Effective ECG fingertip sensor based biometric identification

Abstract

After 9/11, security and safety become one of the main concerns of governments around the world. Automatic accurate individual identification and authentication systems are becoming more critical in day-to-day activities like money transactions, access control, travel, medical services, and numerous others. The most prominent individual identification methods are ID cards, passwords, fingerprint, tokens, and signatures. Despite the large-scale deployment, these methods are vulnerable to identity falsification. The electrocardiogram (ECG) signal is very robust against identity forgery. However, many recent ECG systems demand longer time for recognition, which makes it hard to deploy an ECG based biometric system as a commercial product. This thesis studies a fast and effective ECG fingertip identification system in real time. The objective of the study is to reduce identification time. It is implemented in two steps, first is feature extraction, 3 features in a heartbeat are identified, they are simple but prominent features with discriminate characters. Second is segmentation where signals are sliced into 5 heartbeats to reduce the acquisition time. Then, 5 classification algorithms used to achieve up to 96% accuracy. A popular deep learning algorithm is also used for classification purpose and yields 94.12% accuracy. Through experiments, it is concluded this fingertip ECG recognition system can be used as an identifier for a small population.