Daily life event segmentation for lifestyle evaluation based on multi-sensor data recorded by a wearable device

Area of Science:

• Human-computer interaction
• Wearable computing
• Health informatics

Background:

• Evaluating lifestyle for health maintenance requires analyzing daily activities.
• Wearable devices like the eButton can continuously record extensive multimedia data.
• Automatic processing of this large dataset is crucial for practical application.

Purpose of the Study:

• To develop an automatic segmentation method for daily life events using multi-sensor data from the eButton wearable computer.
• To identify effective signal and visual features for accurate activity boundary detection.

Main Methods:

• A two-step summarization approach was developed for automatic dataset segmentation.
• The first step utilizes motion sensor data (gyroscope) to find candidate activity boundaries.
• The second step employs visual features (image HSV histogram) to refine and finalize activity boundaries.

Main Results:

• A combination of gyroscope standard deviation and image HSV histogram features yielded satisfactory results in automatic daily life event segmentation.
• The proposed two-step method effectively segments large datasets recorded by the eButton.

Conclusions:

• Simple signal and visual features are sufficient for accurate automatic daily life event segmentation.
• This method provides a viable approach for evaluating lifestyle and supporting health maintenance through wearable technology.