自适应融合特征的人群计数网络  

Adaptive Feature Fusion Network for Crowd Counting

作　　者：左健豪 姜文刚[1] ZUO Jianhao;JIANG Wengang(School of Electronic Information,Jiangsu University of Science and Technology,Zhenjiang,Jiangsu 212003,China)

机构地区：[1]江苏科技大学电子信息学院,江苏镇江212003

出　　处：《计算机工程与应用》2021年第21期203-208,共6页Computer Engineering and Applications

基　　金：国家自然科学基金(61671222)。

摘　　要：针对人群计数方法中存在的尺度变化和多层级特征融合不佳的问题,基于U-Net的编码器-解码器网络结构,提出一种自适应特征融合网络,来进行精准的人群计数。提出自适应特征融合模块,根据解码器分支的需要,高效地聚合编码器分支提取的高层语义信息和底层的边缘信息;提出自适应上下文信息提取器,从不同感受野下提取多尺度的上下文信息并自适应加权融合,提高网络对于人头尺度变化的鲁棒性。在ShanghaiTech、UCF-CC-50和UCG-QNRF上的实验表明,与目前主流的人群计数算法相比,该算法具有更强的准确性和鲁棒性。In an attempt to solve the problems of scale change and multi-level feature fusion in population counting method,inspired by U-Net encoder decoder structure network,an adaptive feature fusion network is proposed to carry out accurate population counting.The Adaptive Feature Fusion Module(AFFM)is proposed to efficiently aggregate the high-level semantic information and low-level spatial detail extracted by the encoder branch according to the needs of decoder branch.The Adaptive Context Extractor(ACE)is proposed to extract multi-scale context information on multiple effective field-of-views,then these features are adaptively fused to improve the robustness of the network to scale changes.By conducting exhaustive experiments on Shanghai Tech,UCF-CC-50 and UCF-QNRF,the results show that the network has high accuracy and robustness.

关 键 词：人群计数 卷积神经网络 密度估计 多层级特征 尺度变化 特征融合 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]