Pothole Detection Using Machine Learning Algorithms
Adiba Masud, Saraban Tasnim Sharin, Khandokar Farhan Tanvir Shawon, Zakia Zaman
Abstract
Potholes are one of the greatest problems on the roads of Bangladesh. Stuck water on the roads and overloaded vehicles are mainly responsible for surface decay and erosion of rocks under the road surface which creates potholes that cause a lot of accidents and risks for general people. A system is needed that will detect potholes not only to alert the drivers but also to alert the authorities. In this paper, we emphasized on detecting the potholes using pothole image data and normal road image data. At first, we collected the data and then preprocessed them by resizing and rescaling. We used MobileNetV2 to extract the features and we reduced the dimension of the features using PCA, LDA, and t-SNE. Finally, for training, we applied five Machine Learning classification algorithms which are Support Vector Machine (SVM), Logistic Regression, Random Forest, Elastic Net, and Decision Tree. The results from Logistic Regression, Elastic Net, and Support Vector Machine (SVM) are relatively better than the other two. After setting up the three of them side by side we observed that Support Vector Machine (SVM) works best for our system and gave an accuracy of 99%.