Real-Time Driver Behavior Recognition on Low-Cost Edge Hardware

This paper addresses a critical need in automotive safety by developing a real-time driver monitoring system (DMS) that can run on inexpensive hardware. The focus on low latency, power efficiency, and cost-effectiveness makes the research highly practical for widespread deployment. The combination of a compact vision model, confounder-aware label design, and a temporal decision head is a well-thought-out approach to improve accuracy and reduce false positives. The validation across diverse datasets and real-world testing further strengthens the paper's contribution. The discussion on the potential of DMS for human-centered vehicle intelligence adds to the paper's significance.