The client is a US-based publication house that has offerings in the education segment for learners and educators. They manage a worldwide network of test centers and deliver high-stake examinations to certify and license individuals.
The Challenge
Meeting the demand for online exams with enhanced exam integrity
Due to the COVID-19 pandemic, the education sector underwent a rapid transformation from offline to online-based systems. As a result, the client recognized an increasing demand for online exams and requested an AI-Assisted Remote Proctoring System to meet the increased demand for online exams, all while ensuring exam integrity.
The Objective
Effective detection of prohibited items and unauthorized person with centralized monitoring
To meet the increased demand for online exams, our client was looking for a solution with centralized monitoring capability and detection of prohibited items and unauthorized persons during exams. The solution had to be compatible with low-configuration computers with basic webcams (<1 megapixel).
The Solution
Empowering online exams with customized AI solution
To address the client's challenges, we developed a customized AI solution that seamlessly integrated with their proctoring system. Our technical approach involved developing a custom AI model for detecting prohibited objects like smart watches, pencils, calculators, and books, and prohibited activities like looking away from the screen for a specified amount of time and moving away from the designated spot.
Our AI experts employed single-shot regression learning to draw bounding boxes and identify class labels accurately. A specialized module was deployed for checking stream quality for AI worthiness.
The solution's plug-and-play microservices-based architecture ensures future scalability. Real-time event tracking is enabled with start and end time markings.
The Impact
Experience efficient remote proctoring, prohibited item detection, and rapid response times
The solution resulted in an efficient remote proctoring system, with real-time analysis of live feeds, enabling central monitoring of large numbers of examinees worldwide.
The solution effectively detected prohibited objects and activities, including items like phones, smartwatches, and calculators, as well as behavioral analytics like eye gaze tracking, face detection and face-based authentication.
The solution delivered a response time of less than 2 seconds for addressing malpractice incidents, with a false positive rate below 10% to minimize unnecessary alerts for proctors. It is also capable of handling a burst load of 10,000 webcam video streams.
Key highlights:
- >5x Increase in efficiency of Examinee-to-Proctor ratio
- <$0.5 Cost per stream analytics
- 10,000 Video streams processing simultaneously
