Year: 2026
Type: preprint
Abstract: To address the safety issues in construction machinery caused by abnormal driving behaviors, this paper proposes a Multimodal Safe Driving Monitoring System (MSDMS). The system consists of two main components. The first is the Multimodal Sentiment Analysis model based on Hybrid Attention (MSA-HA), designed for emotional driving detection. By integrating hybrid attention mechanisms with a Tensor Fusion Network, this model enhances multimodal feature interaction and improves the accuracy of emotion detection. The second component employs the YOLOv8n-DPCLN model for the detection of fatigue and distracted driving. In this model, a Deepwise Partial Convolution Layer Network (DPCLN) replaces the bottleneck within the C2f module to enhance overall performance. Experimental results demonstrate that the MSA-HA achieves binary and ternary classification accuracies of 83.5% and 74.76%, respectively, on the SIMSv2 dataset, outperforming other mainstream models. Compared to YOLOv8n, the YOLOv8n-DPCLN improves [email protected] and [email protected]–0.95 by 1.9% and 3.7%, respectively, while reducing the parameter count by 0.4 M and computational cost by 1.1 GFLOPs.Finally, MSDMS achieves recognition rates of 78.7%, 93%, and 86% for emotion, fatigue, and distracted driving, respectively, thereby satisfying the requirements for safe driving monitoring.
Source: Research Square
Institution: Nanjing University of Posts and Telecommunications
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7235786423
