论文《ResnetCrowd: A Residual Deep Learning Architecture for Crowd Counting, Violent Behaviour Detecti》

原创 2018年04月16日 11:09:58

论文《ResnetCrowd: A Residual Deep Learning Architecture for Crowd Counting, Violent Behaviour Detection and Crowd Density Level Classification》
创新点:提出ResnetCrowd,用于人群计数、暴力行为检测和拥塞密度分类的深度残差架构;多任务方法;
贡献
    数据集方面是构建了一个100张图片的数据集适用于多任务人群计数,并且这个数据集是第一个电脑视觉数据集,完全标注了人群计数、暴力行为检测和密度级别分类;
    定量分析了多目标人群计数系统的好处;
解决问题:对群体分析的多目标学习方法还没有得到充分的调查,主要是因为缺乏一个适当标记的多任务数据集

Related Work
Crowd Counting:
    大多数方法是训练计数回归,将像素值直接映射到一个计数估算值;像素级的基于热图的计数方法已经被证明可以提高人群的计数性能,从而应对高度拥挤的场景。
Crowd Behavior Recognition:
    下面两篇论文都依赖于帧间的运动特性检测行为;

    Kai Kang, Chen Change Loy, and Xiaogang Wang. Deeply Learned Attributes for Crowded Scene Understanding.
    Tal Hassner, Yossi Itcher, and Orit Kliper-Gross. Violent flows: Real-time detection of violent crowd behavior.
Crowd Density Level Estimation:
    一个拥塞场景的拥塞层次通常表示为离散(0-n)或连续值(0.0-1.0);最透明的方案是将离散密度级标签直接从真实的人群计数值中推断出来,产生一个具有主观性和人为误差最小化的直方图样式分布。

Multi Task Crowd Dataset

    多任务人群数据集的核心目标:生成一组适合于训练和验证一个模型的图像,用于同时计数、暴力行为识别和人群密度等级分类。
    建造数据集的几个标准:
    产生所需的多任务人群数据集的最有效方法是将新的标签应用到现有的数据集上。
    数据集示例:
    crowd heatmap的产生方式同MCNN中的方法;全部的GT图像均被下采样到160*90,目的是匹配ResnetCrowd模型中的预测heatmap分辨率;


ResnetCrowd

基于论文:Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun.Deep residual learning for image recognition.

Behaviour Recognition:
sigmoid activation;Binary cross entropy;

Density Level Classification:

Regression Based Crowd Counting:

    Relu activation, Mean Squared Error;
Heatmap Based Crowd Counting:

total loss:


实验部分



Conclusion
    Future work will look to include unsupervised learning techniques to overcome the lack of labelled crowd data and further increase model generalisation.

人群分析--ResnetCrowd: A Residual Deep Learning Architecture

ResnetCrowd: A Residual Deep Learning Architecture for Crowd Counting, Violent Behaviour Detection a...
  • zhangjunhit
  • zhangjunhit
  • 2017-09-27 16:25:39
  • 607

ResnetCrowd A Residual Deep Learning Architecture for Crowd Counting,Violent Behaviour Detection

ResnetCrowd A Residual Deep Learning Architecture for Crowd Counting,Violent Behaviour Detection and...
  • qq_19598705
  • qq_19598705
  • 2018-01-02 19:10:35
  • 236

论文笔记 A Large Contextual Dataset for Classification,Detection and Counting of Cars with Deep Learning

ECCV 2016的文章,首先建立了一个从上到下照的车辆影像数据集(即鸟瞰视角),并提出ResCeption神经网络进行训练,进一步建立residual learning with Inception...
  • u012905422
  • u012905422
  • 2016-09-28 22:52:57
  • 752

人群密度估计-Crowd Density

一. 应用背景      在安防大背景下,对敏感区域人流量的管控是一个重要的课题,防止人群骚乱、踩踏现象的发生,对非预期的人员汇聚进行预警等等,最常用的方法是检测到每个目标,然后借助 Perspect...
  • linolzhang
  • linolzhang
  • 2017-12-18 23:47:49
  • 1507

Deep Residual Learning 论文解析

Deep Residual Network(DRN) 是微软亚研 何凯明组在15年10月份挂在arXiv 上的文章。所用的网络在ILSVRC&COCO 2015比赛中取得第一名的成绩(识别,分类,分割...
  • yaoxingfu72
  • yaoxingfu72
  • 2016-02-29 19:21:18
  • 8146

人群密度估计--CrowdNet: A Deep Convolutional Network for Dense Crowd Counting

CrowdNet: A Deep Convolutional Network for Dense Crowd Counting published in the proceedings of AC...
  • zhangjunhit
  • zhangjunhit
  • 2017-09-28 14:26:20
  • 904

Deep Residual Learning for Image Recognition(阅读)

好久没整理论文了,感觉有点耗时间,总结总结还是不错的,便于记忆和细节学习。 废话少说, 何恺明:地址 PPT讲解:地址 代码:地址 1、Residual networks赢了 ImageNet的det...
  • langb2014
  • langb2014
  • 2016-07-01 21:38:09
  • 3300

Deep Residual Learning for Image Recognition(译)

仅供参考,如有翻译不到位的地方敬请指出。论文地址:Deep Residual Learning for Image Recognition译文地址:http://blog.csdn.net/wspba...
  • wspba
  • wspba
  • 2017-02-25 11:41:29
  • 6436

论文笔记:A survey of recent advances in CNN-based single image crowd counting and density estimation

A survey of recent advances in CNN-based single image crowd counting and density estimation目标人群计数目标是...
  • super_chicken
  • super_chicken
  • 2017-12-22 14:33:28
  • 195

DL去燥之 DnCNNs:Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising

网络结构:DnCNNs: feed-forward denoising convolutional neural networks 学习策略Residual Learning:通过残差学习策略,在网...
  • qq_23968185
  • qq_23968185
  • 2017-06-09 00:02:10
  • 2832
收藏助手
不良信息举报
您举报文章:论文《ResnetCrowd: A Residual Deep Learning Architecture for Crowd Counting, Violent Behaviour Detecti》
举报原因:
原因补充:

(最多只允许输入30个字)