0

I am working on a Project that has to detect vehicles and classify them according to their shape (Pickup,Muv,Sedan,etc).
Its my minor project and I need to submit it in a month(obviously not much time left)
I am confused that should I use Image or video for input.

  • If Image->then can anyone suggest me some dataset that have 2 sets of each image (1.vehice+background,2.only background) so that the background subtraction is easier.
  • If Video->then tell me where should I start from so that I be able to segment moving cars and extract features from them for training.


The frames I am getting after using the "mixture of gaussian" contain vehicle that are not perfectly white/segmented (considering rest is black) so I cannot use their shape as a feature. can anyone suggest some good method for this and what things I need to learn in order to complete this project.
Any post of interest would be of great help,
Thanks in advance.

drift
  • 1
  • 2
  • Your question is too general to get much help here. You'd be better off coming back when you have problem with your implementation. – Ed Chapel Oct 06 '13 at 13:23
  • I cant get a start actually I am sorry for acting like noob but is it possible to get the above described dataset I dont know any other place for help...I know you guys are experienced so I am asking for just a little help...Plz – drift Oct 06 '13 at 14:02
  • It might help to share more about what you are trying. You mention "mixture of gaussian"... how are you doing this? What code can you share? What isn't working as you intend? – Ed Chapel Oct 06 '13 at 19:40

1 Answers1

0

There is already a similar problem for static images that is discussed here: object (Car) Detection and segmentation Also this paper provides good idea of the low level vision features used: http://people.csail.mit.edu/xiaoxuma/proj/vehi_reco/MaGrimson_ICCV05_VehicleReco.pdf

For the datasets: http://cogcomp.cs.illinois.edu/Data/Car/ and http://lear.inrialpes.fr/people/marszalek/data/ig02/

For the video part - i guess one can obtain different views of the same car, thus one can obtain a better classification of the car. For monocular cameras you have http://www-users.cs.umn.edu/~martin/papers/its_trans.pdf. Thus its important mention what are you parameters of the scene and camera(static monocular multiview etc) to better define your problem.

best of luck!

Community
  • 1
  • 1
beedot
  • 652
  • 5
  • 12