Reliability of smart textile for biometric recognition using electrocardiogram signal

Abstract

The reliability of smart textile garments for electrocardiogram (ECG) based biometric recognition presents a promising yet underexplored avenue in the development of wearable authentication technologies. While ECG signals offer unique physiological characteristics ideal for biometric verification, real-world deployment poses challenges related to signal variability, consistency, and classifier performance. This research directly addresses these issues by evaluating the reliability of two commercially available smart textile shirts of HeartIN Fit and Hexoskin ProShirt for ECG based biometric verification under practical and real-life conditions. Moreover, to thoroughly assess the approach robustness, the study was conducted through three structured experiments involving a total of 22 participants. The first experiment focused on distinctiveness, which evaluated how well the approach could identify individuals under varying physiological conditions such as walking, sitting, standing, and lying down. The second examined permanence, testing whether biometric performance remained stable across different time intervals, which is critical for long-term usability. The third experiment investigated system performance by benchmarking classification accuracy using a wide spectrum of 29 machine learning algorithms, including Quadratic Support Vector Machine (QSVM), Neural Networks (NN), and k-Nearest Neighbour (kNN). In order to ensure signal clarity and consistency, a low-pass Butterworth filter with a 30 Hz cut-off was applied to remove high-frequency noise while preserving key biometric features. Meanwhile, the HeartIN’s 512 Hz ECG signals were resampled to 256 Hz to align with Hexoskin’s sampling rate, enabling a normalised and fair comparative analysis. Subsequently, feature extraction stage captured the unique ECG morphology by detecting R-peaks and segmenting ten data points on each side, a process that successfully enhanced the reliability of biometric classification. Interestingly, the results demonstrated high reliability across all experiments, with the best performing classifier being QSVM, which achieved 98.81% accuracy with a false rejection rate (FRR) of 16.21% and a false acceptance rate (FAR) of 0.50% under different physiological conditions. When tested across time variability, it accomplished 99.20% accuracy with a FRR of 9.68% and a FAR of 0.27%. In addition, across the comparison with 29 classifiers, it attained 97.4% accuracy with a FRR of 2.60%, further confirming its robustness and reliability in diverse scenarios. Thus, these findings also provide strong evidence that ECG signals acquired through smart textile garments can deliver highly reliable biometric performance in dynamic and real-world environments. Therefore, through combining commercial wearable technology with broad machine learning evaluation, this study advances ECG biometrics and positions smart textiles as a secure, adaptable, and scalable platform for next-generation biometric verification.