A novel illumination-invariant change detection method of shading model based on Cauchy distribution for visual surveillance systems is proposed. It is assumed that the observed temporal intensity variation of each pixel in background images is caused by white noise. After each image being normalized by an initialized Gaussian background model, the distribution of the intensity ratios between corresponding pixels of two background images obeys a Cauchy distribution. The parameter estimation of the Cauchy distribution model is simplified. Based on the change detection, the intensity, hue, and saturation in the YCbCr color space are employed to recognize and eliminate shadows and reflections in video sequences. The experimental results demonstrate that the proposed method of background modeling can tolerate the whole or local sudden or slow changes in illumination, and noises caused by some small motions, shadows or reflections in a background scene.