Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It only takes a minute to sign up.
Sign up to join this community
I have tried asking this on another forum, but after 4 months there are no posted answers, so I am asking it here:
What is the current state of the art approach for determining the pose of a person (including close poses like a head shot, and hand shot, etc...), and for segmenting the person from the background?
The setting here is single still images.
First of all, speaking about pose estimation I would remember Kinect sensor from Microsoft Research. They use depth-map, but the approach seems to be general: generate a HUGE amount of synthetic data and use it to train a HUGE Random Forest pixel-classifier. And do it for different body parts independently. For just still images you will use different features, including HOG as well. For further information have a look here: http://research.microsoft.com/en-us/projects/vrkinect/
The second thing that I want to mention is the research of the group of Vittorio Ferrari: http://groups.inf.ed.ac.uk/calvin/publications.html . They do a lot of stuff for body detection and pose estimation. The basic idea is to build an appearance model of the body. As of particular interest for your problem there is a paper: http://groups.inf.ed.ac.uk/calvin/Publications/eichner12ijcv.pdf . I would pay some attention to how they use spatio-temporal features. And of course it is always good to check references.
Hope it will help you to start. If I'll remember some different approaches - I will update the answer.
I was reading about "Histograms of Oriented Gradients for Human Detection" according to Dalal, there are a lot of derivative works, I guess is very close to what you are looking for.
