ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

finding the highest match between two images and/or point clouds

asked 2021-11-12 08:03:08 -0500

Delbina gravatar image

hi

which kind of methods are used for finding the highest match between two images and/or point clouds? I would be grateful if anybody can support.

ros, kinetic 16.04

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
1

answered 2021-11-12 09:07:07 -0500

osilva gravatar image

updated 2021-11-12 09:25:49 -0500

The best program for matching point clouds that I am aware is: 3DSmoothNet (https://github.com/zgojcic/3DSmoothNet)

It uses a siamese deep learning architecture and fully convolutional layers using a voxelized smoothed density value (SDV) representation with 94%+ matching for rotation invariant point clouds. There is also a very interesting paper, highly recommend to read on the subject. This can be part of your pipeline when processing point clouds from different sources.

One of the things to be aware when using is that both point clouds need to have the same scale, otherwise it will be an issue.

edit flag offensive delete link more

Comments

Thanks a lot @osilva, and do you know what are the requirements for installing this pkg? and what about images? because I have heard that working with point clouds is really hard, so first I want to start with images. and find the image which has the highest match with the one saved as reference.

Delbina gravatar image Delbina  ( 2021-11-12 09:59:29 -0500 )edit

For images, start with the basics using RANSAC application in homography, perhaps learn with OpenCV first. The video tutorial is for Matlab but you will get the idea: https://ros-developer.com/2017/12/26/....

Working with pointclouds it's more challenging but it's a very exciting field that merges Computer Vision and Robotics. Hope this is what you were looking for and if so kindly accept the answer.

osilva gravatar image osilva  ( 2021-11-12 10:10:56 -0500 )edit

The installation steps for 3DSmoothNet are in the repo but you need to be familiar with deep learning to use it.

osilva gravatar image osilva  ( 2021-11-12 10:12:37 -0500 )edit

Thanks a lot @osilva yes, it was the one that i am searching for

Delbina gravatar image Delbina  ( 2021-11-12 10:13:39 -0500 )edit

@osilva, sorry, do you know any reference or link for learning deep learning that explains clearly?

Delbina gravatar image Delbina  ( 2021-11-12 11:04:45 -0500 )edit

i'd recommend to learn PyTorch or Tensorflow there are many great books. It's a very broad field, specifically for Robotics the area of Reinforcement Learning and Computer Vision with Deep Learning will give you some strong foundation

osilva gravatar image osilva  ( 2021-11-12 12:57:54 -0500 )edit

Question Tools

2 followers

Stats

Asked: 2021-11-12 08:03:08 -0500

Seen: 74 times

Last updated: Nov 12 '21