图像融合的运动目标检测算法研究  

Moving object detection algorithm based on image sequences fusion

作　　者：吴君钦[1] 刘昊[1] 罗勇[1] 

机构地区：[1]江西理工大学信息工程学院,江西赣州341000

出　　处：《计算机工程与设计》2012年第12期4614-4618,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(11062002);江西省自然科学基金项目(2008GZS0028)

摘　　要：为了改进常用的运动目标检测算法易受噪声和光线变化的影响、易出空洞、阴影和假边缘等现象,提出一种基于连续五帧帧间差分与Surendra背景边缘差分相融合的运动目标检测算法。该方法先采用Surendra自适应背景提取算法建立运动区域模型,通过优化的Canny算子进行背景边缘检测差分运算,再与五帧差分法相融合,通过双向模板填充和后期处理获得完整、准确的运动目标区域并完成背景的实时更新。实验结果表明,该算法快速、准确,能满足实时性检测的要求。To improve the common moving object detection algorithm which is susceptible to noise and light changes, easy to the phenomenon of inanition, shadow and false edge, a novel moving object detection algorithm based on five-frame-differencing and background edge-characteristic difference is presented. This algorithm adopts Surendra algorithm to obtain a background model, and then the moving object edge is gained using background edge-characteristic difference with improved Canny algorithm, which is combined with five-frame differencing to detect the moving object. At last, the complete and accurate object is extracted with morphologic filtering and connectivity detecting and update background real-time. The result shows the present algorithm runs quickly and veraciously, and is fit for real time detection.

关 键 词：五帧差分 边缘检测 背景边缘检测差 背景更新 运动目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]