High-Precision Point Clouds Pose Estimation with Open3D Global Registration
ฝัง
- เผยแพร่เมื่อ 30 ต.ค. 2021
- Inside my school and program, I teach you my system to become an AI engineer or freelancer. Life-time access, personal help by me and I will show you exactly how I went from below average student to making $250/hr. Join the High Earner AI Career Program here 👉 www.nicolai-nielsen.com/aicareer (PRICES WILL INCREASE SOON)
You will also get access to all the technical courses inside the program, also the ones I plan to make in the future! Check out the technical courses below 👇
_____________________________________________________________
In this video 📝 we are going to take a look at how to do Global Registration for Pose Estimation of Point Clouds. First of all, we will talk about global registration and the idea behind it. We will then jump into code and see some examples of how to do pose estimation on point clouds, apply transformations and align two point clouds with global registration.
If you enjoyed this video, be sure to press the 👍 button so that I know what content you guys like to see.
_____________________________________________________________
🛠️ Freelance Work: www.nicolai-nielsen.com/nncode
_____________________________________________________________
💻💰🛠️ High Earner AI Career Program: www.nicolai-nielsen.com/aicareer
⚙️ Real-world AI Technical Courses: (www.nicos-school.com)
📗 OpenCV GPU in Python: www.nicos-school.com/p/opencv...
📕 YOLOv7 Object Detection: www.nicos-school.com/p/yolov7...
📒 Transformer & Segmentation: www.nicos-school.com/p/transf...
📙 YOLOv8 Object Tracking: www.nicos-school.com/p/yolov8...
📘 Research Paper Implementation: www.nicos-school.com/p/resear...
📔 CustomGPT: www.nicos-school.com/p/custom...
_____________________________________________________________
📞 Connect with Me:
🌳 linktr.ee/nicolainielsen
🌍 My Website: www.nicolai-nielsen.com/
🤖 GitHub: github.com/niconielsen32
👉 LinkedIn: / nicolaiai
🐦 X/Twitter: / nielsencv_ai
🌆 Instagram: / nicolaihoeirup
_____________________________________________________________
🎮 My Gear (Affiliate links):
💻 Laptop: amzn.to/49LJkTW
🖥️ Desktop PC:
NVIDIA RTX 4090 24GB: amzn.to/3Uc7yAM
Intel I9-14900K: amzn.to/3W4Z5Cb
Motherboard: amzn.to/4aR6wBC
32GB RAM: amzn.to/3Jt2XVR
🖥️ Monitor: amzn.to/4aLP8hh
🖱️ Mouse: amzn.to/3W501GH
⌨️ Keyboard: amzn.to/3xUGz5b
🎙️ Microphone: amzn.to/3w1F1WK
📷 Camera: amzn.to/4b4Ryr9
_____________________________________________________________
Tags:
#Open3D #PointClouds #ComputerVision - วิทยาศาสตร์และเทคโนโลยี
Join My AI Career Program
www.nicolai-nielsen.com/aicareer
Enroll in My School and Technical Courses
www.nicos-school.com
Thank you! As always, very interesting material!
Hmm can't find the code for this video on your github.
I run all three algorithms with my custom color point clouds. Firstly, It takes a couple of minutes to run feature (nearest neighbours) based algorithm. Secondly, there is no improvement in point cloud aligning. the RANSAC is returning identity matrix. WHat should be the possible issue and how can I determine correct voxel size and distance threshold according to my point cloud data.Thanks
Hello there! I've some question about the usage of this algorithm. I hope someone can clarify my doubts. I'd like to use the described approaches to perform an "object detection" having a point cloud of the scene and a point cloud of the object to detect inside the scene. Usually the objects I'm working with are no bigger than 10 cm, and the scene is small as well. I'm wondering what changes I should do in order to use the global registration to do the point cloud alignment (object detection) and get the transformation from the object model to the scene. Can someone help me for my project? Thanks!!
Hey,
I have nearly the same task to do and unfortunately the video doesn't help.
So did you solve your problem, if yes can you tell me how ? I don't know what to do anymore.
Can you please provide a link to the jupyter notebook? I can't find it anywhere.
I think there are two coefficient to validate matching result: RMSE and Fitness, I want to ask you is which is more important. If Fitness value not equal 1, Can it be considered a good matching? because all examples I see not reach 1 with Fitness value (may be Fitness value is about 0.4-0.5 but they look like a good matching).
Don't think u will ever come close to a fitness value of 1. I'd look at the error value instead and see how low that is
- You mean RMSE is more important than Fitness.
- My data after being put into open3D was wrongly sized (scaled up), and had to use voxel sizes up to 1-2 to match (I think meter units). I want to ask you about the units used in Open 3D. How to convert meters to minimeter.
I would not say one of them is more important, but if u have a really low error that's great. U should look at both metrics and also visually and then determine if something is odd.
The unit of the points u have in ur point cloud depends on the sensor u have used and how u generated the point cloud. But the absolute positions and distances are with respect to ur sensor used to generate it.
Thanks.
Thanks for the support!
Wait, you are intentionally applying rotation transformations to a second point cloud to create an example case for RANSAC to solve? I understood that correctly?
Yeah that is correct
How can we add texture to mesh
Hai. How to change the data with my own data? I only load my data but cannot registered
where is pose estimation happening here?
Nice video. But do you have any example where the source and target point clouds are different ? so they do not match 100 %
Hi in this example the point clouds are similar, but in a lot of cases they do not match if u want to find an object in a scene for example