基于组合树形结构的多特征协同识别行人方法  

Multi-feature Collaboration Pedestrian Recognition Method Based on Combination of Tree-structure

作　　者：孟晓莉[1] 陈大伟[2] 

机构地区：[1]江苏海事职业技术学院信息工程系,江苏南京211170 [2]东南大学交通学院,江苏南京211170

出　　处：《电视技术》2014年第23期152-157,179,共7页Video Engineering

基　　金：美国能源基金会资助项目(G-1408-16758);江苏省教师素质提高研究计划项目(2013SJB880020)

摘　　要：采取Edgelet特征和聚集型B-Haar特征相结合,协同进行特征提取,设计开发出具有树形组合结构的行人识别模型。该模型的上层结构为:通过改进具有Haar特征(此处称为聚集型B-Haar特征),在完全二叉树架构的基础上,同局部二元模式相结合,对候选人目标进行提取,最终提高检测识别率;该模型的下层结构为:在贝叶斯原理和Edgelet特征相结合的基础上,构建树状决策结构,对多部位进行检测,找寻出行人。实验结果与传统的串并联结构和树状结构进行比较,设计开发的多特征协同树状组合决策结构行人识别方法更具优势,能够更好地保证实时性,降低虚警率,提高检测率。A feature extraction of an aggregated B - Haar and Edgelet feature coordination is proposed and a double - layer pedestrian recognition model is designed. The upper layer of the model ： A Haar feature model is improved with local binary pattern in the complete binary tree structure （ an aggregated B - Haar feature）, combining with local binary pattern and extracting target candidates to get a higher detection rate. The lower layer of the model is ： build a decision - making structure using Edgelet feature and Bayesian principle, analyze multiple sites of a candidate target to determine whether the target is a pedestrian, and achieve low false alarm rate and high real - time performance at last. Experiments results reveal that the multi - feature and combination of tree - structure pedestrian recognition algorithm designed in this paper has obvious overall advantages in real - time performance, detection rate and false alarm rate compared with the traditional tree structure or serial - parallel structure,

关 键 词：行人识别 树状决策结构 贝叶斯原理 整体优势 

分 类 号：TN959[电子电信—信号与信息处理]