Some recent research studies have addressed the attention assessment of participants during online meetings. Advances in artificial intelligence and the isolation period due to the COVID-19 pandemic have favored such research. However, measuring participants’ attention in fully presential meetings, particularly of students in large classrooms, is challenging and has yet to be undertaken. Another demanding scenario for determining the students’ attention is when professors start a course with a new group of students. Artificial intelligence algorithms can be trained with experience acquired from professors to help with that issue. Such algorithms with an appropriate camera distribution can assess the attention of all students in a classroom, providing the professor with a dashboard with information about the students’ attention. As a result, professors can adopt new teaching strategies according to the group’s level of engagement. In this work, we propose a computer system to provide real-time information about students’ attention in a classroom during a lecture. This system performs three computer vision tasks: pose classification, activity classification, and emotion classification from faces. Such algorithms implement deep neural networks using existing libraries (YOLO and EfficientNet), which provide fast object detection and image classification as required for real-time attention estimation. Our proposal focuses on attention estimation as one of the components of more complex concepts like student engagement.