Intelligent Attendance System: Combining Fusion Setting with Robust Similarity Measure for Face Recognition

Facial recognition attendance and timekeeping systems are increasingly widely applied in classrooms or businesses. However, when applying the attendance system in real-life scenarios, there are still several challenges, such as the accuracy of predictions, effective data collection methods, and the recognition time on resource-constrained hardware. Most identification methods compare features at the instance level and cannot utilize the aggregated features of all faces with the same attendance. This paper presents an attendance system that utilizes face recognition technology. To enhance the accuracy of face recognition, we employ intelligent data collection techniques and propose improvements to face detection. A clustering algorithm is employed to select relevant facial data for subsequent feature comparison in the intelligent data collection process. By optimizing face detection accuracy, we combine results across multiple frames and introduce a novel approach to information retrieval by combining distances in centroid-based and instance-based settings. Logistic Regression is employed to determine the weights for the method above. We evaluate our combined approach on the VN-celeb dataset, demonstrating its superior accuracy compared to previous centroid-based and instance-based. Furthermore, we extend the application of our proposed attendance system to an intelligent attendance system, utilizing hardware components such as Raspberry Pi 4 and supplementary devices. This system is supported by a database infrastructure and a user-friendly website interface, allowing users to conveniently display attendance-related information.