In a problem of face recognition, I have a flow of owner's ids from the same face. This flow is generated from the application of the recognition algorithm on the stream of faces from the stream of images. On a given time, the following data structure is generated:
$$\{ \\ 1: \{ 0.90, 0.89, 0.70, 0.99, 0.95, 0.92, 0.93 \} \\ 2: \{ 0.85, 0.88 \} \\ 3: \{ 0.20, 0,75 \} \\ 4: \{ 0.82, 0.90, 0.99, 0.75 \} \\ \} $$
In this data structure, the outer number is the best guess of the recognizer about who is the the owner of the face. In the example, the same face from different images were considered from the ids (1, 2, 3, 4). For each id, there is a probability associated with how confident the recognizer if about its guess of who is the face owner. In the example, the recognizer analyzed one same face from 15 images and associated these face to the ids already mentioned.
Well, my question is: How can these stream of information be modeled in the way that I could get a consolidated probability that would allow me select the most probably owner face.