摘要
针对压缩跟踪算法中目标尺度不能自适应更新和分类器性能弱的问题,提出一种尺度自适应的加权压缩跟踪算法.首先,提取正负样本的压缩特征,根据两者的巴氏系数对分类器进行加权,分类能力越强的特征在分类器中的权值越大.然后,将候选样本输入贝叶斯分类器,通过分类得到目标位置.最后,根据目标位置求得不同尺度下的样本特征,组成尺度金字塔,将尺度金字塔作为相关滤波器响应的特征输入,将响应值最大的尺度作为当前估计的尺度值.选取6组视频序列测试改进算法,实验结果表明,与传统压缩跟踪算法等4种算法相比,此算法能够解决尺度变化和目标旋转问题,且满足实时性要求.
As the target scale cannot adaptively update and classifier performance is not strong in the compressive tracking algorithm,a weighted compression tracking algorithm is proposed with adaptive scales.Firstly,the compressive features of positive and negative samples are extracted and the classifier is weighted according to the Bhattacharyya coefficients.The greater the classification ability is,the greater the weight of the feature in the classifier is.Secondly,the candidate samples are input into the Bayesian classifier and the target position is obtained by classification.Finally,the sample features at different scales are obtained according to the target position.A scale pyramid which consists of these features is input as a feature of the correlation filter score.The new scale is then found by maximizing the score.The proposed algorithm is tested by six video sequences.Experimental results show that the proposed algorithm has better tracking effect for the scale change problem in comparison with traditional compression tracking algorithm and other three algorithms,which could solve target rotation problem and meet real-time requirement.
作者
李晓行
陈金广
马丽丽
王明明
王伟
LI Xiaoxing;CHEN Jinguang;MA Lili;WANG Mingming;WANG Wei(School of Computer Science,Xi′an Polytechnic University,Xi′an 710048,China)
出处
《西安工程大学学报》
CAS
2018年第1期105-113,共9页
Journal of Xi’an Polytechnic University
基金
国家自然科学基金资助项目(61601358)
中国纺织工业联合会科技指导性计划项目(2017058)
陕西省教育厅科研计划项目(17JK0329)
关键词
压缩跟踪
尺度更新
相关滤波器
加权分类器
目标跟踪
compressive tracking
scale update
correlation filters
weighted classifier
object tracking
