Hi,
I am using EM class to train a single GMM depicting a small object in a video sequence for tracking use case. Provided that I get object samples for few number of frames initially, I use the RGB values of every pixel belonging to the object appearing in all the frame as feature vector and train the GMM with number of clusters = 3. Using the trained GMM model I try to find out which foreground blob in the next frame is most likely to represent the object that I have trained for and willing to track.
When I try to do so using EM predict2 function, I sometimes get positive values for loglikelihood returned. I am unable to comprehend the positive nature of the values since likelihood represents probability ranging from 0 to 1. Can someone please help me out with the basic understanding I am lacking about GMM / EM. How can I interpret the positive/negative values and take a decision whether the blob under consideration can represent the object or not.