IPhoneDataCapture

App + script to read iphone ARkit position and read and send images over tcp network to python script CaptureData.py. They to visualize images and positions of recordings with Qt app.

To capture trajectory data

• Install Iphone app on your IPhone located in ARKitStreamer folder
• Set phone IP (visible in app) in python script IPHONE_IP in arkit_loop() (make sure phone and laptop are connected to same network)
• Run: python CaptureData.py and press start and stop record, visualize 3d trajctory afterwards.

---

Extra: Generate point clouds from 2D images

To create 3D point cloud from 2D images (advised to run in collab for CUDA and max speed)

Create 3d point cloud with SfM pipeline (colmap)

• Install colmap (macOs): brew install colmap
• export trajectory and images for colmap (with button) (4 hz, total 52 images)
• run feature extraction (colmap): colmap feature_extractor --database_path database.db --image_path colmap_export/images --ImageReader.camera_model PINHOLE (3 sec)
• run feature matching: colmap exhaustive_matcher --database_path database.db (50 sec)
• copy camera.txt and images.txt to colmap as prior:

mkdir -p colmap_export/sparse/0
cp colmap_export/sparse/cameras.txt colmap_export/sparse/0/ 
cp colmap_export/sparse/images.txt colmap_export/sparse/0/`

• run colmap 3d reconstruction: colmap mapper --database_path database.db --image_path colmap_export/images --output_path colmap_export/sparse --Mapper.init_image_id1 1 (1 min 40 sec)
• visualize running: colmap gui, press File -> Import Model -> (select "sparse/0" folder)

TODO: use DISK and lightglue and compare with current colmap only approach

Create dense 3d point cloud with MVS pipeline (colmap)

TODO: Implement dense pointcloud with with colmap (MVS), run in collab

Export to ply, visualize with open3D

• export to ply format (for gaussan splatting): colmap model_converter --input_path colmap_export/sparse/0 --output_path output.ply --output_type PLY
• Visualize again with open3d: python visualize_colmap.py