您的位置:首页  > 论文页面

基于视觉注意的人体动作检测

发表时间:2011-04-30  浏览量:1670  下载量:781
全部作者: 刘顺彬,郑慧诚
作者单位: 中山大学信息科学与技术学院
摘 要: 生物视觉注意机制能将视觉信息处理集中在场景的显著区域,从而有效提高处理效率。将这一原理运用到计算机人体动作检测中,利用一个基于灰度、颜色和运动等时空特征的视觉注意模型,有效地提取出视频中包含运动信息的显著区域。将这一视觉注意模型同一个基于时空梯度特征提取与子块匹配的人体动作检测相结合,可弥补传统方法耗时大的不足,并提高对噪声的鲁棒性。实验表明:利用该方法能有效提高人体动作检测的效率和准确率。
关 键 词: 计算机视觉;视频分析;动作检测;视觉注意
Title: Action detection based on visual attention
Author: LIU Shunbin, ZHENG Huicheng
Organization: School of Information Science and Technology, Sun Yat-sen University
Abstract: Visual attention in biological vision systems helps to focus processing on salient regions of the visual field. In this paper, visual attention is applied to action detection by combining with a patch-correlation approach. A visual attention model based on intensity, color and motion information is implemented to extract salient spatial-temporal regions in video. During the action detection procedure, only the salient patches or regions are correlated between the query video and the target video, which reduces dramatically the computational cost and improves the robustness. Experimental results of human action detection in video demonstrate the effectiveness of the proposed method.
Key words: computer vision; video analysis; action detection; visual attention
发表期数: 2011年4月第8期
引用格式: 刘顺彬,郑慧诚. 基于视觉注意的人体动作检测[J]. 中国科技论文在线精品论文,2011,4(8):698-704.
 
0 评论数 0
暂无评论
友情链接