This program automatically grades multiple-choice tests that have been done using the following template:
At least 2 exams are needed to use the application. One exam to use as the answer key, and another exam to grade relative to the answer key.
After running the GUI.py file, a window will open. It displays the video of a detected camera in the computer:
When the answer key is shown in the feed, the "Select answer" button can be pressed. This will detect where all the correct answers are, it will also mark them in order to verify that they have been detected correctly.
After this is done, exams can be shown in front of the camera and corrected with the "Grade exam" button. At the bottom left of the window the number of correct, incorrect and not answered questions will be shown.
In order to correct more exams just click the "Next exam" button and repeat the last step for every new exam.
exam-correction.mp4
This project can be improved in many ways.
As of now the detection of correct and incorrect answers is only 99.5% accurate, in order to increase the accuracy the following things could be done:
-
Increase the number of training samples (A total of 3600 samples were used for the training of the neural network, which differentiates between marked and not marked answers)
-
Reduce video distortion. Due to the high FOV of the camera, straight lines in real life might appear curved in the display, this increases the distance between the calculated positions of the answers and the actual position. This can be clearly seen as the answers are not in the middle of the green or red boxes that surround them:
Centered answer:
Not so centered answer:
The interface could also be improved.
More templates could be added. Right now only a 60 questions template with 3 choices per question is possible.