摘要
背景建模是背景减法技术提取前景的基础,背景建模方法的好坏直接影响目标的提取和跟踪.对于视频中同一像素块向量在时间序列上的变化,用三维矩阵水平、垂直方向表示块的位置,纵向表示块的聚集范围.根据像素块向量在一定范围内出现的次数调整其在纵向的位置,出现次数越多,其位置向前越靠近.最终使出现次数较多的向量集中在矩阵纵向1～ d/2的范围内,并根据矩阵这部分值对前景和背景进行分割.该方法中采用动态阈值,能实时动态更新背景,无须对视频进行专门学习.实验证明该方法能有效记录像素块向量的变化,使频繁出现的向量有效聚集,建立的背景模型稳定可靠,提取的前景具有较好的鲁棒性.
Background modeling is the foundation of extracting foreground by background subtract technology,which influence on the extraction of target and tracking directly.For the same pixel block changing in the time series,three dimensional matrix can express the position in horizontal and vertical direction,and gathered range in longitudinal of pixel block.According frequency of occurrence of the pixel block vector in a certain range to adjustment position in the longitudinal,more appear,more forward.Finally,the pixel block vector whose occurrence number is greater concentrated in the longitudinal 1 ～d/2 range,and foreground and background are segmented by the score matrix.Dynamic threshold are used in the method which can dynamic update background real-time without learning.The experiment show that the method can effectively record pixel block vector changes,gather appeared frequently vector effectively,background model is stable and reliable through the method,extracted foreground with the method has good robustness.
出处
《电视技术》
北大核心
2013年第19期53-56,86,共5页
Video Engineering
基金
陕西省科技攻关计划项目(2011k 06-32)
中央高校基本科研业务费专项资金项目(CHD2011TD012)
长安大学科技创新重点项目
陕西省道路交通智能检测与装备工程技术研究中心开放基金项目
西藏民族学院青年学人培育计划资助项目(12myQP07)
关键词
像素块向量
背景建模
实时性
动态阈值
pixel block vector
background modeling
real-time
dynamic threshold
