基于改进高斯混合建模和短时稳定度的运动目标检测算法.pdfVIP

第 34 卷第 10 期 电 子 与 信 息 学 报 Vol.34No.10 2012 年 10 月 Journal of Electronics Information Technology Oct. 2012 基于改进高斯混合建模和短时稳定度的运动目标检测算法 张 超① 吴小培*① 周建英① 戚培庆① 王营冠② 吕 钊① ①(安徽大学计算智能与信号处理教育部重点实验室 合肥 230039) ②(中国科学院上海微系统与信息技术研究所 上海 200050) 摘 要：传统高斯混合建模方法中前景模型转换为背景模型的风险随着模型权值在一定更新率下的迭代累加而增 大，使得传统方法难以有效检测低速运动目标。该文对高斯混合建模中背景匹配失败时生成的前景模型加以利用并 引入短时稳定度指标进行综合判断，在深入挖掘前景模型中包含的运动目标信息和像素点级稳定性的基础上实时区 分各像素点的状态。多场景下的实验结果表明，该方法对低速目标达到了更高的检出率。 关键词：运动目标检测； 高斯混合建模； 前景模型； 短时稳定度 中图分类号：TP391 文献标识码： A 文章编号：1009-5896(2012)10-2402-07 DOI: 10.3724/SP.J.1146.2012.00158 A Moving Object Detection Algorithm Based on Improved GMM and Short-term Stability Measure ① ① ① ① ② ① Zhang Chao Wu Xiao-pei Zhou Jian-ying Qi Pei-qing Wang Ying-guan Lü Zhao ① (Key Lab of Intelligent Computing and Signal Processing Anhui University, Hefei 230039,China) ② (Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 200050, China) Abstract: In traditional Gaussian Mixture Modeling (GMM) algorithm, the risk that foreground model changes into background model rises with the cumulating of model weight under certain learning rate. That makes the algorithm unable to deal with slow moving object. This paper proposes an algorithm which takes advantage of the foreground models and employs an index of short-term stability measure to make a compound judgment. Each pixel status is decided real-timely considering the information of moving objects contained in foreground models and the pixel-level stability status. The results from different experiments veri