Evaluación comparativa de algoritmos de reconocimiento facial para optimizar el control de asistencia de estudiantes de una universidad pública de Perú

Marco Rocca; Williams Muñoz-Robles; Wilder Leandro-Vadillo; Nilton Vicente-Guerra; Maruja Porras-Rojas · Investigación Valdizana (2026) · DOI: 10.33554/riv.19.4.2617

Resumen

This research comparatively evaluates four facial recognition algorithms—FaceNet-X, MiniArc, LBPH-CUDA, and Enhanced Eigenfaces—to determine which offers the best performance for optimizing student attendance control at a public university in Peru. Three hundred individual images were analyzed using a repeated measures experimental design, evaluating two dimensions: accuracy and speed. The results showed that the Deep Learning-based algorithms (FaceNet-X and MiniArc) achieved the highest levels of accuracy (0.98 and 0.97, respectively), significantly outperforming the classical methods. In terms of speed, LBPH-CUDA obtained the lowest times (≈5 ms), followed by MiniArc (≈28 ms), although with lower accuracy. Normality, Friedman, and ANOVA tests confirmed statistically significant differences between the algorithms in both dimensions. Overall, MiniArc presented the best balance between accuracy and speed, while FaceNet-X stood out as the most accurate. It is concluded that Deep Learning models are the most suitable for institutional implementation due to their stability, accuracy, and adaptability to real classroom conditions.

Instituciones: Universidad Nacional Daniel Alcides Carrión