光照变化下的前景检测

Foreground Detection under Illumination Change

前景检测是图像处理领域的基础研究之一,广泛运用于视频监控等领域。目前的前景检测算法大都能在光照缓慢变化的场景下具有较好的检测结果,但往往在光照变化剧烈的场景下(如开关灯)产生错误检测结果。提出一种基于具有光照不变性的颜色空间的前景检测算法,在新的颜色空间中通过对光照变化区域进行混合高斯拟合并在得到初步结果,最后对前景图像与背景图像的领域像素差值对比确定最终的检测结果。实验表明,所提出的算法在光照剧烈变化的场景下对前景的检测具有较好的检测结果。

Foreground detection is a fundamental problem in image processing, it＇s widely applied to video surveillance filed. In nowadays, most of the foreground detection algorithm can get satisfactory detection result in the scene that illumination changes slowly, but they often produce wrong detection result under illumination change dramatic scene（such as turn on and off the light）. Presents a foreground detection algo- rithm based on the light invariance of color space, uses the mixing Gaussian to fit the illumination change zone in the new color space, and then compares the pixel of foreground and background image to produce the final foreground detection result.