Cheating Recognition in Examination Halls Based on Improved YOLOv8

With the advancement of artificial intelligence technology, smart proctoring has gradually supplanted traditional manual invigilation and becomes the dominant mode of examination supervision. However, existing technologies mostly rely on singular object detection algorithms or deep learning techniques, which are inadequate in addressing the complex and varied conditions of examination environments. In this paper, we design a multi-level intelligent recognition system for candidates' cheating behaviors, integrating an optimized YOLOv8 object detection method based on multilayer perceptron (MLP) with the ResNet deep learning framework. This system mines key frames from surveillance videos to precisely capture candidates' positional information and automatically tags those suspected of engaging in cheating activities. Our model's development relies on a custom-tailored dataset, the cheating and normal (CAN) dataset, which includes instances of academic misconduct alongside standard behavior for training purposes. The model's performance is then validated by assessing its effectiveness on real-life surveillance videos from examination halls. The resulting intelligent analysis model is capable of real-time, meticulous tracking and evaluation of every movement of each candidate within the examination venue, accurately discerning the nature of their actions. Our approach represents a significant step forward in enhancing the adaptability and effectiveness of AI-powered exam supervision systems.