The Cerberus cameras in my LT1 digital twin perform both crowd localisation and crowd counting. Since they rely on face detection, they can also be used to track student engagement—if a student's face is directed towards the screen, it is inferred that they are paying attention.

During lectures, students move their faces frequently, introducing noise into the data. Instead of filtering out this variation, head movement is treated as a proxy for engagement. Over time, this generates a binary timestamp list of when faces were up or down, forming a unique barcode-like pattern for each seat.

Since space on the floorplan is limited, traditional barcodes are transformed into circular charts, where each slice represents a time step, similar to a clock’s minute hand. This merges the concepts of a barcode and a clock into a dynamic temporal pie chart, updated continuously as new data arrives from the Cerberus cameras.

1-Minute Data Collection

60-Minute Data Collection Spanning Entire Lecture

The original design featured black dots with grey slices, which were difficult to interpret. Additionally, it wasn’t immediately clear whether the visualisation represented Cerberus data or other environmental sensors. To improve clarity, emoji masks were added over occupied seats, and an explanatory illustration was included in empty spaces.

Immediate insights emerge from the visualisation, such as a student in the bottom row switching seats midway through the lecture.

However, this system also highlights a limitation of the Cerberus sensors. Seats at the edges of the camera’s field of view may suffer from distortion due to the wide fisheye lens, leading to noisier data in these regions. This could give the false impression that some students are less engaged than they actually are.