<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[qq_33690342的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/qq_33690342</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; qq_33690342]]></copyright><item><title><![CDATA[ PA-MPJPE、G-MPJPE 和 Per-metre drift 三个指标]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/145809870</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/145809870</guid><author>qq_33690342</author><pubDate>Sun, 23 Feb 2025 14:45:06 +0800</pubDate><description><![CDATA[在消除平移、旋转和缩放影响后，计算预测姿态与真实姿态之间各关节点的平均欧氏距离误差。在全局坐标系下（未经对齐），直接计算预测姿态与真实姿态各关节点的平均欧氏距离误差。在连续运动过程中，系统每移动1米，累积的位置误差增长量（通常以米/米或百分比表示）。]]></description><category></category></item><item><title><![CDATA[逆向运动学（IK）与正向运动学（FK）详解]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/145807421</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/145807421</guid><author>qq_33690342</author><pubDate>Sun, 23 Feb 2025 12:03:14 +0800</pubDate><description><![CDATA[FK & IK]]></description><category></category></item><item><title><![CDATA[单目摄像头的相机参数估计中translation 和相机深度是什么关系]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/145807163</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/145807163</guid><author>qq_33690342</author><pubDate>Sun, 23 Feb 2025 11:33:18 +0800</pubDate><description><![CDATA[单目摄像头的相机参数估计中translation 和相机深度关系]]></description><category></category></item><item><title><![CDATA[P_all: 投影矩阵（Projection Matrix）]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/145420672</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/145420672</guid><author>qq_33690342</author><pubDate>Sun, 02 Feb 2025 16:15:41 +0800</pubDate><description><![CDATA[每个摄像头的投影矩阵 PiPi​ 是一个 3×43×4 的矩阵，用于将世界坐标系中的 3D 点 X=[X,Y,Z,1]TX=[X,Y,Z,1]T 投影到该摄像头的 2D 图像平面上的点 u=[u,v,1]Tu=[u,v,1]T。通过构造超定方程组 AX=0AX=0，利用所有摄像头的投影矩阵 PiPi​ 和对应的 2D 点 (ui,vi)(ui​,vi​)，求解 3D 点 XX。X=[X,Y,Z,1]TX=[X,Y,Z,1]T 是世界坐标系中的齐次 3D 坐标。是所有摄像头的投影矩阵的集合。]]></description><category></category></item><item><title><![CDATA[复现SMPLify-X: Ubuntu22.04, Cuda-11.3, GPU=3090Ti]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/144176361</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/144176361</guid><author>qq_33690342</author><pubDate>Tue, 03 Dec 2024 00:42:53 +0800</pubDate><description><![CDATA[3090Ti  CUDA 最低支持版本需要>=cuda-11.1  Ubuntu 22.04。]]></description><category></category></item><item><title><![CDATA[.pkl文件的反序列化]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/144154712</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/144154712</guid><author>qq_33690342</author><pubDate>Sat, 30 Nov 2024 14:41:06 +0800</pubDate><description><![CDATA[【代码】.pkl文件的反序列化。]]></description><category></category></item><item><title><![CDATA[SMPL 、 SMPL-X、SMPLify-X、SMPLer-X三维人体模型]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/144154032</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/144154032</guid><author>qq_33690342</author><pubDate>Sat, 30 Nov 2024 14:01:34 +0800</pubDate><description><![CDATA[特性SMPLSMPL-XSMPLify-XSMPLer-X功能类型人体模型综合的人体、面部和手部模型2D 到 3D 关键点重建扩展的 2D 到 3D 重建框架参数类型形状和姿态参数多种参数（包括手和面）优化参数（基于 2D 关键点）适应多模态数据的优化参数应用领域运动捕捉、动画制作游戏、电影等多媒体应用计算机视觉中的 3D 重建更复杂的数据重建包括大规模应用研究时间2015201920202021复杂性基本中等高更高。]]></description><category></category></item><item><title><![CDATA[Openpose]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/142171058</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/142171058</guid><author>qq_33690342</author><pubDate>Thu, 12 Sep 2024 11:41:42 +0800</pubDate><description><![CDATA[OpenPose基于Part Affinity Fields（PAFs，部件亲和场）的概念，这是一种非参数化的表示方法，用于学习图像中个体的身体部位之间的关联。是否在某个肢体（limb）上的方法主要依赖于部件亲和场（Part Affinity Fields，简称PAFs）。OpenPose是一个开源库，可以用于身体、脚、手和面部关键点的实时检测，广泛应用于需要2D人体姿态估计的计算机视觉和机器学习应用。采用多阶段CNN架构，每个阶段都会迭代预测PAFs和置信度图，并通过中间监督来细化预测结果。]]></description><category></category></item><item><title><![CDATA[DWPose & RTMPose]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/142168904</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/142168904</guid><author>qq_33690342</author><pubDate>Thu, 12 Sep 2024 11:03:05 +0800</pubDate><description><![CDATA[RTMPose 基于 MMPose，旨在解决工业界中对轻量级、高速度和高精度姿态估计的需求。]]></description><category></category></item><item><title><![CDATA[SimCC]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/142136377</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/142136377</guid><author>qq_33690342</author><pubDate>Wed, 11 Sep 2024 12:42:59 +0800</pubDate><description><![CDATA[SimCC（Simple Coordinate Classification）的核心思想是将人体关键点的位置估计问题分解为两个独立的分类问题：一个是水平坐标（x坐标）的分类，另一个是垂直坐标（y坐标）的分类。]]></description><category></category></item><item><title><![CDATA[mmpose 安装及使用pretrained model 运行inferencer_demo.py Running RTMPose,ViTPose,YoloPose模型]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/140950073</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/140950073</guid><author>qq_33690342</author><pubDate>Tue, 06 Aug 2024 11:23:18 +0800</pubDate><description><![CDATA[mmpose 安装运行human pose estimation 模型ViTPose，RTMPose, Yolopose]]></description><category></category></item><item><title><![CDATA[Alphapose 安装 CUDA11.3+pytorch1.11.0+torchvision0.12.0+torchaudio0.11.0+gcc8.4.0+setuptools57.5.0]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/140676030</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/140676030</guid><author>qq_33690342</author><pubDate>Thu, 25 Jul 2024 00:02:22 +0800</pubDate><description><![CDATA[2. Install PyTorch (除了下述torch相关的包用conda 安装，剩余跟Python相关的包要用pip 安装，因为有些软件如果跟Python的版本不兼容的话，在安装其他包时，会修改Python和torch的版本，我曾用conda install numpy，结果conda自动安装了其他版本的Python和torch)setuptools的版本也要注意！其余的我基本上都是按照上面的那篇博客安装的。如果编译的时候遇到什么罕见的乱七八糟的问题，基本上都是软件版本并不兼容导致的。]]></description><category></category></item><item><title><![CDATA[Openpose install: Ubuntu20.04, RTX A4000 16G, CUDA11.7.1,CuDNN8.5.0,OpenCV3.4.5]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/140572556</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/140572556</guid><author>qq_33690342</author><pubDate>Sat, 20 Jul 2024 15:51:46 +0800</pubDate><description><![CDATA[]]></description><category></category></item><item><title><![CDATA[视觉里程计学习笔记]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/130320468</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/130320468</guid><author>qq_33690342</author><pubDate>Tue, 09 May 2023 15:35:22 +0800</pubDate><description><![CDATA[从E分解到R, t的过程中，如果相机发生纯旋转，导致t为0，那么，得到的E也将为0，这将导致无法求解R。可以依靠H（单应矩阵）求取旋转，但仅有旋转时，无法用三角测量估计特征点的空间位置。3D-2D方法不需要使用对极约束，又可以在很少的匹配点中获得较好的运动估计，是一种最重要的姿态估计方法。PnP使求解3D到2D点对运动的方法。特征点的3D位置可以由三角化或者RGB-D相机的深度图确定。根据已经求出的本质矩阵E，恢复出相机的运动R,t. 可通过奇异值（SVD）分解得到。一组为3D，一组为2D。]]></description><category></category></item><item><title><![CDATA[Segment Anything论文阅读笔记]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/130006468</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/130006468</guid><author>qq_33690342</author><pubDate>Fri, 07 Apr 2023 11:38:15 +0800</pubDate><description><![CDATA[Segment Anything (SA) project: a new task, model, and dataset for image segmentation.we built the largest segmentation dataset to date (by far), with over 1 billion masks on 11M licensed and privacy respecting images. The model is designed and trained to b]]></description><category></category></item><item><title><![CDATA[AirVO环境搭建ubuntu20.04.5]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/129023435</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/129023435</guid><author>qq_33690342</author><pubDate>Tue, 14 Feb 2023 11:44:00 +0800</pubDate><description><![CDATA[AirVO 环境搭建]]></description><category></category></item><item><title><![CDATA[EDPLVO:Efficient Direct Point-Line Visual Odometry note]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/128827098</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/128827098</guid><author>qq_33690342</author><pubDate>Fri, 03 Feb 2023 10:42:44 +0800</pubDate><description><![CDATA[This paper extends the photometric error for lines.  we proved that the 3D points of the points on a 3D line are determined by the inverse depths of the endpoints of the 2D line, and derive a closed-form solution for this problem.   This property can signi]]></description><category></category></item><item><title><![CDATA[视觉slam线匹配算法]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/128799717</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/128799717</guid><author>qq_33690342</author><pubDate>Mon, 30 Jan 2023 11:49:57 +0800</pubDate><description><![CDATA[线特征广泛存在于人为环境中，可以提供额外的约束。视觉SLAM中线特征的挑战是先特征的匹配问题。LSD是一种局部提取直线的算法，速度比Hough要快。LSD是一种直线检测分割算法，它能在线性的时间内得出亚像素级精度的检测结果。该算法被设计成自适应模式，无需手动调参。opencv使用头文件及命令空间。]]></description><category></category></item><item><title><![CDATA[AirVO: An Illumination-Robust Point-Line Visual Odometry 论文笔记及源码阅读]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/128421239</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/128421239</guid><author>qq_33690342</author><pubDate>Fri, 23 Dec 2022 17:50:10 +0800</pubDate><description><![CDATA[AirVO,  一个基于点、线特征具有光照鲁棒性行且精确的双目视觉里程计系统。为了实现对于光照变化的鲁棒性，我们引入了基于学习的特征提取和匹配方法，并设计了一个新颖的视觉里程计管道：包括特征跟踪、三角测量、关键帧选择和图形优化等。我们还采用了环境中的长线特征来提高系统的准确性。我们提出了一个光照鲁棒性的线追踪方法，其中点特征追踪、点分布和线特征被用来实现线的匹配。]]></description><category></category></item><item><title><![CDATA[SLAM学习笔记(二)]]></title><link>https://blog.youkuaiyun.com/qq_33690342/article/details/128066194</link><guid>https://blog.youkuaiyun.com/qq_33690342/article/details/128066194</guid><author>qq_33690342</author><pubDate>Tue, 29 Nov 2022 18:24:15 +0800</pubDate><description><![CDATA[相机将三维世界中的坐标点（单位米）映射到二维图像平面（单位为像素）的过程中能够用一个几何模型进行描述。]]></description><category></category></item></channel></rss>