Dataset for "Understanding Freehand Cursorless Pointing Variability and Its Impact on Selection Performance"

Data collection method:

For the complete methodology used in the data collection, please refer to the paper. We used a repeated measures within-subject design with Pointing Style (accurate or casual) and Focus (focused or distracted) as independent variables (IV). We asked participants to "point as accurately and precisely as possible" for the accurate pointing condition. In contrast, in the casual pointing condition participants were instructed to "point as casually and relaxed as possible...". The distracted condition involved participants completing a Stroop effect test while pointing to targets, while the focused condition had pointing as the primary and only task.

Technical details and requirements:

Please refer to the README file for instructions to visualise the recorded pointing gestures or use any of the other provided scripts. Please refer to the README file for an explanation of the dataset structure and fields within specific files. For the technical details of the study setup and data collection, please refer to the paper. In summary, we utilised a set of 135 targets, which were grouped into 15 clusters of 9 (3×3), arranged into a 3 × 5 (rows × columns) array, with each target within a cluster spaced 8.4cm apart. The middle row was located 1.4 m from the floor and 2 m away from the participant, and the top and bottom rows were pitched ±25° from the middle row. Each column was yawed ±35° relative to the adjacent column. We used 12 Arqus infrared (IR) tracking cameras, and 10 Miqus cameras capturing RGB images at 1080p (4:3), with recording and marker tracking managed by QTM. The cameras provided coverage of a 4 m wide × 3 m deep × 2.5 m tall volume, within which the participant would be placed 2 m from the shorter edges, and ∼1 m from the long edge. The system was calibrated at the start of each day, with an average residual of 0.732 mm and standard deviation of 0.174 mm. We used 28 6.5 mm IR reflective markers to track all fingers on both hands; two for each finger and four to capture the palm and wrist. To aid in the tracking of the hands, we employed QTM’s AIM models and skeleton-assisted labelling. All sensing apparatus sampled data at 100Hz.

Additional information:

Motion capture data is grouped into zips for each participant, where each participant has been given a unique ID (GUUID). Each zip contains files representing each captured trial for the given participant, where the name of the file details the condition and trial number, e.g. ACCURATE_FOCUSED_3.csv, where 'ACCURATE_FOCUSED' is the condition and '3' is the trial number within that condition. Metadata for the trials is located within the GestureAnnotations.json. Additionally, and encoding of each valid gesture can be located within the encoded_gestures.csv Please refer to the dataset README file for an explanation of the dataset structure and fields within specific files.

Methodology link: