Remote Proctoring System Using Video, Audio and System Scenario Analysis

The onset of the global pandemic saw most academic events shift to an online mode of conduction. The lack of offline interaction gives room for individuals to participate in acts of malpractice which are very difficult to monitor given the lack of access to view the student’s full physical environment. These are also laborious to monitor and can prove to be an expensive and time-intensive affair to have human proctors monitor long exams/interviews. Our work entails a novel approach to building an efficient remote proctoring system using machine learning that can be used to proctor these academic events to ensure that they are conducted fairly and without any form of discrepancy and malpractice. The paper highlights our pipeline of analysing audio, video and system processes and cumulating them into ‘scenarios’ which we then classify into different levels of malpractice. Our work also thoroughly explores system design for handling high volumes of such audio and video data using a combination of edge and cloud processing for the same. Even with our comprehensive outlook for processing data i.e. audio, video and system as a scenario we have achieved an accuracy of 88.91 per cent making our system more accommodative compared to other systems in the market.