Ultraclear/USG_6dof_probe_measurements

6DoF Ultrasound Tracking Dataset

This dataset contains synchronous recordings from video tracking of an ultrasound probe, performed using a Dobot robotic arm with precise and fully repeatable trajectories. The dataset is intended for verifying the precision, robustness to motion blur, and calculating the drift error (RMSE) in a six degrees of freedom (6DoF) space.

Nomenclature and Folder Structure

Each folder in the dataset defines a single experiment ("Measurement") and is named according to a unified parametric schema, which can be parsed on the fly by the evaluation system.

Name format: <movement_name>_<repetitions>x_<robot_speed>sp__<evaluated_axes>

Practical example: square_4x_5sp__x,y What each part means:

1. movement_name (here: square) – The type of geometric path programmed on the Dobot robotic arm. Included paths are square, triangle, horizontal, vertical, cross, and roration (rotation).
2. repetitions (here: 4x) – Specifies how many full loops of the given movement the robot performed in one continuous sequence. More repetitions test the accumulated drift of the tracker over time.
3. robot_speed (here: 5sp) – The defined movement speed percentage used in the Dobot software panel (e.g., 5sp, 8sp). This helps test the tracking algorithm's robustness against camera motion blur.
4. evaluated_axes (here: __x,y) – The parameters following the double underscore (__) specify which native spatial axes or rotational angles are crucial for a given recording and evaluated in the final Euclidean distance. These are mostly the dominant active axes in the trajectory for the Ground Truth (such as x, y, pitch, etc.). Undefined dimensions are logically ignored / zeroed out.

Note on Evaluated Axes: Throughout this dataset, the analyzed translations are strictly limited to the X and Y axes, as the Dobot robotic arm has mechanical limitations preventing complex, dynamic movements along the Z unconstrained axis. Similarly, for rotational tracking, only the Pitch rotation is evaluated, which is also a direct result of the Dobot arm's hardware constraints.

Test Stand Setup and Dimensions

The experimental setup was precisely configured to ensure consistent, controlled motion and reliable tracking evaluation:

• Moving Camera (probe_camera): The first camera was rigidly attached to the ultrasound probe and was moving directly across the surface, simulating a typical scan. The distance from this camera's lens to the tracking surface below was strictly maintained at 7 cm.
• Stationary Reference Camera (ground_truth): The second camera acted as the objective reference. It was mounted completely stationary on a tripod, observing the moving probe from outside the robot's workspace. The distance from this reference camera to the ArUco marker on the robotic arm was precisely 41 cm.

Internal Structure (Measurement Contents)

Inside each measurement folder, there are two fixed subfolders containing recorded videos (named video.mp4 or similar) matching the setup described above:

• probe_camera/ – Contains the First-Person View (FPV) video recording from the moving camera mounted on the probe (7 cm above the surface). This video is intended to be processed by any custom tracking estimator. Note that algorithm tracking results are purposely not included in this public dataset.
• ground_truth/ – Contains the distant perspective video recording from the stationary reference camera (41 cm away) tracking the ArUco marker. This acts as an objective evaluation base (Ground Truth). In addition to the raw video, this folder provides the pre-computed reference coordinate file position.csv (which is generated from the offline analysis of the underlying video recording) for repeatability.

Ground Truth Data Structure

The position.csv reference files contain the tracking data collected securely from the stationary ArUco marker. The CSV structure is standardized across all measurements and consists of 7 columns:

• Frame – The consecutive video frame index (0-indexed).
• X, Y, Z – The calculated relative translation components (in millimeters) comparing the probe's current position to its calibrated starting point (0, 0, 0).
• Roll, Pitch, Yaw – The calculated 3D Euler angles (in degrees) representing the probe's rotation relative to its initially registered spatial orientation.

Note: Due to the parametric axis filtering mentioned in the nomenclature section, any unspecified tracking dimensions in a given measurement folder are intentionally forced to 0.000 to prevent unintended skewing of metrics during algorithmic evaluation.