Distraction Evaluation by Facial Landmark Detection with Lightweight Multi-Task Neural Network

EasyChair Preprint no. 11300

Abstract

Recently, most of distraction detection research results focus on the driver's distraction detection in a car and the detection object is almost the one. However, the detection object may be more than two in some application scenarios. For example, using a camera to perform attention detection on multiple students, how to accurately assess students' attention without disrupting their learning becomes our primary research goal. In this paper, the behavior of student was defined as the student's distraction while the student turns his head, yawning, and eyes closing in the classroom. In this condition, most of existed researches only could be used for one-object detection but could not be used for multi-object detection through one detector. Moreover, less of researches could be used for facial landmark detection with multi-object. Hence, a distraction evaluation by facial landmark detection for multi-object by multi-task neural Network is required. To reduce the cost and space, and improve the ease of installation, our proposal is designed with the embedded system. Therefore, a distraction evaluation by facial landmark detection with lightweight multi-task neural network, DEFLD-LMTNN, was proposed in this paper to address the above issues. In DEFLD-LMTNN, the distraction detection could be applied for multi-object. When the behavior was evaluated as our defined distraction, the student will be marked as distracted in the monitor screen and an alert could be notified to teacher immediately. The teacher also could track student's learning status based on the number or frequency of distraction afterwards by our DEFLD-LMTNN. In the experimental results, the accuracy of DEFLD-LMTNN could be up to 90%. It was proved that distraction evaluation by facial landmark detection with lightweight multi-task neural network, DEFLD-LMTNN, proposed in this paper could be applied for distraction evaluation with multi-object in the embedded system with low cost and space.

Keyphrases: Distraction Detection, embedded system, Facial Landmark Detection, Lightweight Multi-Task Neural Network, multi-object detection

@Booklet{EasyChair:11300,
  author = {Xie Xin-Hua and Wu Yi-Chao},
  title = {Distraction Evaluation by Facial Landmark Detection with Lightweight Multi-Task Neural Network},
  howpublished = {EasyChair Preprint no. 11300},

  year = {EasyChair, 2023}}