On the Feasibility of Low-Cost Wearable Sensors for Multi-Modal Biometric Verification

Abstract

Biometric systems designed on wearable technology have substantial differences from traditional biometric systems. Due to their wearable nature, they generally capture noisier signals and can only be trained with signals belonging to the device user (biometric verification). In this article, we assess the feasibility of using low-cost wearable sensors—photoplethysmogram (PPG), electrocardiogram (ECG), accelerometer (ACC), and galvanic skin response (GSR)—for biometric verification. We present a prototype, built with low-cost wearable sensors, that was used to capture data from 25 subjects while seated (at resting state), walking, and seated (after a gentle stroll). We used this data to evaluate how the different combinations of signals affected the biometric verification process. Our results showed that the low-cost sensors currently being embedded in many fitness bands and smart-watches can be combined to enable biometric verification. We report and compare the results obtained by all tested configurations. Our best configuration, which uses ECG, PPG and GSR, obtained 0.99 area under the curve and 0.02 equal error rate with only 60 s of training data. We have made our dataset public so that our work can be compared with proposals developed by other researchers.