Impact of Sensor Misplacement on Dynamic Time Warping Based Human Activity Recognition using Wearable Computers.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Human-Computer Interaction

Background:

• Daily living activity monitoring is crucial for early disease detection and enhancing elderly quality of life.
• Wireless wearable inertial sensor networks enable observation of daily human motions for activity recognition.
• Dynamic Time Warping (DTW) is a robust signal processing method for time-series pattern matching, but sensitive to sensor placement.

Purpose of the Study:

• To investigate the performance of DTW for human daily activity recognition, specifically the 'sit to stand' movement, under sensor misplacement.
• To assess the impact of varying sensor locations and orientations on DTW classification accuracy.
• To determine the limits of DTW's tolerance to sensor misplacement in activity recognition.

Main Methods:

• Utilized DTW distance as a feature for real-time human activity detection.
• Employed a marker-based optical motion capture system to generate simulated inertial sensor data.
• Created diverse sensor configurations with varied locations and orientations to simulate real-world misplacements.

Main Results:

• DTW's accuracy in activity recognition is significantly affected by sensor misplacement and changes in orientation.
• Simulated data generation using motion capture provided a practical method to study a wide range of sensor configurations.
• Identified critical sensor location variations that degrade DTW performance.

Conclusions:

• DTW, while flexible in time and speed, requires precise sensor placement for reliable human activity recognition.
• The study highlights the need for robust methods or sensor fusion to mitigate the effects of sensor misplacement in wearable systems.
• Findings inform the design of more resilient activity recognition systems for healthcare applications.