Sport-Related Human Activity Detection and Recognition Using a Smartwatch

Area of Science:

• Sports Science
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Sport-related activity monitoring is crucial for health and performance analysis.
• Current methods often use sliding windows for human activity recognition, which struggle with complex or non-periodic movements.
• Accurate detection of meaningful motion state durations is needed for better monitoring.

Purpose of the Study:

• To develop advanced methods for sport-related human activity detection and recognition.
• To address limitations of the sliding window approach in monitoring activities with complex motion states.
• To accurately determine the duration of meaningful motion states in both non-periodic and weakly periodic activities.

Main Methods:

• Proposed an interval-based detection and recognition method for non-periodic activities, generating candidate intervals to determine motion state durations.
• Developed a classification-based periodic matching method for weakly periodic activities, utilizing periodic matching for motion state segmentation.
• Focused on improving human activity recognition accuracy in sports monitoring.

Main Results:

• The proposed interval-based method accurately determines the duration of target motion states for non-periodic activities.
• The classification-based periodic matching method effectively segments motion states for weakly periodic activities.
• Both proposed methods demonstrated superior performance compared to the conventional sliding window approach in experimental results.

Conclusions:

• The developed interval-based and periodic matching methods offer significant improvements for sport-related activity monitoring.
• These novel approaches enhance the accuracy of human activity detection and recognition, particularly for challenging activity types.
• The findings suggest a more effective way to analyze sports activities, benefiting health and performance insights.