Closed kukuruza closed 9 years ago
This is a bug in Matlab. With the new data flow, background is used only at the first step -- to generate background mask video from original video. After that, it's not used. So saving/loading the background object is no longer required.
It is better to train and save a background model separately. After that in the pipeline or during testing detectors, it can just be loaded. However, there is a problem with save/load.
The problem has been described here by me, and apparently no solution http://www.mathworks.com/matlabcentral/answers/169710-how-can-i-save-load-vision-foregrounddetector-object#answer_164713
We need to find a way to use trained models somehow