computer vision - Draw ellipses on people using Expectation Maximization with OpenCV -

i have few doubts how approach goal. have outside camera recording people , want draw ellipse on every person.

right feature points of people frame (i them using mask have feature points on people), set em algorithm , train samples (the feature points extracted). number of clusters twice number of people image (i before start em algorithm using other methods such pixel counting codebook).

my question is

• (a) have train first frame , use predict in following frames? or,
• (b) use train feature points in every frame?

right doing option b) (i don't use predict) because don't know how use predict.

if a), can me , after how draw ellipses?. if b), can me drawing ellipse every person? since right know got different ellipses same person using cov, mean, etc (one arm, example).

what want achieve paper using gaussian model: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=5580105&tag=1&url=http%3a%2f%2fieeexplore.ieee.org%2fxpls%2fabs_all.jsp%3farnumber%3d5580105%26tag%3d1

if draw bounding boxes, rather ellipses, use function grouprectanlges merge different bounding boxes.

but, more important - people detection, can use opencv's person detector (based on hog) or latent svm detector person model.