pySLAM-D is a SLAM repository for RGB-D images in python that computes the camera trajectory and yileds a 3D reconstruction of the scene. This code is suitable for Reinforcement Learning purposes and utilizes existing C++ libraries for fast visual odometry, loop-closure detection, and pose graph optimization. We have tested this code for visual SLAM in the Habitat-Sim environement.
We have tested this library in Ubuntu 18.04 and CentOS 7.
We suggest using a conda environment.
conda update -q conda
conda create -q -n pyslamd python=3.7 opencv=3.4.* numpy boost py-boost cmake eigen
conda activate pyslamd
The gtsam library is used for pose graph optimization. Make sure to build it with python bindings.
We use Open3D for 3D data handling.
conda install -c open3d-admin open3d
We make use of the TEASER++ library for a robust and certifiable front-end. For loop-closure detection, we use the FBOW library and a modified version of pyfbow for python bindings. These two libraries are included in the Thirdpary folder. To install them, follow these instructions:
cd Thirdparty
chmod +x build.sh
./build.sh
run habitat.py