一种图像序列平稳性和相关性检验的天气场景分类方法

Stationarity and Correlation Test of Image Sequences Based Classification on Scenes with Different Weather Conditions

提出一种利用图像序列的平稳性和相关性检验的天气场景分类方法.首先,给出天气场景的定义和分类标准;其次,该方法通过分段逆序平稳性检验,提出图像均值子序列逆序总数数学期望和方差的计算方法,将天气场景分为平稳性天气场景和非平稳性天气场景;最后,提出自相关函数的分类检验方法,建立对待分类场景图像序列的激变描述,完成对其所属静态或动态场景的分类.该方法为非参数检验方法,推断分类标准时无需估计总体分布的参数,并能在线学习所得的分类标准.实验结果表明,该方法可准确完成对天气场景的动态分类.

Classification on different weather conditions provides first step support for outdoor scene modeling, which is a core component in many different applications of image sequence analysis and computer vision. In this paper, an objective classification method on scenes with different weather conditions is presented with two steps based on stationarity and correlation test of image sequences. First of all, scenes with various weather conditions are cdnsiderably described, on which an objective classification standard is accentuated. Secondly, based on the stationarity test on sub-sequences of intensity averages with counter order, the corresponding expectation and deviation of patterns are formulated and proved. Therefore, scenes with different weather conditions are primarily classified into stationary and nonstationary ones. Finally, a correlation test on autocorrelation function of intensity values in image sequences with different weather conditions is organized. Moreover, descriptions on sudden change of the autocorrelation function are established. Consequently, a classification on static or dynamic scene is ultimately accomplished. The two-step method needs no parameters, which avoids estimating the parameters of population distribution, when the inference of classification standard is in progress. This method is demonstrated to be effective using experiments on seven videos with different weather conditions, which contributes to latter applications such as scene modeling with different weather conditions.