team name:rp++
members:
XIA Shujun (Leader)
XU Yuan
NG Cheuk Yiu
CHEN Yihuan
Ng Pak Lam
In the foreseeable future, online classes will continue to be used as a complementary means of teaching and learning. One of the disadvantages of online classes is that the teacher cannot be in front of the students as in an offline classroom and thus follow the students' expressions to see how they are doing. When a student is confused and not keeping up, the teacher cannot detect it in real-time. Also, students are easily distracted and it’s hard to be monitored by the course instructors online. Such a situation calls for a mechanism to detect the expression of students to return real-time feedback to teachers and thus improve class interaction.
Our idea is to develop a real-time facial expression detection tool which analyzes facial attributes to calculate the confusion or attention level of students and give feedback to teachers.
- Input video sequence of student
- Extract features like head pose, eye direction, lip movement etc.
- Calculate the confusion/attention level using the deviations
- Return the feedback
sleepy count:
Blinking count: eye aspect ratio of 0.2 for 3 consecutive frames
Yawning count: mouth aspect ratio of 0.5 for 3 consecutive frames
Sleepy nod count: pitch(x) rotation angle of 0.3 for 3 consecutive frames
Confusion count:
For the first 59 consecutive frames:
The algorithm takes the distance between the eyebrows for each frames and calculates the average distance for 59 frames.
Calculation:
-
For each 20 consecutive frames, a period score will be recorded
-
Period score = 100-total_sleepy_cnt * 8 - total_confused_cnt + 60
-
The total score is the average of the period score.
-
Full score is 100**
By processing the data of each student to obtain an overall level, we can know the learning effectiveness of the whole class, which provides a reference for teachers to improve teaching methods.
This system helps teachers comprehensively understand students’ attentiveness and improve the teaching quality by promptly tracing students’ situations.