2024, Vol. 5, Issue 2, Part A

Potholes and unevenness in road surfaces tend to be accident-prone. Potholes disrupt the flow of traffic as drivers navigate around them cautiously for the reason that they pose hazards to drivers and pedestrians. Addressing these hazards promptly and efficiently is crucial to ensure the safety of road users. Therefore, periodic maintenance of roads should be made to keep roads safe and sustainable for the people. However, manual road inspection leads to several challenges such as vast road networks, limited manpower, outdated equipment and budget constraints.

Eventually, automating the process of pothole detection is suggested to improve quick road maintenance and to avoid accidents. Automation provides a faster assessment of road conditions compared to manual inspections and prevents major repairs through early detection.

Ultimately, this research work aims to improve the detection process using deep learning technique by refining the VGG16 convolutional neural network model. Modifications are made to the VGG16 network by removing some convolution layers and using different filter sizes. Additionally, this paper utilizes the YOLOv8 model for comparison of results with those obtained from VGG16. The models are trained using a pothole dataset taken from kaggle, which includes both normal road conditions and images containing potholes. The paper analyses the performance of pre-trained VGG16, VGG16 architecture trained without using pre-existing weights, modified VGG16 and YOLOv8 models on the dataset. This work compares the performance of models by evaluating accuracy, model size and inference time. The focus of this study is to achieve a balance between accuracy and speed in pothole detection, thereby enhancing road maintenance processes.