It is good to see that you have approached the computer vision task in a systematic manner.

I would suggest the following as a reading for further directions

http://arxiv.org/pdf/1303.4803.pdf

http://sigport.org/sites/default/files/report.pdf

http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5674053&tag=1

Dear Muhammad

I found to you some useful new and good papers , please follow the next links:

paper:- http://www.ncbi.nlm.nih.gov/pubmed/20426028

I suggest to contact above paper author directly bu follow the personal link

https://www.researchgate.net/researcher/13201919_Bram_Platel/publications/2

and request full text from the link:

https://www.researchgate.net/search.Search.html?type=publication&query=Cervical+vertebrae+tracking+in+video-fluoroscopy+using+the+normalized+gradient+field.

I attached to you some very useful papers

Please do not hesitate in any inquiry

Thanks

In my research I had once used the anisotropic diffusion for the MRI medical images. See if it helps u

use anisotropic diffusion

Ma'am Poorvi I need help in object tracking, I'm done with the segmentation part.

If you have a segmentation of the first frame, then the tracking can be done using a segmentation method that allows you to rigidly displace each vertebra segmentation within a small area to be adapted to the position of the vertebrae in the new frame. This takes into account that the vertebral shape should not change in the different video frames.

For instance, you can use a modified version of Active Shape Models (ASM). The idea is to initialize the segmentation of each following frame with the segmentation of the previous frame and then apply the ASM method, but instead of projecting to the statistical shape model space, you can just use a rigid registration instead, use procrustes alignment without the scaling factor. You could also do the same with Active Contour Models and Procrustes afterwards.

You could also try to use a set of rigid registrations (one for each vertebra), penalizing the translation factor, since you want each vertebra to get adapted to the one that is closest  to it in the next frame. However, this is usually computationally more expensive. 

Hope it helps.

Hi Muhaammad

I'm probably very late to this and you have already found your solution but i thought i'd highlight some of our work in which do exactly what you ask for. 

http://www.medengphys.com/article/S1350-4533(16)30042-X/pdf 

Also here is a PhD Thesis by Jonathan Branney on using these techniques to quantify cervical motion using video fluoroscopy

http://eprints.bournemouth.ac.uk/22039/