A Comprehensive Evaluation of YOLO11n Optimization Techniques for Driver Drowsiness Detection

DOI

10.2991/978-94-6239-828-3_8How to use a DOI?

Keywords

Driver drowsiness detection; YOLO11n; Pruning; Quantization; Ghost Module; Edge computing

Abstract

Driver drowsiness is a major contributor to road traffic accidents, highlighting the need for accurate and computationally efficient detection systems suitable for real-time deployment on resource- constrained edge devices. This study presents a comprehensive evaluation of YOLO11n optimization strategies for driver drowsiness detection using the NTHU Driver Drowsiness Detection (NTHU-DDD) dataset. The investigated approaches include TensorRT-based quantization (FP16 and INT8) and unstructured magnitude pruning (10%, 20%, and 30% sparsity). All models were trained under identical hyperparameter settings and evaluated using detection accuracy and com- putational efficiency metrics, including precision, recall, mean Average Precision (mAP), inference speed, and model size. Experimental results demonstrate that each optimization strategy offers distinct trade-offs be- tween accuracy and efficiency. FP16 quantization provided the best balance between detection performance and real-time inference, achieving the highest throughput of 35.91 FPS while maintaining an mAP@50 of 0.9721. INT8 quantization produced the smallest model size (4.80 MB) with only a marginal reduction in detection accuracy, making it suitable for memory-constrained edge devices. Moderate pruning levels (10% and 20%) slightly improved generalization, whereas 30% sparsity caused sub- stantial performance degradation. Overall, FP16 TensorRT quantization emerged as the most practical optimization strategy for real-time edge deployment, offering an effective balance between detection accuracy and computational efficiency. These findings provide practical guidance for developing efficient driver monitoring systems for intelligent transportation applications.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Achmad Rohmandzoni
AU  - Aulia Riza Mufita
AU  - Nada Nadhira Najwa Mazaya
AU  - Fitra Abdurrachman Bachtiar
AU  - Novanto Yudistira
PY  - 2026
DA  - 2026/11/04
TI  - A Comprehensive Evaluation of YOLO11n Optimization Techniques for Driver Drowsiness Detection
BT  - Proceedings of the 2026 11th International Conference on Sustainable Information Engineering and Technology (SIET 2026)
PB  - Atlantis Press
SP  - 77
EP  - 87
SN  - 2589-4900
UR  - https://doi.org/10.2991/978-94-6239-828-3_8
DO  - 10.2991/978-94-6239-828-3_8
ID  - Rohmandzoni2026
ER  -