改进压缩特征的实时压缩跟踪算法.pdfVIP

Computer Engineering and Applications 计算机工程与应用 1 改进压缩特征的实时压缩跟踪算法 崔灿，王民钢，李立，张希铭 CUI Can, WANG Mingang, LI Li, ZHANG Ximing 西北工业大学 航天学院,西安710072 School of Astronautics, Northwestern Polytechnical University, Xi’an 710072, China CUI Can, WANG Mingang, LI Li, ZHANG Ximing. Advanced real-time compressive tracking based on analyzing low-dimensional features. Computer Engineering and Applications Abstract ：An advanced algorithm (RCT) is proposed based on Real-time Compressive Tracking (CT). By analyzing low-dimensional features, an advanced random measurement matrix gives a balanced reflecting to gray level features and texture features is designed to RCT. RCT started with obtaining low-dimensional feature vectors byusing the advanced random mea surement matrix. Then it classifies the image sequence into two categories (foreground and background) byusing Naive Bayes Classifier. Making a comparison among CT, BCT (a known improved algorithm based on CT) and RCT, according to tracking simulation results to different image sequences, it turns out that RCT shows very robust tracking results without losing the real -time character. Key words ：Real-time Compressive Tracking; random measurement matrix; low -dimensional feature vectors 摘要：提出了一种改进的实时压缩跟踪算法（RCT ）。该算法基于实时压缩跟踪算法（CT，Real-time Compressive tracking），构造出一个改进的随机测量矩阵，使降维后得到的压缩特征包含的灰度特征信息 和纹理特征信息比例相等。RCT 算法首先将图像序列的特征用改进的随机测量矩阵转化为低维度特征，再用 朴素贝叶斯分类器对低维特征进行目标和背景的分类，从而实现对目标的跟踪。将原始算法（CT），一种改 进算法（BCT）和本文创新的改进算法（RCT）进行对比，实验表明：RCT 算法保持了原始算法的实时性，并 且在各实验图像序列中跟踪目标的鲁棒性最好。 关键词：实时压缩跟踪；随机测量矩阵；低维特征 doi:10.3778/j.issn.1002-8331.1603-0060 文献标志码: A 中图分类号: TP391 [4] 1 引言 间，得到低维观测向量，再通过求解凸