neuromorphic-polito/siddha

SIDDHA

Smart Inertial Device Data from Human Activities

SIDDHA is a curated dataset derived from smartphone and smartwatch inertial measurement units (IMUs). It contains uniformly sampled accelerometer and gyroscope data at 20 Hz, for 51 subjects performing 18 activities across two devices:

• —Phone IMU data: accelerometer + gyroscope (x, y, z axes)
• —Watch IMU data: accelerometer + gyroscope (x, y, z axes)

All samples are timestamped, aligned, and stored in multiple formats:

• —Parquet (multidevice format)
• —Binary structured dictionaries
• —Text (.txt) WISDM-like format
• —Spike-based event representations (11 encoding methods)

This dataset is suitable for:

• —Human Activity Recognition (HAR)
• —Time-series classification
• —Sensor fusion and wearable ML
• —Neuromorphic computing and SNN research

List of Activities

The dataset includes 18 human activities, grouped into three categories.

Non–Hand-Oriented Activities

• —A — Walking
• —B — Jogging
• —C — Stairs
• —D — Sitting
• —E — Standing
• —M - Kicking (Soccer)

General Hand-Oriented Activities

• —F — Typing
• —G — Brushing Teeth
• —O — Playing Catch (Tennis)
• —P — Dribbling (Basketball)
• —Q — Writing
• —R — Clapping
• —S — Folding Clothes

Eating Hand-Oriented Activities

• —H — Eating Soup
• —I — Eating Chips
• —J — Eating Pasta
• —K — Drinking from Cup
• —L — Eating Sandwich

Devices Used for Recording

SIDDHA contains synchronized inertial data from two devices, each equipped with a full IMU:

Smartphone (Google Nexus 5/5X or Samsung Galaxy S5)

• —Worn in the front trouser pocket
• —Records:
• —Accelerometer (x, y, z)
• —Gyroscope (x, y, z)

Smartwatch (LG G Watch)

• —Worn on the dominant-hand wrist
• —Records:
• —Accelerometer (x, y, z)
• —Gyroscope (x, y, z)

Both devices contribute 6 inertial channels each, for a total of 12 channels per sample, uniformly sampled at 20 Hz.