An Extended Color Difference Histogram Method for Object Detection
Xiaoyu Zhang · Journal of Information and Computational Science · 2014
As current algorithms used for object detection under complex environments have deficiencies in accuracy, memory and computational complexity, an extended color difference histogram method is proposed in this paper. This algorithm detects foreground in a way of coarse to fine. In the coarse detection step, the probable background pixels are removed in real time to reduce computational complexity needed by subsequent steps. In the fine detection step, the potential foreground pixels are further classified cautiously. It effectively decreases the complexity of computation and conditionally solves the problem of camera shaking or illuminations variations, which cause many background subtraction methods failing. Experiments show that the proposed approach performs well in various situations such as moderate, sunny, dim, camera shaking cases and a light switch case which contains drastic illumination variations.