International Research Journal of Engineering and Technology (IRJET) Volume: 09 Issue: 04 | Apr 2022
www.irjet.net
e-ISSN: 2395-0056 p-ISSN: 2395-0072
Online Exam Proctoring using Deep Learning Jai Pawar1, Jyoti Prasad1, Prathamesh Shanbhag1, Charmi Chaniyara2 of Information Technology, Atharva College of Engineering, Maharashtra, India Professor, Dept. of Information Technology, Atharva College of Engineering, Maharashtra, India ---------------------------------------------------------------------***--------------------------------------------------------------------2Assistant
1Dept.
Abstract - The Covid-19 pandemic has affected the
speech recognition. The approach helped us as a reference for creating our system.
educational system worldwide, leading to the near-closures of schools, universities, and colleges. This has resulted in increased demand for E-learning, and the subsequent growth in the remote learning industry. With E-learning, exams are also undertaken remotely. Research suggests most examinees hold the perception that it is easier to subvert the system in online exams than in traditional ones. To ensure the smooth running of exams, Online Exam Proctoring will be essential, where it would ensure fair, unbiased proctoring of exams. The goal of the project is to be able to develop a website that will track examinee activity during the exam to prevent any malpractice.
In [3] J. Huang, V. Rathod, C. Sun, M. Zhu, A. Korattikara, A. Fathi, et al compares the Speed and accuracy of multiple popular modern convolutional networks, specifically object detection networks. The metric selected to evaluate the above models was mAP, while speed was calculated in milliseconds. The findings from this paper helped us finalize an object detection model that best fits our use case. In [4] Liu, W. et al proposed a method for detecting objects in images using a single deep neural network, the Single Shot Detector. SSD is simple when compared to existing methods that require object proposals, like RCNN because it completely eliminates proposal generation and subsequent pixel or feature resampling stages and encapsulates all computation in a single network. As a result of this, SSD is simple to train and integrate into systems that require a detecting component.
Key Words: SSD, coco, proctoring, EfficientNet
1. INTRODUCTION This paper proposes a system to address and mitigate the difficulties faced by proctors during the execution of examinations online. The proposed system is a noninvasive software, that automates routine remote exam proctoring tasks. It is in the form of a website, which allows examinees to attempt the exam with their webcams switched on. The system tracks examinee presence, mobile phone presence, presence of more than one individual, and book presence in the webcam feed using Deep Learning techniques, and immediately warns the examinee if any of the above are found. The Deep Learning approach used for object detection is a Single Shot Detector (SSD), which was pre-trained on the coco dataset. The teacher can schedule tests via google forms links by inputting them to the website, and the examinees can access active test links from the attempt test section.
3. ANTI CHEATING MEASURES Common cheating methods observed after an anonymous survey of examinees were: 1.
Copying questions in text format and pasting them into a group chat in another window.
2.
Capturing an image of the questions via a mobile phone and circulating it to a group of coconspirators.
3.
Referencing books/study material from in front of the computer screen itself.
4.
Moving away from the computer screen to reference books/study material.
5.
Inviting co-conspirators to attempt the test on the same laptop/machine.
2. LITERATURE REVIEW In [1] Nigam, A., Pasricha, R., Singh, T. et al. reviews more than 40 research articles on automated proctoring software and present drawbacks of earlier methods as well as advantages of newer methods. It served as a guide to keeping track of advancements in this sub-field with updated info, considering it was published in 2021.
Measures undertaken by our software mitigate/prohibit the above cheating methods:
In [2] Y. Atoum, L. Chen, A. X. Liu, S. D. H. Hsu and X. Liu proposed a system, which employs two cameras to record the examinee from two different angles, along with a microphone to record audio data. Data obtained from the above input devices are used for gaze detection and
© 2022, IRJET
|
Impact Factor value: 7.529
1.
to
Mandatory enforced Fullscreen mode: Prohibits the examinee from opening another tab or window, eliminating Section 3, point 1.
|
ISO 9001:2008 Certified Journal
|
Page 3254