This is an implementation for the practical course Multi-Camera Computer Vision and Algorithms at TUM.
Watch the pipeline run on the KITTI benchmark on Youtube
-
OpenCV 3.3 or later
-
Ceres 1.13 or later
-
Dlib 19.9 or later
mkdir build
cdir build
cmake ..
cmake --build .
Run binary with path to configuration file as argument. The configuration may look like this:
[Settings]
fancy_video = 1
verbose = 1
video_path = ../../../tracker.avi
error_path = ../../../error.txt
[Odometry]
; When extracting 2d features, tries to extract at least this amount
min_tracked_features = 400
; Tolerated number of seen 3d points before triangulating new features
tracked_features_tol = 150
; Number of frames used to initialise odometry pipeline
init_frames = 5
; Number of frames to track
frames = 600
; Number of frames used for bundle adjustment
bundle_size = 5
[ceres]
max_iterations = 5
[KITTI]
image_dir = D:\Odometry\dataset\sequences\07\image_0
; Number of camera in calibration file
camera = 0
camera_calibration = D:\Odometry\dataset\sequences\07\calib.txt
poses = D:\Odometry\dataset\poses\07.txt