SeatID: Design, Functional Characterization, and Preliminary Computational Validation of a Piezoresistive Smart Cushion Toward Continuous Biometric Authentication in Virtual Examinations

Introduction: Verifying student identity during online examinations is still a challenge for educational institutions. Traditional methods such as webcams and human proctoring may not detect situations in which a different person replaces the student outside the camera's field of view. To address this problem, we developed SeatID, a smart cushion that uses pressure and motion information to continuously verify the identity of the user during an examination. If a change of user is detected, the system can automatically block the session.

Methods: SeatID uses an ESP32 microcontroller connected to a 14×14 Velostat pressure sensor matrix and multiplexers to record the pressure pattern produced while sitting. User identification is based on individual seating characteristics, including body weight distribution and anatomical differences such as the distance between the ischial tuberosities. An Inertial Measurement Unit (IMU) was also included to monitor movement and detect unusual behavior that could indicate a user replacement attempt. Sensor data were transmitted via Wi-Fi to a Python server for processing. User classification was performed using the k-Nearest Neighbors (k-NN) algorithm. The model was trained with simulated data and later tested using the physical prototype.

Results: A dataset of 1,800 records was collected under controlled conditions. Using 10-fold cross-validation, the k-NN classifier achieved an average accuracy of 99.94% with a standard deviation of ±0.17% across 360 test samples. The system also obtained an Equal Error Rate (EER) of 0.00%, indicating effective discrimination between authorized users and impostors.

Conclusions: The proposed system successfully combined pressure and motion data to provide continuous user authentication. The developed prototype achieved reliable performance under controlled conditions and detected user replacement attempts in real time.