Hi Florent, very nice video as always ! It's quite funny you bring it today as coincidently last few days i worked on a prototype using o3d ransac to detect roof planes from aerial lidar files. It looks quite efficient for theses simple usecases that don't necessarily need heavy NN to detect 3D shapes. I'll share that with you when ready
I am currently working on a 3D search problem and need to address the computational efficiency when dealing with a large 3D database. Specifically, I have a dataset consisting of many 3D Point Clouds corresponding to components in the manufacturing industry. My goal is to find components with the most similar geometric structure. So far, I have tried using eigenvalues and eigenvectors, moving the cluster center to the origin coordinates to find the transformation that minimizes the Hausdorff distance. This method yields fairly good results compared to the current approach, but I am encountering issues with computational efficiency for a large 3D database. I am considering converting the 3D Point Cloud into an embedding and using the Cosine Similarity measure. If the embedding is well-represented, we would only need to perform a dot product calculation. Could you please advise if this method is appropriate? Or is there a better solution I should consider? Any feedback you provide would be highly valuable to me. Thank you very much for your assistance.
Hello Sir, I am working on a project in which i have created a 3D point cloud of rectangular sheet having some circular holes (Sheet metal part with holes) in google colab using open 3D and save the file in pcd file format. Now i want to process my file using RANSAC for measuring the hole parameters like the dia, ovality etc.. So how can i do this
when I use a real point cloud obtained from a Drone, I get a big sphere on top of my point cloud, what could be the reason? the scale? how can I fix it?... Could be applied this for real time detection?
Yep, this may be due to haze effects for example. You can definitly get rid of it. If it is very continuous, then this approach work. If this is a bit more random, then noise fitlering techniques may work better
please do work on sfm generated point clouds too, I tried to measure dimension using your technique from one of your previous videos but didn't work please if it's possible can you make a video on that.
![Lp. Сердце Вселенной #60 РОЖДЕНИЕ ЛОЛОЛОШКИ [Финал] • Майнкрафт](http://i.ytimg.com/vi/YoR0pAV9FVQ/mqdefault.jpg)
Hi Florent, very nice video as always ! It's quite funny you bring it today as coincidently last few days i worked on a prototype using o3d ransac to detect roof planes from aerial lidar files. It looks quite efficient for theses simple usecases that don't necessarily need heavy NN to detect 3D shapes. I'll share that with you when ready
Nice work! A tutorial on learning based point cloud registration would be helpful, too. Thank you.
added to the roadmap!
I am currently working on a 3D search problem and need to address the computational efficiency when dealing with a large 3D database. Specifically, I have a dataset consisting of many 3D Point Clouds corresponding to components in the manufacturing industry. My goal is to find components with the most similar geometric structure.
So far, I have tried using eigenvalues and eigenvectors, moving the cluster center to the origin coordinates to find the transformation that minimizes the Hausdorff distance. This method yields fairly good results compared to the current approach, but I am encountering issues with computational efficiency for a large 3D database.
I am considering converting the 3D Point Cloud into an embedding and using the Cosine Similarity measure. If the embedding is well-represented, we would only need to perform a dot product calculation.
Could you please advise if this method is appropriate? Or is there a better solution I should consider? Any feedback you provide would be highly valuable to me.
Thank you very much for your assistance.
Hello Sir, I am working on a project in which i have created a 3D point cloud of rectangular sheet having some circular holes (Sheet metal part with holes) in google colab using open 3D and save the file in pcd file format. Now i want to process my file using RANSAC for measuring the hole parameters like the dia, ovality etc.. So how can i do this
Beautiful project, you can shoot an email through my website: learngeodata.eu, and we will figure this out
@@FlorentPoux Thanks for the reply.
Hi, you mention an article from 2019. Which article is that?
This is the article: www.sciencedirect.com/science/article/abs/pii/S0926580522001236
when I use a real point cloud obtained from a Drone, I get a big sphere on top of my point cloud, what could be the reason? the scale? how can I fix it?... Could be applied this for real time detection?
Yep, this may be due to haze effects for example. You can definitly get rid of it. If it is very continuous, then this approach work. If this is a bit more random, then noise fitlering techniques may work better
please do work on sfm generated point clouds too, I tried to measure dimension using your technique from one of your previous videos but didn't work please if it's possible can you make a video on that.
Great proposal! I will!
I am creating object detection on Lidar LAS dataset how to do that
Did you find a solution?