Template-Based Recognition of Human Locomotion in IMU Sensor Data Using Dynamic Time Warping

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Human Activity Recognition

Background:

• Physical activity (PA) is crucial for health, necessitating accurate methods for its measurement.
• Sensor-based human activity recognition (HAR) offers a promising avenue for quantifying PA levels.
• Differentiating similar locomotion activities like walking and stair climbing remains a challenge in HAR.

Purpose of the Study:

• To develop and validate a two-layer locomotion recognition method using inertial sensor data.
• To accurately differentiate between walking, ascending stairs, and descending stairs.
• To assess the performance of the proposed method using a video-validated dataset.

Main Methods:

• Utilized dynamic time warping (DTW) on inertial sensor data.
• Developed a two-layer recognition algorithm.
• Tested the algorithm on the ADAPT dataset, recorded at the L5 position during free-living conditions.

Main Results:

• Walking was recognized with 88% sensitivity and 89% specificity.
• Stair climbing exhibited higher specificity but lower sensitivity compared to walking.
• Misclassifications primarily involved mistaking stair climbing for walking (0.2-5% unassigned).

Conclusions:

• The proposed two-layer method demonstrates a promising approach for recognizing and differentiating human locomotion.
• Accurate HAR of daily activities can be achieved using inertial sensor data.
• Further refinement may improve sensitivity for distinguishing between stair climbing and walking.