Lightweight monocular SLAM from scratch with OpenCV and Open3D for visualization

Dependencies

pip install numpy opencv-contrib-python open3d

Usage

#examples
# focal length 500 and 3d with colors
F=500 COLORS=1 python3 slam.py videos/test_countryroad.mp4

# focal length 270 (default) open3d window detached and 3d in rgb green; keyframes with pose matrix ON
O3D_OUT=1 KF=1 python3 slam.py videos/city_tram.mp4

Params	Description
F=500	Sets focal length (default: 270)
COLORS=1	Enables colorized point cloud instead of default green
O3D_OUT=1	Displays Open3D window separately
DETECTOR=ORB	Uses ORB instead of default GFTT + BRIEF for feature detection
KF=1	Shows keyframes as wireframe frustum squares
F_MASK=0.7	Uses only bottom 70% of the image for feature detection, negative values use top portion
SKY_AUTO=1	Auto-skips top 40% if features are detected above (like clouds)

WSL2 Fix

export XDG_SESSION_TYPE=x11