A Driver's Visual Attention Prediction Using Optical Flow.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Automotive Safety

Background:

• Driver visual attention is crucial for road safety and is often estimated using computer vision models.
• Existing models primarily focus on scene appearance, with limited research on the impact of motion.
• Motion analysis in videos, representing object and scene movement, is a key visual cue.

Purpose of the Study:

• To investigate the effectiveness of motion information in estimating driver visual attention.
• To determine if motion features enhance the accuracy of driver attention prediction models.
• To address the gap in research regarding motion's role in driver attention estimation.

Main Methods:

• Developed a deep neural network framework utilizing optical flow maps to represent video motion.
• Extracted motion features from video sequences to predict driver attention locations and levels.
• Compared the performance of the motion-based model against state-of-the-art models using RGB frames.

Main Results:

• Experimental results on a real-world dataset confirmed that motion information significantly improves prediction accuracy.
• The proposed motion-based model demonstrated superior performance compared to models relying solely on appearance features.
• Motion features offer a substantial margin for enhancing driver attention prediction accuracy.

Conclusions:

• Motion is a critical factor in accurately estimating driver visual attention.
• Integrating motion analysis into computer vision models is essential for advancing driver attention prediction.
• Future research should leverage motion cues for more robust and reliable driver monitoring systems.