中文摘要：

针对安防监控视频中出现的树叶遮挡异常,提出一种颜色与纹理特征相融合并结合支持向量机（SVM）的视频树叶遮挡异常检测算法。提取视频图像的HSV非均匀量化直方图作为颜色特征;采用分块思想提取图像块的均匀局部二值模式（ULBP）,将图像所有块的ULBP特征进行整合作为图像的纹理特征;将颜色特征和纹理特征进行整合,利用支持向量机进行训练建模,实现树叶遮挡异常的检测。采用真实安防监控视频库进行验证,验证结果表明,该方法识别率达到85.56%,受试者工作特征（ROC）性能指标达到0.9433,融合ULBP特征与颜色特征能很好地减少场景的干扰,对树叶遮挡异常视频进行有效识别。

英文摘要：

Leaf occlusion is a serious problem in security surveillance videos,especially in road surveillance videos.To solve this problem,a method based on color and texture feature fusion and support vector machine（SVM）was proposed.The HSV normalization histogram for the video frame was extracted as the color features.The video frame was divided into blocks.The uniform local binary patterns（ULBP）features were extracted from the image blocks and stitched together as the texture features.The color features and texture features were integrated to image features.Image features were trained and modeled using SVM.The proposed method was tested on the real security surveillance video library.Experimental results show that the detection accuracy reaches 85.56%and the receiver operating characteristic（ROC）reaches 0.9433.The proposed method can reduce the interference of the scene by feature fusion and detect the leaf occlusion effectively in security surveillance video.