Low-Cost and Device-Free Human Activity Recognition Based on Hierarchical Learning Model.

Area of Science:

• Computer Science
• Artificial Intelligence
• Ubiquitous Computing

Background:

• Human activity recognition (HAR) is crucial for smart home human-computer interaction.
• Recognizing diverse and similar human actions remains a significant challenge.
• Existing methods often require costly sensors or complex setups.

Purpose of the Study:

• To propose a novel, low-cost, device-free HAR methodology using WiFi sensing.
• To achieve high accuracy in recognizing human activities, even similar ones.
• To enhance the efficiency of HAR systems in smart environments.

Main Methods:

• A hierarchical deep learning framework with two-level perception modules was developed.
• Received Signal Strength Indicator (RSSI) data was collected using ESP8266 WiFi sensors.
• A coarse-level Support Vector Machine (SVM) and a fine-level Gated Recurrent Unit (GRU) model were employed.

Main Results:

• The proposed method achieved high recognition accuracies of 96.45% and 94.59% for six activities across two environments.
• The hierarchical approach effectively discriminated between similar human activities.
• Performance surpassed traditional pattern-based methods.

Conclusions:

• The developed hierarchical learning method offers a cost-effective, sensor-based HAR framework.
• It significantly enhances recognition accuracy and modeling efficiency for smart homes.
• This approach provides a robust solution for challenging HAR tasks.