Faulty Exercises IMU Dataset: Correct and Faulty Bodyweight Exercise Recordings

181 raw IMU recordings capture 923 annotated repetitions of three bodyweight exercises performed by 16 participants. The data includes accelerometer and gyroscope readings from sensors mounted on the limbs, covering correct executions and common faults. Recorded in December 2020 by Felix Rauterberg under supervision of Philipp Niklas Müller.

Use Cases

• Train classifiers to detect correct vs. faulty exercise form based on IMU sensor data.
• Develop models for repetition counting and segmentation in bodyweight workouts.
• Analyze biomechanical differences between correct and faulty execution of squats, pushups, and bent-over rows.
• Benchmark sensor fusion algorithms for wearable motion tracking.

Strengths

• Includes 923 annotated exercise repetitions across 181 recordings, providing a substantial basis for supervised learning.
• Covers 13 distinct execution classes (correct plus common faults) for three exercise families, enabling fine-grained classification.
• Specifies detailed sensor configuration: accelerometer range +/-16 g, gyroscope range +/-2000 deg/s, and a 95 Hz sampling rate.
• Provides demographic details for 16 participants (12 male, 4 female, age range 13-56 years).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The participant pool of 16 individuals is relatively small, which may limit model generalizability.
• Data was recorded in a single month (December 2020), potentially lacking seasonal or long-term variation.

Provenance

Source

figshare

Collection Method

Recorded using four Silicon Labs Thunderboard Sense 2 boards per participant, mounted above both elbows and both knees.

Time Range

December 2020

Freshness

Last updated 2026-05-17 14:18:10; freshness should be verified.