摘要
针对传统的核密度估计在运动目标检测中需要进行复杂的运算,并且背景模型无法自适应更新等问题,提出了基于关键帧采样的核密度估计背景建模算法。结合间隔视频序列的平均背景和相似性原理,提取具有关键背景信息的样本建立背景模型,大大缩短了背景建立的时间。同时引入融合背景更新策略,实现了背景的自适应更新,克服了光照变化对背景重建的影响。在此基础上,检测系统结合梯度和聚类消除了运动阴影。实验结果表明,该方法具有检测精度高,运行速度快等特点,更好地满足了实时性要求。
A kernel density estimation background model based on key sampling is presented in the moving object detection.A crucial background information samples,which is obtained by means background of interval sequence and pixel similarity theory,is employed to build background model for making the background of the establishment of the time significantly shortened.Meanwhile a fusion updating mechanism is applied for auto-updating background model,which overcame the light change on the background of the impact of the reconstruction.Moving shadow is then removed completely by combining of gradient and clustering method.Simulation experiment indicates that the proposed method has characteristics of strong precision and fast speed,and further increased real-time performance.
出处
《计算机技术与发展》
2010年第8期45-48,共4页
Computer Technology and Development
基金
国家"十一五"计划课题(FIB070335-B8-04)
关键词
核密度估计
关键帧
梯度
聚类
kernel density estimation
key frame
gradient
clustering
