一种高可靠性的头部校验夜间行人快速识别方案  

A rapid and high reliable dentify program for nighttime pedestrians

作　　者：王丰斌[1] 赵喜玲[1] 何勇[1] 杨世凤[2] 

机构地区：[1]信阳农林学院计算机科学系,信阳464000 [2]天津科技大学自动化学院,天津300222

出　　处：《四川大学学报（自然科学版）》2015年第4期785-792,共8页Journal of Sichuan University(Natural Science Edition)

基　　金：国家"十二五"科技支撑计划课题资助项目(2012BAH32F06)

摘　　要：针对夜间行人检测成像尺度不一等因素所引起的类内方差较大、实时性不足等问题,本文在统计学习的应用原理下,设计了基于熵加权和FCSVM优化的头部校验夜间行人快速识别方案.该方案应用熵加权原理改进梯度直方图特征,引入了三分支结构的支持向量机对目标进一步识别,同时利用快速分类支持向量机(FCSVM)降低运算所需的开销,确保实时性,最后通过头部校验方法对误检目标进一步分析评估,进一步提高图像匹配的准确度.实验结果表明,该方案在夜间环境下能有效区分远红外行人目标,在充分确保行人实时性的基础上,在市区、郊区等不同应用环境中,均具有良好的实用性.For the larger intra-class variance and inadequate real-time problems caused by factors of ima- ging scales difference in nighttime pedestrian detection. The paper designs a rapid dentify program for nighttime pedestrians based on entropy weight and header checksum of FCSVM optimization under the the application of the statistical learning principles. The program utilizes entropy weighted to improve the feature of gradient histogram, introduces three branch structure SVM to identify the target further, and uses rapid classification FCSVM to reduce the overhead required of computational and to ensure real- time, finally through the header checksum method to analysis and assess error detection goals, to fur- ther improve the accuracy of image matching. Experimental results show that the scheme can distinguish far infrared pedestrian goals effectively at night environment, and have good usability in urban, subur- ban and other different application environments on the basis of ensuring pedestrian real-time fully.

关 键 词：夜间行人检测 统计学习 熵加权 快速分类支持向量机 头部校验 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]