Predicting Human Motion Signals Using Modern Deep Learning Techniques and Smartphone Sensors.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Wearable sensors and smartphone technology are increasingly used in healthcare applications.
• Human activity and motion recognition often involve time-series sensor data, typically pre-processed into images.
• Converting time-series data to images, such as recurrence plots, is a common pre-processing step.

Purpose of the Study:

• To predict human motion signals obtained from wearable sensors for healthcare applications.
• To evaluate the effectiveness of the Fourier neural operator for motion signal prediction.
• To compare the performance of the Fourier neural operator against Convolutional Neural Networks (CNNs) for this task.

Main Methods:

• Human motion signals from wearable sensors were converted into image formats using the recurrence plot method.
• A deep learning model, the Fourier neural operator, was utilized for predicting subsequent motion signals.
• The Fourier neural operator's performance was compared to a standard Convolutional Neural Network (CNN) model.

Main Results:

• The Fourier neural operator demonstrated superior performance compared to the CNN model in predicting motion signals.
• The proposed method showed potential for quicker detection of accidental falls through predicted motion signals.
• Recurrence plot images served as effective input for the deep learning models.

Conclusions:

• The Fourier neural operator is a promising deep learning model for analyzing time-series sensor data in healthcare.
• Converting motion signals to recurrence plots enables effective deep learning analysis for human motion prediction.
• This approach offers a pathway for faster and more accurate detection of accidental falls, enhancing patient safety.