Visual Odometry

Welcome to this lab in the computer vision course TEK5030 at the University of Oslo.

This lab will be similar to the lab Pose estimation and augmented reality, but we will now create our own 3D maps instead of relying on known planar world points. This will also enable us to implement a very naive visual odometry method.

Start by cloning this repository on your machine. Then, open the lab project in your editor.

The lab is carried out by following these steps:

1. Get an overview
2. Calibrate the camera
3. Finish TwoViewRelativePoseEstimator
4. Further experiments

You will find our proposed solution at https://github.com/tek5030/solution-simple-vo.

Please try to solve the lab with help from others instead of just jumping straight to the solution ;)

Start the lab by going to the first step.

Prerequisites

As in lab-stereo, you will need to install the tek5030 camera-library for the lab. Head over to the repository and follow the installation instructions.
Note: The camera-library is preinstalled on the lab computers and in the tek5030/devcontainer Docker image.

For this lab, we can unfortunately not rely on conan to install all required OpenCV modules (namely the viz module for 3D visualization). You have a few other options:

• Solve the python lab (recommended)
• Use the lab computers
• Install OpenCV using homebrew (option for mac and linux). (See also Getting started on MacOS.)
• Try Docker toolchain in CLion (fairly experimental)
• Try devcontainer in VS Code (fairly experimental)
• Rely on virtualbox and our prepared linux image with dependencies preinstalled (see Canvas: Setting up your computer for the labs)