摘要
低分辨率视频因像素密度低,关键帧细节被平均化,难以捕捉人体行为边缘轮廓的细微变化,导致行为特征区分度不足,使得难以清晰区分正常与异常行为。为此,提出低分辨率视频图像关键帧异常行为识别方法。对低分辨率视频图像展开增强,以提高像素密度,提取增强后视频图像的前景,并根据提取结果有效确定视频图像中的关键帧,获取关键帧图像内行人目标的边缘轮廓特征,从而确保视频图像质量能够保留足够的细节,提高行为特征区分度,以便清晰地捕捉和区分人体动作的细微变化,提高异常行为识别的准确性。将获取的边缘轮廓特征作为支持向量机的输入,对人体行为特征实施精准分类,以此完成低分辨率视频内人体异常行为的有效识别。实验结果表明,利用上述方法开展低分辨率视频异常行为识别时,识别精度高、效果好。
Due to the low pixel density of low-resolution video,the details of keyframes are averaged out,and it is difficult to capture the subtle changes in the edge contours of human behavior,resulting in insufficient differentiation of behavioral features and making it difficult to clearly distinguish normal and abnormal behavior.For this reason,the key frame abnormal behavior recognition method is proposed for low-resolution video images.En⁃hancement of low-resolution video images is carried out to improve the pixel density,extract the foreground of the en⁃hanced video images,effectively determine the keyframes in the video images according to the extraction results,and obtain the edge contour features of the pedestrian target in the keyframes,so as to ensure that the quality of the video images retains sufficient details,improve the differentiation of behavioral features,so as to clearly capture and distin⁃guish the subtle changes in human movements,and improve the accuracy of the recognition of abnormal behaviors.The obtained edge contour features are used as inputs to the support vector machine to accurately classify the human behavioral features,so as to complete the effective identification of abnormal human behaviors in low-resolution vide⁃os.The experimental results show that the above method is highly accurate and effective in identifying abnormal be⁃haviors in low-resolution videos.
作者
杨国萍
侯庆
蓝善根
YANG Guo-ping;HOU Qing;LAN Shan-gen(Guizhou Communication Industry Service Co.,Ltd.Guizhou Guiyang 550000,China)
出处
《计算机仿真》
2025年第7期533-537,共5页
Computer Simulation
基金
贵州省2023年科技支撑计划项目(黔科合支撑[2023]一般366)
贵州省高层次创新型人才(黔科合平台人才-GCC[2023]101)。
关键词
特征融合
低分辨率视频
分辨率优化
异常行为识别
Feature fusion
Low resolution video
Resolution optimization
Anomalous behavior recognition
作者简介
杨国萍(1991-),男(汉族),贵州贵阳人,硕士,工程师,主要研究领域为计算机视觉;侯庆(1980-),男(汉族),贵州贵阳人,博士,正高级研究员,主要研究方向:机器视觉、行为分析;通讯作者:蓝善根(1982-),男(畲族),贵州贵阳人,博士,高级工程师,主要研究方向:多模态、行为分析。
