基于SVM的公交人数统计方法研究  

People counting based on SVM in bus scene

作　　者：张文涛 宋焕生[1] 李莹[1] 郑宝峰 严腾 张向清 ZHANG Wentao;SONG Huansheng;LI Ying;ZHENG Baofeng;YAN Teng;ZHANG Xiangqing(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《中国科技论文》2018年第2期143-148,共6页China Sciencepaper

基　　金：国家自然科学基金资助项目(61572083);陕西省自然科学基础研究计划资助项目(2015JZ018;2015JQ6230);中央高校基本科研业务费专项资金资助项目(310824152009;310824163411)

摘　　要：为解决实际公交场景中人数统计精确度不高的问题,采用基于支持向量机(support vector machine,SVM)分类的方法对疑似目标的三维轨迹进行分析,通过提取真实目标与伪目标轨迹的特征信息,进一步分类真实目标与伪目标。首先通过相机标定将深度相机获取的深度图像转换为三维空间中的俯视图;然后采用局部高度最大值方法提取疑似人头目标区域,并利用卡尔曼滤波跟踪得到三维轨迹;最后利用SVM训练正负样本得到强分类器,对目标轨迹进行分类,实现人数自动计数。实验表明,所提方法能够有效地提高目标轨迹分类和人数统计的精度。To solve the precision problem of the counting system in the real bus scene,this paper proposed a method,which is based on support vector machine(SVM)classification to analyze the 3D trajectory of the suspected targets and use the feature of trajectory to classify the real target and pseudo target.Firstly,camera calibration is performed,which converts the depth image obtained by depth camera to a top-view image in 3D space.Secondly,the local height maximum method is used to extract the area of suspected head,and Kalman filter is used to track the 3D trajectory.Finally,SVM is used to train the positive and negative samples to obtain the strong classifier.It is able to classify the target trajectories and achieve the automatic counting.Experiments show that our method can effectively improve the accuracy of target trajectory classification and pedestrian statistics.

关 键 词：SVM分类 摄像机标定 轨迹特征 卡尔曼滤波 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]