基于自适应混合高斯模型全方位视觉目标检测  

Adaptive mixture Gaussian model based target detection using omnidirectional vision

作　　者：刘士荣[1] 王凯[1] 邱雪娜[2] 

机构地区：[1]杭州电子科技大学自动化研究所,浙江杭州310018 [2]华东理工大学自动化研究所,上海200237

出　　处：《浙江大学学报（工学版）》2010年第7期1387-1393,共7页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(60675043);浙江省科技计划资助项目(2007C21051);杭州电子科技大学科研启动基金资助项目(KYS09150543)

摘　　要：针对全方位视觉传感器视野范围大的特点,提出一种基于自适应混合高斯模型的全方位视觉目标检测系统.该系统通过Hough变换检测全方位图像的中心,基于图像中心对全方位图像进行展开.对展开后的图像利用混合高斯模型进行背景建模,并自适应地更新背景模型,通过前景分割可以有效地分割出运动目标.在图像展开及混合高斯建模时,通过调整系统的采样频率可以较好地改善目标检测的实时性.实验结果表明,该系统可以在复杂环境中有效地检测运动目标,具有较强的准确性和鲁棒性.Considering the advantage of omnidirectional vision sensor with large field-of-view,an omnidirectional vision target detection system was presented based on the adaptive mixture Gaussian model.A Hough transform algorithm was adopted to detect the center of the scene,and the omnidirectional image was transformed around the center.An adaptively updated background was modeled by the mixture Gaussian model for the transformed images.The moving object can be effectively segmented by the foreground segmentation method.The real-time performance of target detection can be improved by adjusting the sample frequency during image transforming and mixture Gaussian modeling.Experimental results demonstrate that the system can effectively detect the moving object under complex environments,and has more accuracy and robustness.

关 键 词：全方位视觉传感器 HOUGH变换 图像展开 混合高斯模型 目标检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]