摘要
针对传统的背景建模会产生空洞和阴影误检测的问题,提出了基于矢量特征的动态背景建模方法。该算法分为初始背景模型学习和更新背景模型学习两个部分。初始背景模型把图像RGB 3个特征对应到球坐标中的方向特征,并取前若干帧图片通过矢量均值聚类算法算出K个聚类,认为这K个类为这一像素点的背景模型,当一张新的图片对应像素的矢量特征落在这K类中的任何一个时,就认为其为背景;更新算法是初始模型的后续,它除了对新来的图片进行背景分析之外,也将其用来更新背景模型。该算法能够有效避免空洞现象和阴影误检测,并且当场景改变时能及时有效地更新背景。
For the problem that the traditional background modeling results in empty hole and wrong shadow detection,we proposed a dynamic background modeling method based on vector characteristics.Learning algorithm is divided into two parts namely initial background model and updating background model.The initial background model maps the RGB feature to the direction feature in spherical coordinates.K clustering centers are calculated by the lastest images using the method of mean vector clustering algorithm,and the K clusters are considered to be the background model of the pixel.When a new image's corresponding pixel falls into any one of the K clusters,the new pixel is considered to be background.The updating algorithm is the successor of the initial background model algorithm.Besides doing back-ground analyse for the new image,it also uses the new image to update the background model.The algorithm can effectively reduce the empty hole and wrong shadow detection,and can update the background timely when the scene changes.
出处
《计算机科学》
CSCD
北大核心
2016年第3期291-295,共5页
Computer Science
关键词
矢量方向特征
背景建模
矢量均值聚类
更新背景
Vector direction feature
Background modeling
Mean vector clustering
Update background