Motion Detection with Adaptive Background and Dynamic Thresholds

Human motion detection is a fundamental part for many computer vision tasks and various methods have been proposed. Background subtraction is a very popular method where the classification of pixels into motion pixels and background pixels is based on thresholding the difference image between a background image and a current image. The choice of threshold is crucial and many methods have been employed, from a preset threshold for the whole image to adaptive thresholds for each pixel. In this paper, a motion detection method by dynamically thresholding the difference image is proposed. We keep a static background image that is updated periodically, and compare the incoming frame with it to obtain the absolute difference image. If through a preliminary judgement that there are moving objects in the scene, Otsu's threshoding method is used to find the threshold to binarize the difference image. Experimental results demonstrate the effectiveness of our method.