ST-HAR: STretchable-only Human Activity Recognition

The ST-HAR dataset introduces a novel approach to human activity recognition (HAR) using only a stretchable capacitive sensor, the Bando C-Stretch, without relying on inertial or optical systems. This dataset offers measurements of human leg motion during walking and running, captured solely through elastomeric deformation. The C-Stretch sensor is based on a flexible elastomeric structure that accurately responds to elongation and surface strain. Its capacitance changes linearly with deformation, making it ideal for continuous biomechanical monitoring. The sensor is mounted on the knee joint of five healthy subjects while they perform treadmill-based walking and running at variable speeds ranging from 1.5 km/h to 14 km/h. Key features of the dataset: Single-sensor HAR: Data exclusively from a stretchable strain sensor. High fidelity: Includes both nominal activity data and noisy samples (in a dedicated others folder) to support research on robustness and signal compensation. Detailed user metadata: Each subject is characterized by anthropometric and demographic data (age, height, weight, percentiles, sensor placement). Rich annotation: Time-stamped activity transitions are marked and each trial is clearly labeled by speed and repetition. Whether you're developing machine learning models for wearable sensors, testing signal processing techniques, or exploring new directions in soft electronics, ST-HAR provides a focused dataset for advancing sensor-driven activity recognition. 📁 Includes: .mat files per subject and speed | 🧍‍♂️ 5 Subjects | ⚙️ Speeds: 1.5–14 km/h | 📚 Sensor datasheet & user manual included