A Novel Dataset for Gait Activity Recognition in Real-World Environments

Area of Science:

• Biomechanics
• Sensor Technology
• Machine Learning

Background:

• Falls are a major global health concern, with gait analysis crucial for fall risk identification.
• Wearable sensors and deep learning offer potential for remote gait analysis, enhancing data quality and automation.
• Accurate Human Activity Recognition (HAR) and terrain classification are essential for real-world gait analysis.

Purpose of the Study:

• To address the lack of suitable datasets for terrain classification in gait analysis.
• To present the Context-Aware Human Activity Recognition (CAHAR) dataset, the first activity- and terrain-labeled dataset.
• To facilitate the development of advanced classification models for remote gait analysis.

Main Methods:

• Collected data from 20 healthy participants using Inertial Measurement Units (IMUs), Force-Sensing Resistor (FSR) insoles, color sensors, and LiDAR.
• Captured data across a diverse range of indoor and outdoor terrains.
• Developed the CAHAR dataset with activity and terrain labels.

Main Results:

• The CAHAR dataset is the first of its kind, integrating both human activity and environmental terrain information.
• The dataset supports the development of models capable of simultaneous HAR and terrain identification.
• Enables progress towards more sophisticated remote gait analysis using wearable sensors.

Conclusions:

• The CAHAR dataset is a significant resource for advancing research in wearable sensor technology for gait analysis.
• It provides a foundation for creating more accurate fall risk prediction models.
• Facilitates the development of intelligent systems for real-world gait monitoring and fall prevention.