Integrated deep learning framework for driver distraction detection and real-time road object recognition in advanced...

Area of Science:

• Artificial Intelligence
• Computer Vision
• Automotive Safety

Background:

• Road user safety is a global concern, with driver distractions being a primary cause of accidents.
• Existing Advanced Driver Assistance Systems (ADAS) often lack comprehensive integration of driver monitoring and real-time environmental awareness.

Purpose of the Study:

• To develop an integrated system for real-time driver distraction detection and road object recognition.
• To enhance the capabilities of Advanced Driver Assistance Systems (ADAS) by providing context-aware warnings and improving situational awareness.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) and transfer learning to categorize driver behavior (physical, visual, cognitive distractions).
• Employed YOLO (You Only Look Once) for real-time detection of vehicles, pedestrians, lane markers, and traffic signals.
• Integrated a decision-making module to assess danger levels based on combined driver and road data.

Main Results:

• Achieved high accuracy and computational efficiency in driver distraction detection using CNNs and transfer learning.
• Demonstrated real-time performance with YOLO for essential road object recognition.
• Validated system reliability across diverse conditions (rain, fog, low-light) through data augmentation and model optimization.
• Achieved 25 FPS on an NVIDIA Jetson Xavier NX platform, indicating feasibility for resource-constrained ADAS.

Conclusions:

• The integrated system effectively combines driver monitoring and road awareness for a comprehensive safety solution.
• The developed system enhances ADAS capabilities, leading to reduced accidents and improved driver safety.
• Demonstrated practical feasibility for real-time embedded deployment in safety-critical automotive applications.