CAJ | 학술논문

鉴于目前散货码头运用智能视频监控系统时，由于不同方向人形的方向梯度直方图（Histo-gram of Oriented Gradient，HOG）特征存在较大的变化，使得用传统方法训练获得的少量特异性特征不足以支撑人形的有效分类，因此提出一种基于AdaBoost的针对不同姿势HOG特征的二级分类方法。首先将样本快速分为正（背）面人形和侧面人形，组成第一级分类；然后通过分别为两类样本训练子分类器组成第二级分类；第二级分类对人形进行识别，并对结果进行融合。以天津港干散货码头无人作业区为背景，完成一组人形识别实验。实验结果表明，相较于传统方法，该方法对正（背）面人形具有更高的识别率。二级分类方法整体上提高了人形识别的识别率。
감우목전산화마두운용지능시빈감공계통시，유우불동방향인형적방향제도직방도（Histo-gram of Oriented Gradient，HOG）특정존재교대적변화，사득용전통방법훈련획득적소량특이성특정불족이지탱인형적유효분류，인차제출일충기우AdaBoost적침대불동자세HOG특정적이급분류방법。수선장양본쾌속분위정（배）면인형화측면인형，조성제일급분류；연후통과분별위량류양본훈련자분류기조성제이급분류；제이급분류대인형진행식별，병대결과진행융합。이천진항간산화마두무인작업구위배경，완성일조인형식별실험。실험결과표명，상교우전통방법，해방법대정（배）면인형구유경고적식별솔。이급분류방법정체상제고료인형식별적식별솔。
At present,the intelligent video monitoring system is used in bulk terminals. However,Histo-gram of Oriented Gradient (HOG)features of human body in different directions show great difference, so that a small number of specific features obtained by the traditional method is insufficient to support ef-fective classification of human body. Therefore,a kind of two-stage classification method based on Ada-Boost is proposed for different HOG features of different human postures. Firstly,the samples are rapidly divided into the Front&Back (F&B)human body and the side (not F&B)human body to form the first-stage classification. Then,the sub-classifiers are respectively trained for the two types of samples to form the second-stage classification,where the human body and the non-human body are recognized respec-tively and the results are merged. Taking the unmanned area in the bulk terminal of Tianjin Port as the background,a group of human body recognition experiments is carried out. The experimental results show that,compared with the traditional method,this method is of higher recognition rate for F&B human body. The two-stage classification method improves the recognition rate of human body overall.