I am trying to distinguish between candidates video segments in a shortlist. Say there are N number of candidates, and one of them is an important event.
Based on some training data, I am able to generate a 2-dimensional Gaussian distribution in matlab. This distribution is supposed to model UN-EVENTFULNESS. So, does it make sense if I just take representative data points from each candidate, and then put it into the equation of the Gaussian distribution to get a probability density measurement?
In the end, the candidate with the lowest density is considered the one to contain the desired event (since the Gaussian distrib models UN-EVENTFULNESS).
Does this make sense? Can it simply be done like that?