Binarization Method Based on Pixel-level Dynamic Thresholds for Change Detection in Image Sequences

A new change detection scheme with enhanced tolerance to noises and illumination changes is proposed in this paper. In addition to computing a global threshold for an entire target image, a dynamically adjusted cumulative histogram is computed to find the most suitable threshold for every pixel in the target image. A binarization method is designed to integrate the global threshold and the dynamic thresholds. The adoption of pixel-level dynamic thresholds makes the proposed method more robust. Experimental results show that the proposed scheme is able to locate motion pixels more accurately and work well under various kinds of illumination conditions.