Hi
SIFT , Surf , Variations of LBP methods
Like ALBP captures directional statistical information for rotation invariant texture classification .
LBPV is also well suited for his problem.
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=4761847&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D4761847
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1410446&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D1410446
Hi All,
Thanks a lot Tunc Guven Kaya and Chethankumar BM. I would like to handle the person rotations of 45, 90 and even 180 degrees. In the case of of 180 degree rotation of person, its not possible to find face features. What features shall be useful in that situation? Pls suggest.....
The way I for this is using a tensorial model to explicitly model viewpoint variation, e.g.
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1640960&tag=1
You could try some simpler features like color histograms or something that can account for shape (maybe segment the person and calculate Zernike moments). Color will mostly be preserved with rotation around a person (assuming their clothing is similarly colored from different angles) and some aspects of shape could be preserved (a tall person versus a short person, for example).
You might also want to look at metric learning to see if it can help. I've had some success with it before in a different application. The idea is that if you have a training set where you have pairwise binary labels (i.e. sample n and sample m are the same person, but sample j is different from sample n) you estimate a distance metric such that positively labeled pairs have low distance and negatively labeled pairs have high distance. Once you have an initial feature set, this could help you evaluate the differences between the features from different views.
Hi All,
I have gone different articles on Person Re-Identification and found that they measured performance of their methods by plotting Rank Score versus Recognition Rate. I understand Recognition rate. What is Rank Score? How to compute it?
Hello,
Apart from the interesting previous answers, other features could be applied.
We have recently published a very interesting work about multiple people tracking from single view camera. This work is focused on detecting suspicious behaviours in shopping malls. In this work, we use several features with the aim of re-identifying people in different points of view. Mainly, we compute a combination of GCH, LBP and HOG features. You can check our paper online here:
http://www.sciencedirect.com/science/article/pii/S0957417415004182
In the following video, you can better understand the proposal:
https://www.youtube.com/watch?v=Rqk-vFKzAcQ
Good luck with your research!
----------
For more information, please visit my Webpage:
http://www.robesafe.com/personal/roberto.arroyo/
Article Expert Video-Surveillance System for Real-Time Detection of ...
Hi friends, I think you must use face properties that no changed such as Nose properties
Hello Mr. Mohammadali Ebrahimi,Thanks for your comment on the question. Nose will not be visible when person rotates by 180 degree even not visible on a rotation of 90 degrees. Anyways if you have some idea plz share so that I can explore furher.
Thanks....
- Badges
- Science topic
- Similar topics
- Analytical Chemistry
- Extraction