Data Structure

The main data for the Invisible Map is stored in JSON (JavaScript Object Notation) files. There are two sets of JSON files; the raw data, which is collected by Invisible Map Creator and uploaded to Firebase, and the processed data, which is written by the back-end Invisible Map Generation code and uploaded to Firebase. The processed data is read by Invisible Map so that the user can interact with the finished map.

Raw Map Data

Note as of 6/17/2021: This structure is only implemented in the floor-detection branch of the InvisibleMap repository, and can only be read by the code in the floor-detection branch of invisible-map-generation. Older maps will not work with the floor-detection back-end code and newer maps will not work with the current master branch back-end code.

Note as of 8/3/2021: The code that collects plane data has been commented out due to memory issues during map creation. Maps located in the Firebase invisible-map-sandbox database have empty structures where planes are involved in the data. The commented out code that updates plane data is in ARView.swift in InvisibleMapCreator2 (lines 81-88).

The overall structure of the raw data is a dictionary with 5 keys: map_id, pose_data, tag_data, location_data (the waypoints or points of interest recorded during map creation), and plane_data. Each of these entries with the exception of map_id is another dictionary. The full structure of the file is detailed below in a nested bulleted list.

• "map_id": String, the unique ID for the map
• "pose_data": Dict, the odometry data taken at each frame 10 times per second
• "timestamp": Float
• "id": Int
• "pose": 16 element list (4x4 Float matrix), the odometry pose in world frame
• "planes": [Dict] (List of Dicts), each new plane that was detected when this odom frame was recorded
• "id": Int, id of the plane anchor
• "transform": 16 element list (4x4 Float matrix), the transformation matrix from the odometry to the plane anchor position.
• "tag_data": Dict
• "timestamp": Float
• "tag_id": Int
• "pose_id": Int, id of the odometry frame when this tag was detected
• "tag_pose": 16 element list (4x4 Float matrix), the tag pose in camera frame (local to phone)
• "tag_corners_pixel_coordinates": 8 element List, coordinates for each tag corner in pixels
• "camera_intrinsics": 4 element List, the internal data unique to the camera used to record the map. First two elements are focal length in the x and y directions, third and fourth elements are the pixel coordinates of the center.
• "tag_position_variance": 3 element List, variability in tag position
• "tag_orientation_variance": 4 element List, variability in tag orientation
• "joint_covar": 49 element List (7x7 matrix), covariance
• "location_data": Dict
• "timestamp": Float
• "name": String
• "pose_id": Int, id of the odometry frame when this location was saved
• "transform": 16 element list (4x4 Float matrix), the waypoint pose in world frame
• "plane_data": Dict
• "id": Int
• "pose": 16 element list (4x4 Float matrix), the pose of the plane anchor in world frame
• "boundaries": List of 3 element Lists, the coordinates of the corners of the plane boundary

Processed Map Data

The processed data JSON stores the optimized positions of the odometry, tags and the waypoints. The full structure is detailed below.

• "odometry_vertices": [Dict] (List of Dicts)
• "pose_id": Int
• "adjchi2": Float
• "viztags": Float
• "translation": Dict
• "x": Float
• "y": Float
• "z": Float
• "rotation": Dict
• "x": Float
• "y": Float
• "z": Float
• "w": Float
• "tag_vertices": [Dict]
• "id": Int
• "translation": Dict
• "x": Float
• "y": Float
• "z": Float
• "rotation": Dict
• "x": Float
• "y": Float
• "z": Float
• "w": Float
• "waypoints_vertices": [Dict]
• "id": String, name of the waypoint
• "translation": Dict
• "x": Float
• "y": Float
• "z": Float
• "rotation": Dict
• "x": Float
• "y": Float
• "z": Float
• "w": Float