基于腐蚀-膨胀比的视频场景运动状态鲁棒检测  

A Robust Approach to Detect Motion Status in Video Scenes Based on Ratio of Erosion and Dilation

作　　者：陈立平[1,2] 曹丽萍[3] 黄增喜[1] 刘春玲[4] 

机构地区：[1]四川大学计算机学院,四川成都610065 [2]塔里木大学信息工程学院,新疆阿拉尔843300 [3]四川大学图书馆,四川成都610065 [4]四川大学华西医院,四川成都610041

出　　处：《四川大学学报（工程科学版）》2011年第4期95-100,共6页Journal of Sichuan University (Engineering Science Edition)

基　　金：教育部新世纪人才支持计划资助项目(NCET-08-0370);国家自然科学基金资助项目(60705005);高等学校博士学科点专项科研基金资助项目(20090181110052);博士点新教师基金资助项目(20070610031);四川省国际合作与交流研究计划(2010HH0031)

摘　　要：针对智能视觉分析中视频场景状态检测问题,提出了一种鲁棒的方法。首先,用高斯金字塔算法预处理输入帧;对比3种帧间差计算方法,其中,先灰度化再差分的帧间差计算方案性能最佳。通过分析帧间差灰度图中高亮点在空间分布和形态学上的差异,提出了基于腐蚀-膨胀比的场景状态检测算法REDP。将算法应用于不同场景、亮度、天气条件下的视频序列,实验结果说明了算法不仅可指示获取场景中运动对象位置和轮廓信息的时机,而且,验证了算法对于场景状态检测的有效性。通过增大参与帧间差运算的帧间间距,可进一步提高算法对场景状态检测的鲁棒性。The issue of detecting the status of video scenes was studied.First,Gaussian pyramid algorithm was employed to preprocess input frames.By comparing three consecutive inter-frame difference methods,the method that calculated the inter-frame difference on the gray frame images was adopted.According to the spatial distribution of the difference information as well as the morphological features of high intensity points,a new algorithm named Ratio of Erosion and Dilation per Point（REDP） was proposed to detect status change under different videos,which were captured in different scenes,luminance and weather conditions.The experimental results confirmed the effectiveness of indicating whether there are motions in the video.When the proposed approach was applied to the inter-frames,the detection capability of the algorithm can become more robust under an appropriate frame interval.

关 键 词：智能视觉 运动分析 场景状态检测 腐蚀-膨胀比 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]