FIN: A Deep Multi-task Model for Face Information Detection

Yi-Shan Lai, Yi Chen, Li-Chen Fu, Pei-Yung Hsiao, Yu-Chieh Wang

Abstract: In recent years, with the development of autonomous driving technology, designing a driver monitoring system that can run on vehicles has become increasingly important. The driving system analyzes various factors such as head pose, gaze direction, and eye state to determine the driver’s attentiveness and aid the system in evaluating the driver’s capability to control the vehicle. In this paper, we present a lightweight multi-task deep learning model that utilizes full-face images to detect head pose, gaze direction, and eye state simultaneously. We propose a task-based cross-dimensional attention module (TCAM) that selectively filters and enhances relevant features for each specific task. Additionally, to tackle the lack of diverse head pose angles in the eye state dataset, we introduce a data augmentation technique to generate eye state data under different head poses, enhancing the model’s robustness in detecting eye states under varying head poses. We demonstrate the results of our work on the 300WLP, AFLW2000, BIWI, Gaze360, and MRL datasets. Extensive experiments show that our method can obtain competitive results with lower parameter costs. Furthermore, we establish a baseline for eye state detection using full-face images as the input on the CEW datasets.