Smart System for Monitoring Student Behavior

Detecting and classifying student behaviors within the classroom accurately becomes more beneficial for analyzing students’ performance and improving teaching and learning process effectiveness. There are major challenges within classrooms, including high classroom density and various student behaviors. Furthermore, many current school management systems suffer from many issues regarding accurate verification of whether a student is present or not. Moreover, these systems do not provide a clear view of how the students are attentive and engaged with the instructor during lectures. To address these challenges, this paper proposes a smart student behavior detection system that analyzes the behavior of university students and their participation throughout the lectures and integrates them with their academic performance. This paper proposes a solution that integrates academic performance with real-time behavior analysis and facial recognition in a parallel way to achieve enhanced academic performance. First, student attendance is taken using a two-step verification technique where the student is verified upon entering the campus gate and then reconfirms his attendance through facial recognition as he enters the class. Then, advanced computer models track and analyze the behavior of students in the class to keep a record of how attentive they are and how often they participate in class. This paper proposes an improved YOLOv8-based model to detect student behavior in the classroom with high accuracy. The proposed model presented an improved YOLOv8-based model by enhancing the convolution techniques by importing the SPD-Conv module instead of the original convolution to enhance the model’s attention towards small and medium-sized targets. Also, the proposed system enhances the loss functions by using the Focal-SIoU loss function instead of the original C-IOU loss function to adjust the weights of positive and negative samples to penalize difficult-to-classify samples more heavily. The SCB-5 dataset was used in the experiments for training, validation, and testing. The proposed method uses three metrics to evaluate the system performance: mean average precision (mAP), precision, and recall. The proposed system results demonstrate remarkable precision, recall, and average precision compared to state-of-the-art methods.