Pothole Detection Using Image Processing and Machine Learning

One major cause of vehicle damage, traffic disruption and accidents are road degradation through potholes. Traditionally, detecting and repairing potholes is very labor intensive and inefficient. This paper presents a new way of pothole detection automation using image processing and machine learning techniques. Using computer vision, I suggest to detect the potholes in the road surface using the irregularities seen in the road surface. Then, key image processing methods are used to preprocess and enhance the images; these include edge detection, contour analysis, texture segmentation and the like. We use the lightweight convolutional neural network, MobileNet, architecture for classifying aswell as pothole detection with high accuracy. Hence, the model is integrated with real time data of Road Surveillance Cameras or Vehicle mounted Cameras so that timely alerts to the road maintenance authorities can be made. It is shown that the model can detect with high accuracy and robustness under various lighting and weather circumstances, according to the experimental results. This may increase the road maintenance efficiency by large extent and bring down the costs, as well as improve the road safety.