MultiSenseGolf: A Multimodal Sensor-Based Dataset for Golf Swing Motion and Shot Performance Analysis

MultiSenseGolf is a multimodal dataset for golf swing motion and shot performance analysis collected using wearable and environmental sensors. It contains 1,557 swing records from 24 participants spanning Beginner, Intermediate, and Professional skill levels. The released modalities include full-body joint kinematic estimates from an inertial motion capture (IMC) system, bilateral foot pressure streams, first-person view (FPV) video, gaze coordinates, acceleration and angular velocity from a head-mounted inertial measurement unit (IMU), front-view RGB-D video, and shot outcome data recorded using an optical launch monitor. To support swing-level analyses, we conducted post-hoc data processing by annotating the start and end boundaries of each swing, correcting the observed timestamp offsets between recording systems using different device clocks, and storing the processed streams as swing-level records. Technical validation comprises four analyses: a mathematical sensitivity analysis of forward kinematics (FK)-derived joint positions under assumed orientation errors, a statistical analysis of carry distance distributions across participant attributes, a characterization of observed cross-device timestamp offsets, their within-session trends, and cross-modal timing variability, and a shot outcome prediction benchmark across five input conditions based on different modalities.