Setting and training this network differs very little from the main neural network described in previous topic.
This network has less parameters and is much simpler to train and use.
Using this trained network you will be able to check any given combination in terms of how good it looks as a jackpot combination, based on previous draws.
Main Video Source
Integer value represents the USB web camera. If you have only one web camera connected, the number will be zero. If you have two cameras, the second is the number 1.
Fast detection of faces is shown on a main screen with a yellow box. The detected faces with this fast detection process may not be suitable for processing so you will probably want not to detect all those faces. This limit accounts only faces with the probability of being a face higher than this limit.
Every counted face gets its face encoding extracted and compared for difference with all faces stored in Known faces table. The person will be recognized if its stored face encoding is closest to the newly detected face, but only if it's difference is less than this limit.The smaller the number the smaller the difference.
The face is tracked in subsequent frames, the nearest to the previous position will be added to tracking that face if the distance is less than this number. This is useful if you have more faces in a frame, that should be enough to connect faces to the right faces from previous frames. The number is in pixels, so the correct value depends on your camera positions and monitored people behavior.
The same face won't be detected in every frame. The angle of the face may be inappropriate, it may be partially covered or blured. This number defines how many subsequent frames may be missing a tracked face before that face is considered gone from the input stream and added to the detected faces and processed accordingly.
Faces detected originaly detected are confined close to cover just the most important parts of the face. With this padding more of the face would be included for age and gender detection, but zero padding is quite satisfactory.
