Accurate detection of moving regions via a nested model

A nested model with multiple features is represented to accurately detect moving objects and their shadows in outdoor and indoor scenes, even when the moving objects and their shadows have similar color to the background or when there is background motion. Bag of features (color, texture and movement patterns) is first extracted in each frame of an image sequence, and its temporal variation is calculated in the consecutive frames. States of the bag of features are then selected and updated according to the statistical computing during all the previous frames. Pixels with the most stability are considered as background, and moving objects and their shadows are therefore extracted. Experimental results on our captured image sequences and public image sequences demonstrate the efficiency of our method.