| Abstract |
Visually impaired pedestrians often face challenges navigating outdoor environments due to difficulties in identifying road surface quality. To enhance safety, we propose a deep learning-based archite...cture that can be easily deployed on mobile devices for real-time assistance, potentially reducing road injuries. Our suggested approach builds upon the pre-trained CNN, MobileNetV2, by adding supplementary layers without increasing computational complexity. The model is evaluated on unseen images, with results indicating improved classification performance in terms of F1-score, recall, accuracy, and precision compared to alternative models, including Random Forest, ResNet, EfficientNet, and InceptionNet. Our proposed model achieves 93.20% accuracy relative to MobileNetV2. However, the architecture does not account for obstacles on road surfaces, which could also cause injuries. The modified MobileNetV2 architecture effectively classifies road surfaces to assist visually impaired pedestrians and can be seamlessly integrated into mobile devices. Our work presents a novel, efficient, and low-power system with enhanced accuracy for road quality classification, suitable for deployment on diverse devices such as smartphones.show more
|
Export Mendeley
p706-714