SLAM.jl

Simultaneous Localization and Mapping.

Features

Monocular / Stereo modes.
Bundle-Adjustment over a subset of Keyframes.
Local Map Matching for re-tracking lost Mappoints back into Frame.

Install

]add https://github.com/pxl-th/SLAM.jl.git

Usage

Minimal abstract example of how to use.

using SLAM

camera = Camera(...)
params = Params(; stereo=false, ...)
manager = SlamManager(camera, params)
manager_thread = Threads.@spawn run!(manager)

images = Matrix{Gray{Float64}}[...]
timestamps = Float64[...]

for (time, image) in zip(timestamps, images)
    add_image!(manager, image, timestamp)
    sleep(1e-2)
end

manager.exit_required = true
wait(manager_thread)

For a more detailed explanation, see Tutorial section. If you prefer looking at raw code instead, look at a complete KITTY Dataset example.

To launch the example, download KITTY Dataset, change base_dir and save_dir to you own (base_dir should be the root directory of KITTY dataset folder) and launch in from the REPL:

include("./example/kitty/main.jl")

# Launch SLAM
n_frames = 100
slam_manager, visualizer = main(n_frames)

# Replay saved result
replay(n_frames)

Warning

These are 3 components in the SLAM.jl that need to run on a separate thread. So, remember to launch Julia with at least -t4 flag (+ 1 for the main thread).

Results

Final map on the 00 sequence taken from KITTY dataset in stereo mode.

KITTY 00 sequence