A Deep Machine Learning Method for Concurrent and Interleaved Human Activity Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Human activity recognition is crucial for pervasive computing, ambient assistive living, robotics, and healthcare.
• Current methods struggle with recognizing complex activities like concurrent and interleaved actions.
• Advanced techniques are needed to overcome these limitations in human activity recognition.

Purpose of the Study:

• To propose a novel two-phase hybrid deep learning approach for recognizing complex human activities.
• To address the challenges of identifying concurrent and interleaved activities.
• To enhance the accuracy and robustness of human activity recognition systems.

Main Methods:

• A two-phase hybrid deep learning model combining Bi-directional Long-Short Term Memory (BiLSTM) and Skip-Chain Conditional Random Field (SCCRF).
• Phase 1: BiLSTM for recognizing concurrent activities.
• Phase 2: SCCRF for identifying interleaved activities, leveraging long-term dependencies.

Main Results:

• The proposed framework demonstrated superior performance compared to state-of-the-art methods.
• Achieved an average accuracy exceeding 93% on publicly available smart home datasets.
• The hybrid approach effectively handles both concurrent and interleaved human activities.

Conclusions:

• The proposed BiLSTM-SCCRF hybrid model offers a significant advancement in complex human activity recognition.
• This method provides a robust solution for applications in smart homes and healthcare monitoring.
• The high accuracy validates the effectiveness of the two-phase deep learning strategy for complex activity analysis.