Driver Behavior and Action Prediction in Human-Computer Interaction

Distracted driving is a leading cause of traffic accidents, making real-time driver behavior prediction crucial for road safety. Leveraging deep learning, this study explores three efficient models–MobileNetV2, InceptionV3, and ResNet50V2–known for their strong performance and low computational cost in computer vision tasks. MobileNetV2's lightweight architecture is ideal for mobile applications, while InceptionV3 efficiently extracts features, and ResNet50V2 addresses the vanishing gradient problem in deep networks. To enhance accuracy, we incorporate both the Convolutional Block Attention Module (CBAM), which focuses the model on important image regions through channel and spatial attention, and a dynamic class weighting strategy that reallocates training resources based on validation performance. Our experiments compare the models with and without these enhancements, providing insights into their effectiveness for real-time driver behavior prediction and offering guidance for future practical implementations.