文献类型：期刊文献

中文题名：级联式多尺度行人检测算法研究

英文题名：Study on multi-scale pedestrian detection algorithm based on cascade convolution network

作者：张姗[1];刘艳霞[2];方建军[2]

第一作者：张姗

机构：[1]北京联合大学北京市信息服务工程重点实验室,北京100101;[2]北京联合大学城市轨道交通与物流学院,北京100101

第一机构：北京联合大学北京市信息服务工程重点实验室

年份：2020

卷号：39

期号：1

起止页码：42-45

中文期刊名：传感器与微系统

外文期刊名：Transducer and Microsystem Technologies

收录：CSTPCD;;CSCD:【CSCD_E2019_2020】；

基金：国家自然科学基金资助项目(61602041);北京联合大学人才强校优选计划项目(BPHR2017CZ07);教育部天诚汇智科研创新基金资助项目(2018A03017);北京市教育委员会科研计划基金资助项目(KM201911417007)

语种：中文

中文关键词：多尺度行人检测;级联卷积神经网络;正样本采集;加权损失函数

外文关键词：multi-scale pedestrian detection;cascade convolutional neural network;positive sample collection;weighted-loss function

摘要：针对多尺度行人检测精度不够高的问题,提出了一种级联式多尺度行人检测算法,使用矩形卷积核提取行人特征,根据行人轮廓特征设计候选区域宽高比例;并提出自适应损失函数,使网络聚焦于困难样本,有效缓解了长尾效应在训练网络时带来的不利因素,提高了网络泛化能力。实验结果表明:所提算法对于Caltech数据集中的大尺度行人,漏检率比Adapt Faster Rcnn算法降低了1.36%;对于中小尺度行人,漏检率比Adapt Faster Rcnn算法下降8.82%。
Aiming at the problem that precision of multi-scale pedestrian detection algorithm is not high,a multi-scale pedestrian detection algorithm based on cascade convolutional neural network is proposed.The rectangular convolution kernel is used to extract pedestrian features,and the ratio of width to height of proposals are designed according to pedestrian contour features and the self-adaptive loss function is proposed,which makes the network focus on hard samples and effectively alleviate adverse factors brought by the long tail effect in training network and improves the network generalization ability.Compared with Adapt Faster RCNN,the miss rate of the proposed algorithm declines by 1.36%in large-scale pedestrians of Caltech data set and that decreases by 8.82%in small and medium-sized.