<![CDATA[Manual attendance processes in higher education often face severe constraints regarding time inefficiency and vulnerability to data manipulation, specifically the prevalent issue of proxy attendance. Although Face Recognition technology has been widely adopted, most existing systems utilize a "once recognition" method, which fails to validate the student's presence throughout the entire lecture duration. This study aims to bridge this gap by developing an automatic desktop-based attendance system that integrates Face Recognition with a novel Adaptive Attendance Monitoring (AAM) approach. The proposed system utilizes a robust deep learning pipeline employing the Multi-Task Cascaded Convolutional Neural Network (MTCNN) for face detection and alignment, followed by FaceNet for generating 128-dimensional feature embeddings. To ensure real-time performance, the processing is accelerated by CUDA GPU technology on an NVIDIA RTX 4060 Ti. The system architecture follows a decoupled Client-Server model based on REST API, ensuring scalability and low-latency data transmission. The primary novelty of this research is the AAM algorithm, which continuously calculates the cumulative duration of a student's presence. A student is validated as "Present" only if they maintain visibility for at least 80% of the total session duration, effectively eliminating the "check-in and leave" loophole. Experimental results demonstrate that the system achieves a 100% recognition accuracy at an optimal distance of 1.0 meter under normal lighting conditions, with a processing latency consistently maintained under 100ms. These findings confirm that the proposed desktop-edge architecture significantly outperforms traditional mobile-based solutions in terms of stability, security, and continuous monitoring capabilities.]]>
