Real-Time Full-Body Gesture IMU Recordings from 10 Participants

413 isolated-gesture recordings and 35 continuous-sequence recordings of seven full-body gestures performed by 10 participants. The data was recorded in July 2019 by Connor Sachs under the supervision of Philipp Niklas Müller. Each recording uses four IMU sensors placed on the limbs, configured with a 50 Hz sampling rate.

Use Cases

• Train isolated gesture classifiers based on the seven defined gesture classes.
• Develop sequence models for continuous gesture recognition based on the sample-wise labeled sequences.
• Benchmark model performance across different sensor positions mentioned in the description.
• Study human motion variability based on the participant demographics provided.

Strengths

• Includes 413 isolated recordings and 35 continuous sequences, providing data for both classification and sequence modeling tasks.
• Contains metadata for 9 participants, including age range (20-29 years) and height range (155-200 cm).
• Specifies sensor configuration details, including accelerometer range (+/-2 g), gyroscope range (+/-250 deg/s), and sampling rate (50 Hz).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is relatively small at 7.4 MB, which may limit the complexity of models that can be trained.
• Participant demographics are incomplete for one participant, and the sample size of 10 participants is modest.

Provenance

Source

figshare

Collection Method

Recorded using four Silicon Labs Thunderboard Sense 2 boards per participant.

Time Range

July 2019

Freshness

Last updated 2026-05-17 14:24:04; freshness should be verified.