摘要
首先在网络Conv layers层对图片进行特征提取,在RPN层改变Anchors比例(1∶2),再修正Anchors获得精确的Proposals。在Roi Pooling层收集输入图片特征和Proposals,综合以上信息后提取Proposal feature maps。在Classification层计算Proposal的类别,获得检测框最终精确位置,并统计检测框数量,最终确定图片中的人数。
First the features of an image are extracted in network Conv layer,and Anchors proportion is changed to(1∶2)in RPN layer,Then the Anchors are modified to obtain accurate Proposals.In ROI pooling layer the features and Proposals are collected to extract the Proposal feature maps by synthesizing all information.In Classification layer Proposals are classified to get the final exact position of the detected box.Finally,the numbers of detection boxes are calculated to determine the number of people in the picture.
作者
肖巍
卢劲伉
李博深
吴启槊
白英东
潘超
XIAO Wei;LU Jinkang;LI Boshen;WU Qishuo;BAI Yingdong;PAN Chao(School of Computer Science & Engineering, Changchun University of Technology, Changchun 130012, China)
出处
《长春工业大学学报》
CAS
2020年第4期369-374,共6页
Journal of Changchun University of Technology
基金
国家自然科学基金面上项目(61472049)
吉林省科技发展计划技术攻关项目(20190302071GX)。
