This article presents SLAM with semantic segmentation, a functional monocular SLAM system operating on the basis of package configuration with ArUco initialization. The system is capable of not only creating camera tracking and mapping, but also using a Neural Network to identify stationary objects. Based on the algorithms of recent years, we have developed a new system that selects points and keyframes and creates a map reflecting the sequence of poses of the camera. You can see obtained results in the results folder.

It is a final project for the "Perception in Robotics" course at Skoltech, 2024.

Create an virtual environment:

cd "<project_dir>"
virtualenv -p python3 venv
source venv/bin/activate
pip install -r requirements.txt

Activate the environment:

source venv/bin/activate

And run src/main.ipynb file. Done!