human pose estimation
Posted guantian
tags:
篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了human pose estimation相关的知识,希望对你有一定的参考价值。
2D Pose estimation主要面临的困难:遮挡、复杂背景、光照、真实世界的复杂姿态、人的尺度不一、拍摄角度不固定等。
单人姿态估计
传统方法:基于Pictorial Structures, DPM
? 基于深度学习的算法包括直接回归坐标(Deep Pose)和通过热力图回归坐标(CPM, Hourlgass)
目前单人姿态估计,主流算法是基于Hourlgass各种更改结构的算法。
多人姿态估计
二维图像姿态估计基于CNN的多人姿态估计方法,通常有2个思路(Bottom-Up Approaches和Top-Down Approaches):
(1)Top-Down Approaches,即two-step framework,就是先进行行人检测,得到边界框,然后在每一个边界框中检测人体关键点,连接成一个人形,缺点就是受检测框的影响太大,漏检,误检,IOU大小等都会对结果有影响,算法包括RMPE、Mask-RCNN 等。
(2)Bottom-Up Approaches,即part-based framework,就是先对整个图片进行每个人体关键点部件的检测,再将检测到的部件拼接成一个人形,缺点就是会将不同人的不同部位按一个人进行拼接,代表方法就是openpose、DeepCut 、PAFs。
tricks
- 采用多尺度,多分辨率的网络结构
- 采用基于Residual Block来构建网络
- 扩大感受野(large kernel, dilation convolution, Spatial Transformer Network、hourglass module)
- 预处理很重要(将人放在输入图片的中心,人的尺度尽量归一化到统一尺度,对图片进行翻转、旋转)
- 后处理同样重要
openpose源码中subset输出的关键点顺序是:1鼻子,2脖子,3右肩,4右肘,5右腕,6左肩,7左肘,8左腕,9右髋,10右膝,11右踝,12左髋,13左膝,14左踝,15左眼,16右眼,17左耳,18右耳,19 pt19
输出格式;https://www.aiuai.cn/aifarm712.html
CPM
paper:
https://blog.csdn.net/shenxiaolu1984/article/details/51094959
openPose
GitHub:
Realtime_Multi-Person_Pose_Estimation
https://github.com/CMU-Perceptual-Computing-Lab/openpose
配置:
https://blog.csdn.net/lgh0824/article/details/75949477
https://blog.csdn.net/hk121/article/details/83537350
openPose解析
https://blog.csdn.net/qq_27158179/article/details/82717821
https://www.jianshu.com/c/8602d176d8ea?utm_source=desktop&utm_medium=notes-included-collection
https://zhuanlan.zhihu.com/p/48507352
[OpenPose翻译] Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields ?原文翻译(注释版)
https://blog.csdn.net/kenllf/article/details/79702078
Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields ? 实时多人人体姿态估计论文原理讲解
https://blog.csdn.net/Lin_xiaoyi/article/details/78838393
https://blog.csdn.net/yxr403614258/article/details/77977330
Paper reading: Realtime Multi-person 2D Pose estimation using Part Affinity Fields(1)
https://blog.csdn.net/yengjie2200/article/details/68064095
openpose实验总结
https://blog.csdn.net/qq_20657717/article/details/81707746
肤色检测
https://blog.csdn.net/yangtrees/article/details/8269984
基于颜色检测物体
http://www.cnblogs.com/wangxinyu0628/p/5928824.html
项目编译:
https://blog.csdn.net/zb1165048017/article/details/82115724
姿态估计的应用:
https://blog.csdn.net/itchosen/article/details/77200384
https://blog.csdn.net/shenxiaolu1984/article/details/51094959
https://blog.csdn.net/yeahDeDiQiZhang/article/details/78131566
https://www.cnblogs.com/JillBlogs/p/9098989.html
Stacked Hourglass算法详解
https://blog.csdn.net/shenxiaolu1984/article/details/51428392
代码阅读】OpenPose(Pytorch Realtime Multi-Person Pose Estimation)
https://blog.csdn.net/a529975125/article/details/80991781
pytorch千千问
https://blog.csdn.net/daniaokuye/article/details/78851479
以上是关于human pose estimation的主要内容,如果未能解决你的问题,请参考以下文章
Deep High-Resolution Representation Learning for Human Pose Estimation
论文翻译ViTPose: Simple Vision Transformer Baselines for Human Pose Estimation
human_pose_estimation_demo的再进一步研究