Pothole Detection Using Deep Learning Classification Method

Abstract

Potholes on roads have been a primary reason for road disasters and damage to the vehicles. Currently due to heavy rains and poor structure accoutrements roads surface have flaws. Detecting the potholes manually is time consuming and will not be detecting the flaw impeccably. The proposed system aim is to recognize potholes on muddy roads and high way roads[1] pictures in order to avoid disasters and damage to the vehicles. Deep learning algorithms are used to classify image dataset in order to determine whether the roads are plain or have potholes. Images are collected from internet sources,(muddy roads) dataset and another dataset is taken from Kaggle (highway roads) dataset. Pretrained models Resnet50, IncepectionV2 and VGG19 are used for training the model. Web application is implemented for testing the model to identify the roads whether there is a pothole or plain based on the selected models like Resnet50, InseptionResNetV2, and VGG19 models are trained for the system. The model performances are analysed for better accuracy, precision and recall. Compare to Resnet50, IncepetionResNetV2 models. VGG19 model has given the best accuracy with 97 percentage for highway roads and 98 percentage for muddy roads.