Abstract: It is a hotspot that detecting moving objects from complex traffic video sequences in ITS field. Background-frame subtraction method is widely used currently. To generate and update qualified background images is the core of the whole method. This paper proposes an improved method based on traditional mixed Gaussian modeling algorithm. This method makes use of the relationship between neighbored pixels to adjust the learning rate of parameter estimation in Gaussian model, which strongly enhances the improved algorithm the ability of anti-interference. It also proposes an improved method of shadow removing based on statistical texture character in shadow region. Experimental results show that the combinations of the two methods can accurately detect moving objects from complex traffic video sequences.

Key words: objects extraction, mixed Gaussian model, shadow remove

摘要： 研究复杂交通场景中运动目标提取的背景差减法及背景图像的生成。在传统的基于混合高斯模型的基础上，利用相邻像素之间的作用关系修正高斯模型参数估计的学习速率，使算法有较强的抗干扰能力。给出一种改进的基于阴影区域纹理统计特性的阴影去除方法。实验结果表明，2种方法的结合可以准确地从复杂交通场景中提取运动目标。

关键词: 目标提取, 混合高斯模型, 阴影去除

CLC Number: 

• TP393