[Paper Review] Person Identification using Seismic Signals generated from Footfalls

Bodhibrata Mukhopadhyay, Sahil Anchal+1 more|arXiv (Cornell University)|Sep 24, 2018

Gait Recognition and Analysis3 references3 citations

TL;DR

This paper proposes a Fog computing-based person identification system using seismic signals from footfalls, leveraging a novel basis pursuit-based compression technique (DS8BP) that reduces 8 kHz signals by a factor of 108 while enhancing classification accuracy. The system achieves 90% and 95% identification accuracy using 5 and 10 footsteps per sample, respectively, on an 8-person dataset of 46,000 events, demonstrating high viability for privacy-preserving access control.

ABSTRACT

Footfall based biometric system is perhaps the only person identification technique which does not hinder the natural movement of an individual. This is a clear edge over all other biometric systems which require a formidable amount of human intervention and encroach upon an individual's privacy to some extent or the other. This paper presents a Fog computing architecture for implementing footfall based biometric system using widespread geographically distributed geophones (vibration sensor). Results were stored in an Internet of Things (IoT) cloud. We have tested our biometric system on an indigenous database (created by us) containing 46000 footfall events from 8 individuals and achieved an accuracy of 73%, 90% and 95% in case of 1, 5 and 10 footsteps per sample. We also proposed a basis pursuit based data compression technique DS8BP for wireless transmission of footfall events to the Fog. DS8BP compresses the original footfall events (sampled at 8 kHz) by a factor of 108 and also acts as a smoothing filter. These experimental results depict the high viability of our technique in the realm of person identification and access control systems.

Motivation & Objective

• To develop a privacy-preserving, non-intrusive person identification system using seismic signals from footfalls.
• To reduce bandwidth and energy consumption in wireless transmission of raw seismic signals through efficient compression.
• To implement a Fog computing architecture that shifts computation closer to the sensor for real-time processing.
• To evaluate the impact of number of footsteps per sample, sampling frequency, and classifier choice on identification accuracy.
• To design a secure, de-noising compression technique that preserves classification performance.

Proposed method

• Deployed a network of geophones (vibration sensors) to capture seismic signals from footfalls at 8 kHz sampling rate.
• Proposed DS8BP—a basis pursuit-based compression technique using a learned Gabor dictionary to represent footfall signals with few basis functions.
• Set compression thresholds (L_GC and H_GC) to eliminate signals with poor or excessive compression, ensuring signal integrity.
• Used SVM-RBF, SVM-Lin, and Logistic Regression as classifiers on compressed and decompressed signals to evaluate performance.
• Integrated the system with an IoT cloud for data storage and used XBee 868LR for wireless transmission at 80 kbps.
• Applied event detection and signal recovery using algorithms that reconstruct original signals from compressed measurements via sparse coding.

Experimental results

Research questions

• RQ1 What is the optimal number of footsteps per sample to maximize identification accuracy while minimizing computational load?
• RQ2 How does downsampling from 8 kHz to 500 Hz affect classification accuracy and feature extraction time?
• RQ3 Can a basis pursuit-based compression method (DS8BP) achieve high compression ratios while preserving signal fidelity for classification?
• RQ4 How does DS8BP compare to traditional decimation (DS16) in terms of compression ratio, signal recovery quality, and classifier accuracy?
• RQ5 What is the minimum number of training footsteps per individual required to achieve a target classification accuracy?

Key findings

• The system achieved 90% and 95% identification accuracy using 5 and 10 footsteps per sample, respectively, with SVM-RBF on the 46,000-event dataset.
• DS8BP compressed the original 8 kHz signal by a factor of 108.32, significantly reducing transmission bandwidth and energy.
• Classifiers achieved 2–4.3% higher accuracy when using DS8BP-compressed signals compared to DS16, due to its inherent de-noising and smoothing effect.
• Downsampling to 500 Hz (DS16) did not degrade classification accuracy but reduced feature extraction and event detection times.
• The average compressed footfall event length was 18.51 samples, with a mean compression factor of 13.54 per step.
• DS8BP enhanced system security, as the compressed signal cannot be decoded without knowledge of the Gabor dictionary (D).