Abstarct: Abstract: After overcoming the COVID-19 pandemic, the increasing demand for electronic exams, and the growing use of AI-baxsed services like ChatGPT for answering questions in these exams, video surveillance has emerged as one of the primary methods for enhancing the quality of remote assessments. In such circumstances, reviewing many exam videos can be time-consuming and tedious for proctors. A solution to this problem is automatically summarizing exam videos, focusing on suspicious moments that may indicate overt cheating, to facilitate easier monitoring. This research proposes a novel approach utilizing deep neural networks to analyze the interactions of examinees with objects present in the scene, emphasizing video summarization and optimized processing for identifying moments of overt cheating. Overt cheating includes physical interactions of the examinee with objects in the scene, such as using a mobile phone, notes, or the presence of more than one person in the exam setting to gain an unfair advantage over other examinees. Previous studies baxsed on morphological methods primarily focused on scene changes without considering the impact of these changes on cheating events. Subsequent neural network-baxsed methods provided relatively acceptable scientific responses; however, the limitations of using neural networks include the need for powerful processors, the requirement to train models before use, and privacy concerns, all of which present challenges. The proposed method combines the use of deep neural networks with morphological functions to reduce processing load and eliminate dependence on various scenes and conditions of the examinee. This method consists of three main stages: preprocessing, extracting examinee activities, and identifying unauthorized objects in the scene used for cheating. In the preprocessing stage, by comparing structural similarities between consecutive frxames, the stationary frxames of the examinee are removed from the original video. In the second stage, using skeletal analysis, the movements of the examinee are identified, and another neural network extracts the objects present in the scene along with their positions. Finally, suspicious scenes of overt cheating are identified by evaluating the interactions between the movements of the examinee's body parts and the unauthorized objects in the scene. Key features of the proposed method include high processing speed, flexibility against background changes, and the ability to summarize videos without prior training. Experimental results on 120 exam videos show that this method reduces the video duration by up to 70% while preserving all scenes of cheating or suspected overt cheating.