IMU Dataset: Raw and Labeled Data for Human Activity Recognition

This dataset contains de-identified raw IMU recordings and labeled CSV files for human activity recognition.

The data were collected at a sampling rate of 100 Hz using three Xsens MTw Awinda IMUs attached to the left thigh, right thigh, and pelvis. The raw CSV files include orientation, angular velocity, linear acceleration, and magnetic-field measurements. For public release, the original absolute timestamps were replaced with relative time values at 0.01 s intervals, and the IMU hardware identifiers were replaced with the corresponding sensor attachment locations.

The labeled CSV files contain a relative time column, 18 input channels consisting of three-axis angular velocity and three-axis linear acceleration from the left thigh, right thigh, and pelvis, and an activity label column. The labeled files were generated by selecting activity-relevant intervals from the corresponding raw recordings, reordering the sensor channels, and assigning activity labels manually.

The recordings cover standing, sitting, walking, stairs cycle, turning left/right, sit-to-stand cycles, squatting, and stooping.