Auto-proctoring using computer vision in MOOCs system

Abstract

The COVID-19 outbreak has caused a significant shift towards virtual education, where Massive Open Online Courses (MOOCs), such as EdX and Coursera, have become prevalent distance learning mediums. Online exams are also gaining popularity, but they pose a risk of cheating without proper supervision. Online proctoring can significantly improve the quality of education, and with the addition of extended modules on MOOCs, the incorporation of artificial intelligence in the proctoring process has become more accessible. Despite the advancements in machine learning-based cheating detection in third-party proctoring tools, there is still a need for optimization and adaptability of such systems for massive simultaneous user requirements of MOOCs. Therefore, we have developed an examination monitoring system based on advanced artificial intelligence technology. This system is highly scalable and can be easily integrated with our existing MOOCs platform, daotao.ai. Experimental results demonstrated that our proposed system achieved a 95.66% accuracy rate in detecting cheating behaviors, processed video inputs with an average response time of 0.517 seconds, and successfully handled concurrent user demands, thereby validating its effectiveness and reliability for large-scale online examination monitoring.