Improving BLE-Based Passive Human Sensing with Deep Learning.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Passive Human Sensing (PHS) traditionally uses WiFi, facing challenges like power consumption and deployment costs.
• Bluetooth Low Energy (BLE) offers a promising alternative due to its Adaptive Frequency Hopping (AFH) mechanism.
• Existing PHS methods often require active participation or direct line-of-sight obstruction.

Purpose of the Study:

• To apply Deep Convolutional Neural Networks (DNN) for enhanced analysis and classification of BLE signal deformations for PHS.
• To demonstrate the efficacy of BLE-based PHS in detecting human presence in complex environments.
• To overcome the limitations associated with WiFi-based PHS.

Main Methods:

• Utilized commercial standard BLE devices for signal transmission and reception.
• Implemented a Deep Convolutional Neural Network (DNN) model for processing BLE signal variations.
• Tested the system in a large, articulated room, including scenarios without direct line-of-sight obstruction.

Main Results:

• The DNN-based approach reliably detected human occupants using a minimal number of BLE transmitters and receivers.
• Achieved superior performance compared to existing state-of-the-art techniques on the same experimental data.
• Successfully demonstrated PHS in non-line-of-sight conditions.

Conclusions:

• DNN-powered BLE PHS is a highly effective and efficient method for human presence detection.
• This approach offers a viable, low-cost alternative to WiFi-based PHS systems.
• The proposed method significantly advances the capabilities of passive sensing technologies.