Smartphone Motion Sensor-Based Complex Human Activity Identification Using Deep Stacked Autoencoder Algorithm for...

Area of Science:

• Human-Computer Interaction
• Wearable Technology
• Machine Learning

Background:

• Smartphone accelerometers are used for human motion analysis in health and activity recognition.
• Current methods struggle with varying device orientation, leading to performance degradation.
• Existing algorithms are often application-specific and fail to capture complex activity dynamics.

Purpose of the Study:

• To propose a deep learning approach for accurate complex human activity identification using smartphone accelerometer data.
• To develop orientation-invariant features to address sensor orientation inconsistencies.
• To improve the robustness and accuracy of human activity recognition frameworks.

Main Methods:

• Augmented accelerometer data with magnitude norm vector and rotation features (pitch, roll angles).
• Employed a deep stacked autoencoder (DSAE) for automatic feature extraction.
• Integrated orientation-invariant features with the DSAE deep learning algorithm.

Main Results:

• The proposed DSAE method achieved 97.13% accuracy in complex human activity identification.
• Demonstrated superior performance compared to conventional machine learning and deep belief network algorithms.
• Successfully recognized complex activity details using only smartphone accelerometer data.

Conclusions:

• The integration of deep learning and orientation-invariant features significantly enhances complex human activity identification.
• The proposed method offers a robust and accurate solution for smartphone-based activity recognition.
• This framework has the potential to improve various applications, including health monitoring and fall detection.