基于YUV空间码本模型的运动检测

Motion detection method based on YUV codebook model

以YUV空间码本模型为基础,提出了一种融合码本聚类思想和信号能量分析思想的运动检测算法。先在YUV空间上进行码本建模,然后在码字训练阶段对每个进入码字的像素点的均值以及围绕均值波动的能量值进行统计,最后在前景检测阶段通过能量门限实现前景和背景的分割。实验结果表明,与以往的YUV码本模型相比,该算法除了具备快速实时和准确的特性外,对复杂背景环境中如晃动的树叶、水波和光线的突变等强干扰因素体现出更好的适应能力,并且在阴影的抑制方面也有较强的鲁棒性。

This paper proposes a YUV-color-space codebook-model-based motion detection algorithm which fuses the technique of codebook quantization/clustering with energy analysis of video signal. The proposed algorithm constructs codebook model for each pixel on the YUV-color-space. Then the mean value of each pixel fitting into the codeword and the energy value which fluctuating above and below that mean value is calculated on the stage of codewords training. The image frame is partitioned to foreground and background by means of energy threshold during the phase of foreground detection. Experimental results show that comparing with the YUV-color-space codebook method, not only does the proposed algorithm gain more speed and accuracy on the process of foreground extracting, but also show more resistence to strong disturbance resulted from shaking leaves, water wave and mutation of luminance in the complicated condition. Furthermore it has better robustness on shadow suppression.