基于自适应压缩特征选择的实时目标跟踪算法.pdfVIP

第28 卷 第4 期 模式识别与人工智能 Vol.28 No.4 2015年4月 PRAI Apr. 2015 基于自适应压缩特征选择的实时目标跟踪算法∗ 齐美彬,2 陆 磊 杨 勋 杨艳芳3 蒋建国,2 1(合肥工业大学计算机与信息学院 合肥230009) 2(合肥工业大学安全关键工业测控技术教育部工程研究中心 合肥230009) 3(合肥工业大学电子科学与应用物理学院 合肥230009) 摘 要 针对压缩感知算法中的低维特征对目标重构效果较差的问题,提出基于自适应压缩特征选择的目标跟踪 算法.该算法首先提取满足目标重构要求的高维压缩特征,再通过所提出的特征选择方法选择区分度高的低维特 征作为目标的外观模型,从而降低计算复杂度.为自适应选择特征,采用一种差分方法控制特征维数,满足实时性 要求.实验表明,与其他算法相比,文中算法具有更强的鲁棒性和实时性. 关键词 目标跟踪, 自适应特征选择,方差比,均值差,差分方法 中图法分类号 TP391 DOI 10.16451/ki.iss201504009 Real-TimeObjectTrackingAlgorithm BasedonAdaptiveCompressiveFeatureSelection 12 1 1 3 12 QIMei-Bin ,LULei ,YANGXun ,YANGYan-Fang ,JIANGJian-Guo 1(SchoolofComputerandInformation,HefeiUniversityofTechnology,Hefei230009) 2(EngineeringResearchCenterofSafetyCriticalIndustrialMeasurementandControlTechnology, MinistryofEducation,HefeiUniversityofTechnology,Hefei230009) 3(SchoolofElectronicScienceandAppliedPhysics,HefeiUniversityofTechnology,Hefei230009) ABSTRACT Low dimensionalfeaturesadoptedbycompressivetrackingalgorithmcannotreconstructtheobject effectively. Tosolvethisproblem,areal-timeobjecttrackingalgorithmbasedonadaptivecompressive featureselectionisproposedinthispaper. Thehighdimensionalfeaturesmeetingtherequirementof objectreconstructionareextracted. Thenthelowerdimensionalfeatureswithahigherdiscriminationare selectedasappearancemodeloftheobjecttoreducethecomputationalcomplexity. Toselectfeatures adaptivelyadifferencemethodisadoptedtocontrolthefeaturedimensionality. Theexpe