Abstract: Traditional Gaussian mixture model of RGB has some drawbacks such as bad detection on rate of low contrast pixel. Aiming at that, this paper proposes Gaussian mixture model of chromaticity coordinates, which brings better result in target detection in gait detection. Converting RGB into chromaticity coordinates, color contrast and detection on rate of low contrast pixel are enhanced. It adds lightness information to reduce shadow. In the part of target extracting, mechanism of noise suppressor is accessed. Experimental result shows the improvement gets half error detection rate than before at the same color contrast rate.

Key words: gait detection, Guassian mixture model, chromaticity coordinates, lightness information

摘要： 针对传统的基于RGB通道的高斯混合模型低对比度像素点检测效果较差的问题，提出一种基于色度坐标的高斯混合模型，使之更好地用于步态检测。该算法将RGB色彩值转换到色度坐标上，以强调色彩对比度，提高低对比度像素点的检测率，并增加亮度信息以减小阴影的影响，在前景提取部分，加入噪声抑制机制。实验结果表明，改进后的算法在相同对比度下，误检测率最多可减小一半。

关键词: 步态检测, 高斯混合模型, 色度坐标, 亮度信息

CLC Number: 

• TP391.4