摘要： 从路口视频图像中实时地构建道路背景图像是基于背景减法的车流量检测的前提。由于路口车流信息复杂，图像受外界环境因素影响大，现有的背景估计算法都有不足之处。该文提出了一种改进的频度背景估计算法，通过离散样本图像，实时地构建背景。该方法不仅算法简单、实时性高，而且增强了在背景估计时对光线变化的适应性，提高了背景建模的准确度，可适用于基于视频的路口信号灯的智能控制系统。

关键词: 背景估计, 智能交通, 平均估计, 频度估计, SVM

Abstract: Modeling road junction background from traffic surveillance video is the foundation for real-time traffic flow detection. The speed and accuracy of the models are subject to the complexity of traffic flow at junctions and the environmental influence, e.g. changing illumination. The paper presents an improved frequency estimation algorithm to construct junction background in real-time from image sequences. Experimental results show the proposed method can produce high modeling accuracy under different environmental conditions.

Key words: Background estimation, ITS, Average estimation, Frequency estimation, SVM