Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- de97ff85-feae-4fbf-9e9f-4d6bafc6a170
- Published
- 2026-10-30
Universitas Dian Nuswantoro (UDINUS), through the participation of Winarno, Sri; Alzami, Farrikh; Naufal, Muhammad; Al Azies, Harun; Soeleman, Moch Arief; and Malim, Nurul Hashimah Ahamed Hassain, conducted a study on real-time driver drowsiness detection through sequential image processing. The activity presents the first integration of EfficientNet with time-distributed layers and K Nearest Neighbors (KNN) classification for this purpose. The approach differs from methods that analyse individual frames in isolation by capturing temporal patterns in facial expressions.
The proposed EfficientNet-KNN architecture combines EfficientNet compound scaling for feature extraction with KNN for sequential pattern classification. It processes five-second video segments through time-distributed layers. This process reduces feature dimensionality from 1,228,800 to 1,280 while preserving critical temporal information. Preprocessing incorporates facial detection using Haar Cascade filters and Global Average Pooling for computational optimisation. The study was experimentally validated using the ULg Multimodality Drowsiness Database (DROZY) across dataset proportions of 25%, 50%, 75%, and 100%.
On the complete dataset, the integrated model achieved 99.52% accuracy with an execution time of 0.55 seconds. The study reports a 49.9-fold improvement in computational efficiency compared with baseline KNN approaches while maintaining higher accuracy. Statistical t-test analysis confirmed significant performance differences at p<0.05, and Cohen’s d indicated large effect sizes for improvements in accuracy and speed.
The evaluation remains limited to controlled laboratory conditions and has potential demographic bias. The study therefore identifies the need for comprehensive real-world validation across diverse driving environments and population groups. In relation to the stated goal of three, this UDINUS activity contributes research on a computationally efficient drowsiness detection framework for real-time automotive applications, balancing accuracy with resource requirements in support of driver safety systems.