一种基于二值粒子群优化和支持向量机的目标检测算法  

A Binary Particle Swarm Optimization and Support Vector Machine-based Algorithm for Object Detection

作　　者：潘泓[1] 李晓兵[1] 金立左[1] 夏良正[1] 

机构地区：[1]东南大学自动化学院,南京210096

出　　处：《电子与信息学报》2011年第1期117-121,共5页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(60805002;90820009);航空科学基金(20080169003);东南大学优秀青年教师教学科研资助计划;国家留学基金资助课题

摘　　要：针对复杂场景下目标检测和目标检测中特征选择问题,该文将二值粒子群优化算法(BPSO)用于特征选择,结合支持向量机(SVM)技术提出了一种新颖的基于BPSO-SVM特征选择的自动目标检测算法。该算法将目标检测转化为目标识别问题,采用wrapper特征选择模型,以SVM为分类器,通过样本训练分类器,根据分类结果,利用BPSO算法在特征空间中进行全局搜索,选择最优特征集进行分类。基于BPSO-SVM的特征选择方法降低了特征维数,显著提高了分类器性能。实验结果表明,该文算法不仅有效提高了复杂场景下目标姿态、尺度、光照变化和局部被遮挡时的检测准确率,还大大缩短了检测时间。This paper proposes a novel object detection method,namely the BPSO-SVM-based detection algorithm that combines Binary Particle Swarm Optimization（BPSO） and Support Vector Machine（SVM） techniques to cope with feature selection issue for object detection under complex scenarios.In the proposed algorithm,object detection is regarded as a two-class categorization problem and feature subset is selected using a wrapper model,in which the BPSO searches the whole feature space and a SVM classifier serves as an evaluator for the goodness of the feature subset selected by the BPSO.Using the proposed BPSO-SVM-based feature selection scheme,feature dimensionality is reduced and classification performance of the SVM classifier is greatly enhanced.Experimental results show the increase on detection accuracy of the proposed algorithm for object detection in complex backgrounds with pose,scale,illumination variations and partial occlusions as well as the significant improvement on detection speed.

关 键 词：目标检测 二值粒子群优化 支持向量机 特征选择 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]