6D Pose Estimation Data Annotation

Author: Martin Purgat

Supervisor: RNDr. Zuzana Berger Haladová, PhD.

Consultant: Mgr. Lukáš Gajdošech

Project Goals

Project Description

6D pose estimation is a computer vision task that aims to determine the position and orientation of an object in 3D space. It is a crucial step in many robotic applications, such as bin picking, pick and place, or autonomous driving. The goal of this project is to develop a tool for semi-automatic annotation of 6D pose estimation datasets. The tool will be implemented as a web application with a virtual reality interface. The user will be able to load an RGBD point cloud of a scene, CAD models of objects and annotate it by placing the objects right into the point cloud. The user will be able to correct the pose by moving the object with a controller ("virtual hand" metaphor) right in VR. The tool will also allow the user to refine the pose by using human in the loop version of the ICP algorithm. The user will be able to export the annotated point cloud. Finally, we will conduct a user study to evaluate the usability of the tool.

Project Progress

September 2022

Thinking about doing some VR related project. I met with my supervisor and we discussed some project ideas. Eventually, we decided to do a project related to 6D pose annotation.

October 2022

Started working on the project. I did some research on existing 6D pose estimation datasets and annotation tools. I also started learning about point cloud processing algorithms and frameworks used in this field.

November 2022

We settled on final project goals. I started to play around with libraries and frameworks I was planning to use in the project. I also started to learn about VR development.

December 2022

I took a break from the project to work on some other projects. Also, hey, it's Christmas time!🎄🎁

January 2023

I created a web page for the project and started thinking about the structure, visuals, and general design of the application.

February 2023

I started implementing the application. I decided to use Vue.js and Three.js. Created my first dynamic VR scene!

March 2023

I was implementing the basic features and functionality of the application.

April 2023

I started working on some advanced features, such as experimenting with human in the loop ICP.

May 2023

We conducted a user study to evaluate the usability of the application.

It also started writing the final text.

June 2023

Thesis submitted! Thoughts and prayers please!🙏

Project diary

Text

You can read the final text of the thesis here.

Screenshots of the application

References

All references can be found in the final text of the thesis.