation map is computed grouping the points that have a certain hight. If we get more points in a neighbourhood we consider than point an obstacle and find his boundaries. This found object can be offerd for further processing (to a classifier for example, in order to classifie the object in certain classes, however the classification is not in the scope of this project. There are many methods that do this task of classification on the internet).

The proposed method is invariant to ground issues, unlike other methods( i want to say here that methods like UV disparity are prone to errors, for example if there is a bump in the road or a whole it can detect it as an obstacle, wheras the method proposed is invariant to such issues).

The algorithm will run in real time. ( > 10 fps). I really believe that the method proposed is a must to any system.

The second topic is regarding the image features for machine learning. There are many features that can be extracted from images and furthe be used in computation.
HOG, HOPE, SIFT,HARIS, LBP, Chain Codes, bag of features/ words, features like the mean and variation of the image after applying filters(gray scale filter, laplacian) etc.
I have worked with all the above mentioned filters, using them either from a framework or implementing them from scrach.
My question in this topic would be what type of feature descriptors would you desire ? How many ?

I hooe to hear from you soon!
Regards Paul

[Paul Muresan]

Would anyone please tell me if they think my idea is right or wrong or for the second project what would be more interesting to propose??

[Burcin Erocal]

Hi Paul,

thank you for your interest on working with VIGRA for GSoC.
Unfortunately, we had to remove VIGRA related ideas from our list
since the main developer will not be able to mentor projects this
summer. Also see my post to this list earlier:

https://groups.google.com/d/msg/lmnd-devel/Sb-iLaKcdRs/gmDpJPB18MYJ

Please take a look at the ideas list again and see if anything else
looks interesting.

Cheers,
Burcin