Lightweight Driver Monitoring System Based on Multi-Task Mobilenets

Area of Science:

• Computer Science
• Artificial Intelligence
• Automotive Engineering

Background:

• Driver status recognition systems aim to reduce accidents caused by distraction and drowsiness.
• Deep learning shows promise, but visual monitoring systems face challenges with high processing demands and hierarchical structures.
• Existing systems are not widely adopted due to computational requirements and sensitivity to errors in sequential processing.

Purpose of the Study:

• To develop a lightweight and efficient driver monitoring system for real-time recognition of driver distraction and drowsiness.
• To overcome the limitations of high-performance processors and hierarchical structures in current driver monitoring technologies.
• To improve the accuracy and practicality of driver status recognition using resource sharing.

Main Methods:

• Proposed a novel method using Mobilenets without face detection/tracking to recognize facial behaviors indicating distraction.
• Developed a lightweight system based on Multi-Task Mobilenets (MT-Mobilenets) incorporating a multi-task classifier with three Softmax regressions.
• Leveraged a driver's mobile phone as a resource-sharing device to augment the processing capabilities of a Raspberry Pi.

Main Results:

• The proposed MT-Mobilenets approach effectively recognizes facial behaviors linked to driver distraction, fatigue, and drowsiness.
• Integrating a mobile phone significantly improved the frames per second (FPS) processing rate on the Raspberry Pi.
• The system demonstrated enhanced accuracy in driver status recognition compared to using the Raspberry Pi alone.

Conclusions:

• The MT-Mobilenets based system offers a practical solution for driver status recognition by mitigating computational constraints.
• Resource sharing with a mobile device is a viable strategy to enhance the performance of embedded driver monitoring systems.
• This approach paves the way for wider adoption of advanced driver monitoring systems in the automotive industry.