Enhancing Smart Campus Monitoring: YOLOv8-Based Activity Recognition in Ethiopian Higher Education

The rapid expansion and application of computer vision which demands the installation of surveillance cameras for sophisticated smart classes. This paper describes a real-time activity recognition framework based on deep learning to enhance monitoring and educational quality in higher education. The proposed paper applied using YOLOv8 based model, the state-of-the-art in action recognition, to detect abnormal activities and unauthorized intrusions in higher education. Nevertheless, the implementation of such systems in resource-limited environment, such as Ethiopian higher education institutions, poses unique challenges related to smart class resources and data scarcity. This study focuses in Ethiopian higher education by introducing EthioCAD, a novel dataset to classify academic activities recognition across smart classroom teaching learning process. The dataset was constructed by addressing challenges in gathering real-world scenarios, employing the methodology of AI driven RoboFlow framework for frame extraction, annotation, and dataset organization. The novel EthioCAD dataset contains 4,224 KB of extracted frames and 33,485 MB of images. Where the EthioCAD dataset, available at github.com, serves as a valuable resource for advancing research in academic activity. To train dataset and evaluate its performance, we implemented the YOLOv8 model in high performing power of GPU in Google Colab resource where the accuracy reaching 90.2% and 78.6% on mAP50 and mAP50-95, respectively while these results maintaining high accuracy, ensuring proactive campus safety. Correspondingly results highlight the effectiveness of YOLOv8 for ensuring of safety and quality in teaching learning activity with auto-detected threats and real-time alerts which also a solid foundation for future studies in smart class based recognition.