摘要
针对视频目标跟踪问题,介绍了一种基于局部图像描述的目标跟踪方法.以高斯差值函数图像金字塔中的空间极值点作为图像关键点,在关键点邻接域中,以基于局部二进制模式(LBP)的纹理统计和色彩分布作为特征量,应用特征量在视频序列中的匹配实现目标跟踪.当视频序列中存在目标全局运动、有限度的尺度变化、旋转等复杂因素时,采用纹理和色彩分布得到的局部图像描述具备一定的稳定性,运用权值调整算法寻找稳定特征量集合,能够进一步自适应跟踪目标的外观变化.测试了不同条件下的视频序列,该方法具备良好的跟踪效果和一定的稳定性.
A novel local image descriptor based method was introduced for object tracking from video sequences. -The key points of an image were generated from the extreme points in difference of Gaussian （DOG） image pyramid, and the local image descriptor was based on local binary pattern （LBP） histogram and color histogram. The tracking was accomplished by the features matching in the series of frames. The LBP and color based descriptor were stable even when there are global moving,limited scale changes, or rotations in video sequences. An algorithm for adjusting weight of descriptor was applied to find a stable descriptor set and improve the reliability when the object appearance changes. Tests of several video sequences show that the method has good stability and tracking performance.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2008年第7期1179-1183,共5页
Journal of Zhejiang University：Engineering Science
基金
国家"863"高技术研究发展计划资助项目(2003AAIZ2130)
浙江省科技计划重大科技技术攻关资助项目(2005C11001-02)
