基于改进K均值的运动目标检测算法研究

背景建模是运动目标检测的关键环节，提出了基于改进K均值背景建模的方法，并进行前景提取．该算法在HSV颜色空间对视频流的前N帧中的每个像素样本进行K均值聚类学习，K均值聚类的结果用来表示该像素螅背景模型；接着输入的视频流像素与背景模型比较，进行背景、可能前景和阴影的分离，并提出了一种像素相关的选择性背景更新机制；然后利用TOM（Time Out Map）方法来消除鬼影现象．实验结果表明该算法能够很好地对背景进行建模，较精确地提取出运动目标信息，对光照变化具有较强的鲁棒性．

An Approach Based on Modified K-means for Moving Objects Detection

A key issue of detecting moving objects, an approach based on modified K-means to model back- ground is proposed. It learns from the starting N frames with K-means algorithm; and the results learned the background of the pixels. Following, it performs the separation of background pixels, probable foreground pixels and shadow pixels; through the comparison of the input pixels and the background model, a pixel-based selective mechanism of the background update is proposed. Finally, the ghost effects are eliminated by apply- ing the TOM method. The experimental results show that this proposed approach can well model the back- ground, and more accurately extract the moving objects, as well as more robust to the illumination changes.