Sınavda öğrenci etkinliklerinin etiketlenmesi için derin öğrenme temelli özniteliklerin çıkarılması ve sınıflandırılması

Visual monitoring study findings have improved considerably as a result of research on video description and human activity detection. Exam cheating detection is a fundamental part of any level education program. This work focuses on students' activities labeling in the exam. The framework is developed for labeling students' activities into six different classes, back- watching, front-watching, side-watching, normal, showing gestures, and suspicious. A methodology for predicting cheating activities is proposed in this study. To extract features, feature descriptors such as local binary patterns and texture features are used. The entropy and ant colony optimization (ACO) based feature selection methods are utilized separately on the acquired feature subsets having qualities of both filter and wrapper-based approaches. The features are then combined to form a powerful features subset. Those selected features are trained on several different models. SVM-based and KNN classifiers are showed promising results on the datasets. The classification system accurately labels the student activities into abnormal and normal classifications using the exam activities detection dataset. The findings show that the proposed framework for activity recognition in exams is quite effective and accurate. 92% for SVM and 94% for KNN achieved on the dataset.