摘要
针对复杂场景下用单一特征描述目标导致的目标漂移问题,基于均值漂移(Mean Shift)跟踪框架,构建了一种有效的自适应融合特征(Adaptive Fusion Feature,AFF)描述子,并提出一种自适应融合多特征的跟踪方法。该方法融合了颜色特征和尺度不变特征转换(Scale-Invariant Feature Transform,SIFT),并通过相邻帧间各特征的相似性来自适应动态调整特征的权值。实验结果表明,在复杂场景下多特征自适应融合方法(AFF)比单一特征跟踪方法和经典跟踪方法减少了目标漂移、目标跟踪更加精确鲁棒。
Aimed at using single feature to describe the target often leads to target drift in complex scenes, aneffective Adaptive Fusion Feature （AFF） is constructed based on Mean Shift tracking framework, furthermore, atracking method which used multiple fusion features to describe target adaptively is put forward. This trackingmethod combined color feature and SIFT feature, the similarity between adjacent frames of each feature is used todynamically adjust the feature weights. The experimental results show that the proposed AFF tracking method ismore accurate and robust than single feature tracking and state-of-the-art tracking methods in complex scenes.
作者
龚春红
Gong Chunhong(Department of Information Management, Hunan University of Finance and Economics, Changsha 410205, China)
出处
《湖南文理学院学报（自然科学版）》
CAS
2016年第4期21-26,共6页
Journal of Hunan University of Arts and Science(Science and Technology)
基金
湖南省重点学科建设项目
湖南省教育厅科学研究重点项目(13A010)
