Tiny deep learning models for real-time and efficient embedded driver state detection

doi:10.1007/s11554-025-01836-2

Tiny deep learning models for real-time and efficient embedded driver state detection

2025-12-25

0

PRE

AI

S

Siham Essahraui *

A

Abdussalam Elhanashi *

DOI：10.1007/s11554-025-01836-2

Original

Original request for help

Save

Abstract

En 中文

Driver distraction and inattention are major contributors to traffic accidents, making reliable real-time driver state detection essential for intelligent transportation systems. Although deep learning (DL) has significantly improved detection accuracy, conventional models remain computationally demanding and are often unsuitable for deployment on embedded or low-power hardware. To address these limitations, this study benchmarks seven lightweight architectures—Efficient-Tiny, ESPNetv2-Small, MCUNet, Micro-MobileNet, PhiNets, ShuffleNetLite, and SqueezeNetv11—within a unified evaluation framework tailored for embedded deployment. Experimental results demonstrate that ESPNetv2-Small and SqueezeNetv11 achieved the highest accuracies of 99.50% and 95.81%, with compact model sizes of 0.74 MB and 0.33 MB and inference speeds of 40.49 FPS and 28.60 FPS, respectively. Efficient-Tiny, despite its reduced footprint (0.43 MB), reached 91.64% accuracy. These findings confirm that modern TinyDL architectures can deliver near state-of-the-art performance with minimal computational cost, offering a viable path toward energy-efficient, real-time driver monitoring in next-generation intelligent vehicles.

Keywords:

Deep learning

Lightweight models

Driver distraction detection

Real-time

Computer vision

TinyML

Embedded devices

Tiny deep learning models for real-time and efficient embedded driver state detection

Abstract

Journal

Organization