Human Identification using electrocardiogram ECG signal from wrist mounted sensors (P-0091) (MFN 4664)

Abstract

As the technology for falsification advances, security concerns augment. In today’s era of highly developed technological modernism, instead of using the conventional systems of authentication, biometric based systems are used. These systems provide better security solutions because of their reliability and user acceptance. Biometric systems work on the proposition that resemblance exists between two recordings of the features under study, obtained from same individual. The use of intrinsic biometric traits is trending as it guarantees high level of accuracy and great impossibility of being forged. There are strong substantiations which reveal that ECG (Electrocardiogram) has a good potential- with permanence and distinctiveness, justifying the fact that human’s heart electrical activity alone can be used as a biometric stricture for identity recognition. Since ECG is an intrinsic biological attribute; it provides high accuracy, is hard to falsify. It also confirms the presence and aliveness of the person. The organization of this thesis elucidates a fiducial-detection based framework of our developed prototype of an embedded system which uses relevant information from ECG signals to perform human identification. The system comprises of a wrist band; a wearable biometric identity device, that uses heart’s unique signature to authenticate and confirm the individuality of the person wearing it. The thesis is divided into four parts. The first part explains signal acquisition; capturing single channel ECG using an Analog Front End circuit and performing analysis for recognition using signal processing and machine learning methods. The second part states the preprocessing; the acquired ECG signal is passed through ADC and then transmitted to a Personal Computer via Bluetooth. Filters are applied to remove Power Line and other Noises from the signal and amplification is performed. In the third part, ECG features are extracted applying by Auto-Regressive Coefficients and distance between the peak methods and saved in a database. Finally, in the fourth part, Biometric operation is performed and similarity between the test subjects and database is calculated. The signals are classified using Linear Discernment Analysis. Distance between the peaks gave an accuracy of 80% whereas using AR coefficients, 99.77% accuracy is achieved.