Embedded Iris Recognition System Based on YOLO

In recent years, respiratory infectious diseases have imposed limitations on facial and fingerprint recognition technologies. Consequently, iris recognition has garnered significant attention due to its non-contact operation, accuracy, and security. To address the computational complexity and slow speed of iris localization during recognition, this study proposes an object detection method based on an improved YOLOv7-tiny model. Enhancements to the Backbone, Neck, and loss function components of the YOLOv7-tiny model resulted in improved mean Average Precision (mAP), reduced parameter count, and decreased computational load. Additionally, to mitigate edge noise interference in texture feature extraction, we refined the Local Binary Pattern (LBP) algorithm by recalculating thresholds using a center-point-integrated adjacent pixel approach, thereby enhancing image clarity. The optimized iris recognition algorithm was deployed on a Jetson Xavier NX development board to establish an embedded iris recognition system. Experimental demonstrates that the system achieves robust iris matching performance, with notable improvements in recognition accuracy, efficiency, and operational effectiveness.