中文摘要：

为解决雨雪等动态背景严重影响目标检测和识别的问题，提出一种新颖的利用误检信息消除雨雪干扰的方法．首先采用背景差法进行初始运动检测，对每个像素点进行前背景分类，获得候选前景第一次分类的二值序列；继而统计该二值序列的码元跳变次数和高电平时间，将2个数值作为特征向量；然后利用K均值分类算法进行候选前景的二次分类；最终提取真实运动目标．实验结果表明，相比于背景差法，所提算法能够有效去除雨雪动态背景导致的目标检测干扰．

英文摘要：

To solve the problem that object detection and recognition will be seriously affected by dynamic background such as rain or snow, a novel method of eliminating rain or snow interferences through false detection information was proposed. Firstly, background subtraction was used to conduct an initial moving detection. Every pixel was classified as foreground or background, and binary sequences describing the first classification results of candidate foreground were obtained. After that the numbers of code hopping times and the last time of high level in those binary sequences wer counted and retained as feature vectors. Then a secondary classification was performed on the candidate foreground by K-means method and finally true object targets were extracted. In the experiments, it was shown that dynamic background interferences caused by rain or snow can be effectively removed compared with background subtraction method.