融合空域信息的码本动态背景相减算法

Improving Codebook-based Dynamic Background Subtraction by Integrating Spatial Context

背景动态纹理是导致传统背景相减算法性能下降的一个重要因素。为了区分背景动态纹理导致的图像运动与真实目标导致的图像运动,同时确保目标检测的完整,对码本编码的过程引入空间位置信息,实现近邻像素间码本模型的共享。这样不仅利用了动态背景的时间准周期行为,还兼顾了它们的空间准周期行为。在保证检测率的基础上,降低了虚警的产生。实验结果证实了提出算法的有效性。

A major challenge for most traditional background subtraction algorithms is how to suppress the background dynamics.In order to deal with the disambiguation of image motion induced by the background dynamics and image motion influenced by the truly moving objects,while guaranteeing the detection accuracy,observable pixels are allowed to match codebooks of their neighboring pixels,and therefore neighboring codebooks can share with each other.With the help of the improvement,the new codebook algorithm eliminates the false positives largely without unreasonably raising the false negative rate.Experimental results validate the proposed algorithm.