当前位置:网站首页>Azure Kinect DK 实现三维重建 (PC非实时版)
Azure Kinect DK 实现三维重建 (PC非实时版)
2022-06-27 19:56:00 【lucky li】
Azure Kinect DK 深度相机+NVIDIA jetson实时三维重建系列
前言
利用开源框架open3d的Reconstruction system实现Azure Kinect DK相机的三维重建,已上传代码:
https://github.com/luckyluckydadada/Azure-Kinect-DK-3D-reconstruction.git
环境说明
利用开源框架open3d的Reconstruction system实现Azure Kinect DK相机的三维重建。
开发环境是在Ubuntu18.04笔记本+DK相机上进行,可以携带笔记本连着dk相机(充电宝供电给相机)进行室外采集,采集和重建使用的open3d版本不一样,最好用conda建立两个独立的环境。
另外在win10上可以进行重建,但是利用win10+相机采集数据有些小问题,需要自行解决。
后续会将在jetson+dk相机上进行,最好可以实时。
请确保您完成了Azure Kinect DK 深度相机sdk安装(Ubuntu+ARM64,Ubuntu+AMD64,Windows)和Azure Kinect DK 点云和RGBD图的获取并通过测试,再继续下面的过程。
安装
git clone https://github.com/luckyluckydadada/Azure-Kinect-DK-3D-reconstruction.git
cd Azure-Kinect-DK-3D-reconstruction/open3d_reconstruction
录制和提取数据
cd Azure-Kinect-DK-3D-reconstruction/open3d_reconstruction
pip install -r requirements-0.10.txt # 此处安装 open3d版本为 0.10.0
录制数据,空格开始录制,esc退出录制并保存:
python sensors/azure_kinect_recorder.py --output dataset/name.mkv
提取rgb和depth图像,以及相机config和相机内参config:
python sensors/azure_kinect_mkv_reader.py --input dataset/name.mkv --output dataset/name
进行重建
cd Azure-Kinect-DK-3D-reconstruction/open3d_reconstruction
pip install -r requirements.txt # 此处安装 open3d版本为 0.12.0
如果发生目录移动,请修改dataset/name/config.json 中的"path_dataset" 和 "path_intrinsic"的值。
python run_system.py dataset/name/config.json --make --register --refine --integrate
重建demo
楼梯

彩色点云的结果:

mesh的结果:

办公室

彩色点云的结果:

mesh的结果:

参考
http://www.open3d.org/docs/latest/tutorial/ReconstructionSystem/index.html
边栏推荐
- Yarn中RMApp、RMAppAttempt、RMContainer和RMNode状态机及其状态转移
- Memoirs of actual combat: breaking the border from webshell
- Structured machine learning project (I) - machine learning strategy
- Kill the general and seize the "pointer" (Part 2)
- 爬虫笔记(2)- 解析
- 渗透学习-sql注入过程中遇到的问题-针对sort=left(version(),1)的解释-对order by后接字符串的理解
- Acwing weekly contest 57- digital operation - (thinking + decomposition of prime factor)
- Codeforces Round #722 (Div. 2)
- One to many association in MySQL to obtain the latest data in multiple tables
- 北京邮电大学|用于成本和延迟敏感的虚拟网络功能放置和路由的多智能体深度强化学习
猜你喜欢

Vue+mysql login registration case

Introduce you to ldbc SNB, a powerful tool for database performance and scenario testing

初识C语言 第二弹

Consumer finance app user insight in the first quarter of 2022 - a total of 44.79 million people

Structured machine learning project (II) - machine learning strategy (2)

Dialogue with Qiao Xinyu: the user is the product manager of Wei brand, and zero anxiety defines luxury

爬虫笔记(3)-selenium和requests

【微服务】(十六)—— 分布式事务Seata

It smells good. Since I used Charles, Fiddler has been completely uninstalled by me

6G显卡显存不足出现CUDA Error:out of memory解决办法
随机推荐
Management system itclub (Part 1)
The karsonzhang/fastadmin addons provided by the system reports an error
资深猎头团队管理者:面试3000顾问,总结组织出8大共性(茅生)
Exclusive interview with millions of annual salary. What should developers do if they don't fix bugs?
Oracle obtains the beginning and end of the month time, and obtains the beginning and end of the previous month time
01 golang environment construction
OpenSSL programming I: basic concepts
對話喬心昱:用戶是魏牌的產品經理,零焦慮定義豪華
元气森林的5元有矿之死
CDH集群之YARN性能调优
Deep learning has a new pit! The University of Sydney proposed a new cross modal task, using text to guide image matting
Do you know the full meaning of log4j2 configurations? Take you through all the components of log4j2
Workflow automation low code is the key
Process judgment - ternary operation - for loop
同花顺炒股软件可靠吗??安全嘛?
How many ways does selenium upload files? I don't believe you have me all!
Macro task and micro task understanding
Use logrotate to automatically cut the website logs of the pagoda
crontab定时任务常用命令
Mysql database experiment report (I)