一种改进的基于QPSO的VIBE算法

与传统背景减除建模算法相比,视频背景提取算子(VIBE)算法无需估计背景数据的概率分布。所以,不像传统背景减除算法需要一定数量的训练视频帧,具有较小的运算复杂度与较好的精度,适用于嵌入式实时视频监测。某些场合下,视频流需要适时改变分辨率。然而,传统VIBE算法通常采用固定参数,不同分辨率的视频流,会导致固定参数的VIBE算法的检测精度下降。为此提出一种基于量子蚁群最优(QPSO)参数寻优算法,在初始状态对于视频流进行参数寻优计算,获取相应的最优参数,从而提高VIBE算法的学习能力。实验从定量和定性两个角度验证了本文改进的VIBE算法大幅度提升了针对不同分辨率的视频流分析能力。

An improved VIBE algorithm based on QPSO

Compared with the traditional background subtraction model, the VIBE algorithm does not require an estimation of a probability function for background pixels or a sequence of training samples. So there is a good balance between computation complexity and accuracy performance for the algorithm. It can be usually used in real time video embedded sequence surveillance systems. The resolutions of a video sequence can be changed under some circumstances. However, the fixed parameters method adopted in the traditional VIBE leads to degraded detection accuracy. We present an improved VIBE algorithm which obtains optimal parameters using the QPOS algorithm during initialization in order to improve the generalization ability of the VIBE. Experimental results verify the proposed algorithm quantitatively and qualitatively, which can greatly improve the accuracy performance.