Skip to content

Latest commit

 

History

History
124 lines (96 loc) · 2.85 KB

running_on_own_dataset.md

File metadata and controls

124 lines (96 loc) · 2.85 KB

Running on own dataset

Calibration

To run dsopp_main on your sequence calibration should be obtained

  • photometric calibration
    • gamma correction pcalib.txt
    • vignetting vignetting.png
  • geometric calibration calib.txt

For photometric calibration please refer to ONLINE PHOTOMETRIC CALIBRATION and mono_dataset_code

For geometric calibration for example to OpenCV calibration tutorial

Currently supported camera models:

Pinhole Camera Model

calib.txt file for pinhole camera model:

pinhole
width height
focal_x focal_y
center_x center_y

An example could be found in test/test_data/track30seconds/calib.txt

Simple Radial Camera Model

This model is the same as the radial model in colmap or pinhole from OpenCV but only k1 k2 is non-zero.

simple_radial
width height
focal
center_x center_y
k1 k2

Other files

times.csv contains timestamps and frames ids. For example

0 5.84253863990307
1 5.89253864064813
2 5.94253864139318

If you do not plan to use any sensors, then create timest.csv with number of frames lines with the following content

0 0
1 1
2 2
3 3
...

Video file or Image folder with your images, and videos. If you decided to use images, please name them %06d.png.

[Optional] Mask uses black to exclude region, and white to include.

Then the minimal config mono.yaml would be:

Minimal config 
sensors:
  - id: camera_1
    type: camera
    provider:
      type: video
      video_file: "./images.mkv"
      timestamps: "./times.csv"
      timestamps_start_id: 0
    model:
      calibration: "./calib.txt"
      shutter_time_seconds: 0
      photometric_calibration: "./pcalib.txt"
      vignetting: "./vignetting.png"
    position: world

time:
  type: no_synchronization

tracker:
  type: monocular
  sensor_id: camera_1
  number_of_desired_points: 2000
  keyframe_strategy:
    strategy: mean_square_optical_flow
    factor: 1.0 #the larger the more keyframes are created (e.g., 2 = double number of keyframes)
  marginalization_strategy:
    strategy: sparse
    minimum_size: 5
    maximum_size: 7
    maximum_percentage_of_marginalized_points_in_frame: 0.95

initializer:
  type: monocular
  initializer_type: calibrated
  sensor_id: camera_1
  essential_matrix_ransac_threshold: 0.2 # in pixels at 1280x720 image
  pnp_ransac_threshold: 0.2 # in pixels at 1280x720 image
  se3_inlier_ratio: 0.6
  reprojection_threshold: 1 # in pixels at 1280x720 image
  pnp_inlier_ratio: 0.5
  features_extractor:
    type: ORB
    number_of_features: 500
  keyframe_strategy:
    strategy: wait_for_movement
    sliding_window_length: 4

gnss_alignmenter:
  mode: off