Skip to content

Deep Learning based Smart Driver Assistance System for Automated Detection of Road Anomalies

Oct 2026 · International Journal of Sensors Wireless Communications and Control · 0 citations

Abstract

Road anomalies such as potholes, surface cracks, and speed bumps present a persistent hazard to vehicles, motorists, and pedestrians. Their automatic detection is a significant issue because they damage cars, disrupt traffic, accelerate the deterioration of road infrastructure, and reduce driver comfort. Advanced driver assistance systems address this need by allowing autonomous and semi-autonomous vehicles to identify and respond to road anomalies in real time. The present work proposes a deep learning-based smart driver-assistance system for the automated detection of road anomalies. A 13-layer convolutional neural network (CNN) with 232,034 trainable parameters is used both as an end-to-end classifier and, in a separate hybrid configuration, as a feature extractor whose 512-dimensional output is passed to a light gradient boosting machine, histogram-based gradient boosting, extra trees, random forest, and k-nearest-neighbour classifiers. Images from four public datasets are resized to 256 x 256 x 3, scaled to [0, 1], and class distributions are balanced using the synthetic minority over-sampling technique. The proposed CNN attains 99.00% accuracy, 99.13% precision, and a 98.98% F1 score in the two-class classification of cracks and potholes and 97.67% accuracy, 97.54% precision, and a 97.61% F1 score in the three-class classification of cracks, potholes, and speed bumps. Its computational cost is additionally compared with that of transformer-based classifiers and contemporary convolutional backbones. The proposed network reaches this accuracy with 232,034 parameters, which is between one and three orders of magnitude fewer than the established convolutional and transformer backbones evaluated here. This combination of accuracy and low parameter count is the property that matters for an embedded driver assistance unit, where memory and inference latency are constrained. The results indicate that a compact, purpose-designed CNN is a practical basis for onvehicle road anomaly detection. Thus, this approach could contribute to safer travel, reduced vehicle deterioration, and enhanced performance of autonomous vehicles. The evaluation is confined to offline public image datasets, and on-road validation remains to be carried out.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.