In this paper, we introduce a multimedia analytics system to perform automatic and continuous online exam proctoring (OEP) . The overall goal of this system is to maintain academic integrity of exams, by providing real-time proctoring to detect the majority of cheating behaviors of the test taker. To achieve such goals, audio-visual observations about the test takers are required to be able to detect any cheat behavior.
OEP System Summary
Our system monitors such cues in the room where the test taker resides, using two cameras and a microphone. The first camera is located above or integrated with the monitor facing the test taker. The other camera can be worn or attached to eyeglasses, capturing the field of view of the test taker. In this paper, these two cameras are referred to as the "webcam" and "wearcam" respectively. The webcam also has a built-in microphone to capture any sound in the room. We propose a hybrid two-stage algorithm for our OEP system. The first stage focuses on extracting middle-level features from audio-visual streams that are indicative of cheating. These mainly consists of six basic components: user verification, text detection, speech detection, active window detection, gaze estimation, and phone detection. Each component produces either a binary or probabilistic estimation of observing certain behavior cues. In the second stage, a joint decision across all components is carried out by extracting high-level temporal features from the OEP components at the first stage. These new features are utilized to train and test a classifier to provide real-time continuous detection of cheating behavior. To evaluate the OEP system, we collect multimedia (audio and visual) data from 24 subjects performing various types of cheating while taking a multiple choice and fill in the blank math exam.