Object detection using a moving camera under sudden illumination change

In the recent years, various background subtraction methods have been proposed and used in vision systems for moving object detection and tracking; however most of them are sensitive to illumination change and have difficulty in handling shading and shadows caused by illumination change. Although there are some algorithms to handle illumination change, they need time on the order of several frames to estimate and train the background model and, in the majority of surveillance applications, there is no such time especially when the continuous detection of moving objects after a sudden illumination change is required or if objects of interest move fast. This paper presents a robust background subtraction method which is able to cope with sudden illumination change. Our algorithm is based on the key observation that statistical background model used for object detection right after a sudden illumination change can be inferred from the model before the change sufficiently accurately to allow continued detection without delay for model re-training. The algorithm was tested on both indoor and outdoor video sequences from different datasets. Experimental results show this approach works better than the state-of-the-art algorithms in background subtraction.