Footsteps detection and identification based on distributed optical fiber sensor and double-YOLO model

Area of Science:

• Sensor Technology
• Artificial Intelligence
• Biometrics

Background:

• Traditional footstep detection methods have limitations in range and accuracy.
• Gait analysis offers a unique biometric for identification.
• Distributed optical fiber sensing provides a large-scale sensing capability.

Purpose of the Study:

• To develop an accurate and efficient footstep detection and recognition system.
• To leverage distributed optical fiber sensing for extended range detection.
• To enhance gait-based person identification using advanced AI models.

Main Methods:

• Utilized phase-sensitive optical time-domain reflectometry (Φ-OTDR) for footstep sound detection.
• Employed a double-YOLO (You Only Look Once) model for footstep localization and recognition.
• Analyzed gait parameters like stride and step frequency for walker identification.

Main Results:

• Successfully detected, located, and identified footsteps from three individuals in field experiments.
• Achieved an overall identification accuracy of 86.0%.
• Demonstrated a 12.6% accuracy improvement over the single-YOLO method.

Conclusions:

• The proposed Φ-OTDR and double-YOLO method offers a robust solution for footstep detection and recognition.
• This technology has potential applications in gait-based clinical diagnosis and person identification.
• The system's enhanced accuracy and range surpass traditional methods.