Deep Learning-Based Vibration Signal Personnel Positioning System

• Main Authors: Yang Yu, Marian Waltereit, Viktor Matkovic, Weiyan Hou, Torben Weis
• Format: Article
• Language: English
• Published: IEEE 2020-01-01
• Series: IEEE Access
• Subjects: Vibration signal; localization; pattern recognition; deep learning; privacy protection; robustness
• Online Access: https://ieeexplore.ieee.org/document/9292972/

In this work, we present a person localization system based on ground vibration caused by walking persons. The system is designed for production plants and large buildings to track the movement of workers. Position and movement in these settings are especially safety-relevant in emergencies. Our approach is privacy-preserving, because it requires neither video nor sound. Instead, piezo sensors on the floor measure vibrations, which are analyzed with machine learning to derive a person's position from the vibration signals. This way, our system can determine where a person is moving, but it is not straightforward to attach names to the detected persons. Due to the anisotropic characteristic of the ground vibration wave, classical analysis methods are not applicable. We show that a deep learning-based approach is feasible. Our experiments show that we can determine the position with an average F1 score of 0.95.