Recurrence Quantification Analysis for Human Activity Recognition.

Area of Science:

• * Human Activity Recognition (HAR)
• * Signal Processing
• * Machine Learning

Background:

• * Human Activity Recognition (HAR) is crucial for understanding and predicting behavior, with applications in health monitoring and lifestyle analysis.
• * Inertial Measurement Units (IMUs) are common sensors for HAR, and Recurrence Plot (RP) techniques are emerging for feature extraction.
• * While automated feature extraction from RPs exists, hand-crafted methods remain underexplored for HAR.

Purpose of the Study:

• * To evaluate the feasibility of using Recurrence Quantification Analysis (RQA), a hand-crafted feature extraction method from RPs, for human activity classification.
• * To compare the performance of RQA against machine-auto or end-to-end approaches in HAR.
• * To demonstrate RQA as a simpler and more efficient system for HAR applications.

Main Methods:

• * Utilizing Recurrence Quantification Analysis (RQA) to extract feature vectors from sensor data.
• * Applying a Linear Discriminant Analysis (LDA) classifier to classify human activities based on extracted RQA features.
• * Comparing RQA-based classification accuracy with existing literature, particularly machine-auto methods.

Main Results:

• * RQA combined with LDA achieved a classification accuracy of 95.08% for human activities.
• * This accuracy rate is competitive with, and among the top reported in, the existing literature.
• * RQA demonstrated a less complicated and more lean system compared to machine-auto or end-to-end approaches.

Conclusions:

• * Recurrence Quantification Analysis (RQA) is a feasible and highly effective method for classifying human activities.
• * RQA offers a simpler, more efficient, and interpretable alternative to complex automated feature extraction methods in HAR.
• * Further research into RQA for HAR applications is warranted due to its promising performance and system leanliness.