An Intelligent System for Person Identification Utilizing Ear Biometrics, Jetson Nano‐Based Embedded Device, and Advanced Deep Learnings

ABSTRACT Authentication is an essential component of any access mechanism or system. Biometrics is one of the most widely used technologies for personal identification systems. Biometric systems are built on individual human body traits including fingerprints, iris patterns, and facial identification. Among these characteristics, the ear is unique among all visible organs. This study is a pioneering effort to develop an embedded system for person identification using ear biometrics. We deploy two deep learning models, YOLOv5 and a self‐attention‐based Vision Transformer (ViT) on an embedded device (Jetson Nano) to provide ear biometrics‐based person detection. The primary dataset consists of 2050 images gathered from 10 individuals, split at random into 70% for training, 15% for validation, and 15% for testing. The fine‐tuned ViT reaches 99.86% training and 99.75% offline test accuracy, and a macro‐averaged AUC of 0.89 against 0.84 for YOLOv5. Deployed in real time on the Jetson Nano, the complete pipeline achieves 79.3% identification accuracy on live captures at about 16 frames per second. Furthermore, a retrained Haar cascade classifier is integrated as the automated ear‐cropping stage of the pipeline, reaching a positive predictive value of 99.805% for ear detection.