针对视频图像易受噪声干扰和背景变化复杂的特点,改进传统Census变换特征值对中心像素的依赖问题,建立Census模板以保持Census变换对光线变化的稳健性。将改进后的Census变换特征值、图像像素值、更新频数、最近更新时间和动态指数等多种特征融合,建立了一种新的背景建模算法。利用帧间亮度差,自适应选择融合多种特征更新背景模型,依据动态指数衡量背景变化复杂程度,建立不同的更新规则,提升模型对光线突变和复杂场景处理的稳定性。经测试多组标准视频序列,本算法检测精度优于其他算法,有效改善了光线突变对前景目标提取的影响,提高了对光线突变和复杂场景的稳健性,减少了运动目标的孔洞和像素漂移产生的假前景。

In view of the noise interference to video images and the complexity of background change, the traditional Census transform eigenvalue dependence on the central pixel is improved, and the Census template is established to maintain the robustness of the Census transform to light changes. A new background modeling method is established by combining the improved Census transform eigenvalue, image pixel value, update frequency, latest update time and dynamic index. The background texture difference is adaptively selected and fused with multiple features to update the background model. According to the dynamic index, the background change complexity is established, and different update rules are established to improve the stability of the model for light mutation and complex scene processing. After testing multiple sets of standard video sequences, the detection accuracy of this algorithm is better than that of other algorithms, which effectively improves the influence of light mutation on foreground target extraction, increases the robustness to light mutations and complex scenes, and reduces the false foreground caused by holes and pixel shift of the moving target.