Classification of motor activities through derivative dynamic time warping applied on accelerometer data

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Signal Processing

Background:

• Tele-monitoring systems increasingly focus on physical activity recognition for elderly and disabled populations.
• Accelerometers are commonly used for recognizing activities of daily living (ADLs).

Purpose of the Study:

• To propose a novel algorithm for activity recognition that accounts for movement speed variability.
• To compare the effectiveness of Dynamic Time Warping (DTW) and Derivative Dynamic Time Warping (DDTW) for activity recognition.

Main Methods:

• Developed a new algorithm using dynamic programming for activity recognition.
• Implemented and compared two template matching techniques: DTW and DDTW.
• Utilized a biaxial accelerometer placed on the shin to collect data for gait, climbing, and descending stairs.

Main Results:

• The DDTW approach achieved an average recognition score of 95% using a single sensor channel.
• DTW achieved an average recognition score of approximately 85%.
• Both DTW and DDTW outperformed classical Linear Time Warping (LTW) in classification rates.

Conclusions:

• DDTW offers a highly accurate method for recognizing daily physical activities, particularly gait and stair negotiation.
• The proposed algorithm's ability to handle movement speed variability enhances its utility in tele-monitoring applications.
• Accelerometer-based activity recognition, especially with advanced algorithms like DDTW, shows significant promise for remote health monitoring.