AR Data Format

AR Coordinate System

Coordinate Frame:

• X-axis: Right

• Y-axis: Up

• Z-axis: Forward

Pose : $[t_x, t_y, t_z, q_x, q_y, q_z, q_w]$

• Translation (first 3 values): Position in meters

  • $t_x, t_y, t_z$: X, Y, Z coordinates

• Rotation (last 4 values): Orientation as quaternion

  • $q_x, q_y, q_z$: Imaginary part

  • $q_w$: Real part

Difference between local_pose and global_pose:

• local_pose: Position relative to device starting point

• global_pose: Position in shared world coordinate (when multiple devices collaborate)

Note: When using a single device, local_pose and global_pose are identical

Velocity

Format: $[v_x, v_y, v_z]$

• Calculated as: $\vec{v} = \frac{\Delta \vec{p}}{\Delta t}$

• Unit: meters per second (m/s)

• In the same coordinate frame as pose

Camera Intrinsics

Format: $[f_x, 0, 0, 0, f_y, 0, c_x, c_y, 1]$

Represents 3×3 matrix:

$$K = \begin{bmatrix} f_x & 0 & c_x \\ 0 & f_y & c_y \\ 0 & 0 & 1 \end{bmatrix}$$

Where:

• $f_x, f_y$: Focal length in pixels

• $c_x, c_y$: Principal point (optical center) in pixels

Attention

Camera intrinsics scaling required

The camera intrinsics are originally computed for the original image size 1920 × 1440.

In the current app, the streamed color image size is 640 × 480.

If you use the intrinsics with the streamed images, you must scale the intrinsics to match the new resolution.

(You can follow the method below for implementation.)

from asmagic import ARDataSubscriber
import numpy as np

sub = ARDataSubscriber("192.168.1.100")

for data in sub:
    K = np.array(data.camera_intrinsics).reshape(3, 3, order='F')
    scale_x = 640 / 1920  # = 1/3
    scale_y = 480 / 1440  # = 1/3
    
    K[0, 0] *= scale_x  # fx
    K[1, 1] *= scale_y  # fy
    K[0, 2] *= scale_x  # cx
    K[1, 2] *= scale_y  # cy
    
    print(f"Scaled intrinsics:\n{K}")

Depth Image

• Format: 16-bit unsigned integer (uint16)

• Unit: $10^{-4}$ m (0.1 mm, scaled by 10000 from meters)

• Conversion to meters: $d_{meters} = \frac{d_{raw}}{10000}$

• Range: $0 \leq d_{raw} \leq 65535$ → $0$ to $6.5535$ m

• Access: Use data.depth or data.depth_array to get numpy array

Image Data Format

Color Image: Two formats available

• color_bytes: JPEG compressed bytes

• color_array: RGB numpy array (640×480×3)

• color: Shortcut for color_array

Depth Image: Numpy array only

• depth_array: uint16 numpy array (256×192)

• depth: Shortcut for depth_array