基于时空上下文信息的混合交通视频检测算法  

New Mixed Traffic Detection Algorithm Based on Spatial and Temporal Context

作　　者：王霄维 王殿海[2] 江晟[3] 赵莹莹[1] 

机构地区：[1]吉林大学交通学院,吉林长春130022 [2]浙江大学建筑工程学院,浙江杭州310058 [3]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《北京理工大学学报》2013年第12期1289-1292,1297,共5页Transactions of Beijing Institute of Technology

基　　金：国家"八六三"计划项目(2011AA110304)

摘　　要：前景物体的边缘信息对混合交通视频检测的参数提取和表达具有非常重要的作用.为克服孤立地利用当前图像无法准确提取边缘信息的问题,提出了基于时空上下文表达的混合交通边缘提取新算法.在获取当前边缘信息基础上,结合运动目标的特征属性与时空上下文相关信息,通过检索文本的关联性,进行前景边缘提取.实验表明,文中算法能够准确而充分利用上下文信息实现前景边缘的提取,前景边缘的有效提取率可达95%以上,为后续混合交通视频检测的分类识别和参数提取提供了有效的依据.Foreground edge information has a very important role in the parameter extraction and the expression of mixed traffic video detection. To overcome the problem that the accurate edge information cannot be extracted from current frame solely, a new edge extraction algorithm for mixed traffic based on spatial and temporal context is proposed. Based on the current frame edge information, combining the characteristics of the moving objects with the information of spatial and temporal context and retrieving the relevance of the text, the foreground edge is extracted. Experimental results show that the algorithm can get the extraction of the edge of the prospects accurately and adequately using context, and the effective extraction rate can reach above 95G, which provide an efficient basis for the subsequent classification and parameter extraction of the mixed traffic video detection.

关 键 词：混合交通 视频检测 时空上下文 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] U121[自动化与计算机技术—计算机科学与技术]