Multimodal Signals Subject Authentication System

The ever increasing number of e-services provided to users necessitates the efficient and accurate verification of user identity. In this work, we present a user authentication model based on multimodal signals. More specifically, we investigate the extraction of statistical features using various segment lengths of photoplethysmogram (PPG), electrocardiogram (ECG) and capnogram (CO2) signals and generate a feature vector using different fusion approaches. In addition, adding to the feature vector additional features extracted from the 1st derivative of the corresponding signal is explored. Finally, a classifier is used for identity authentication. The proposed authentication system achieved an average authentication accuracy of 99.76% using a 1 second segment length and a random forest classifier.