Pothole Patrol: Harnessing Machine Learning for Automated Detection and Filling

Abstract

  This project leverages machine learning and image processing to identify and repair potholes, with the goal of enhancing road safety and minimizing maintenance expenses. The system is trained on a dataset of pothole images and, once deployed, it can detect potholes in real-time through a webcam. Upon identifying a pothole, the system guides a vehicle to fill it with appropriate material. This process can be conducted periodically to ensure timely repairs and maintain road quality. An ultrasonic sensor attached to the vehicle measures the pothole depth, and the filling process is fine-tuned using a trial-and-error approach. The vehicle’s movement is controlled by DC motors, while servo motors regulate the discharge of the filling material. Another ultrasonic sensor monitors the material level, and if it falls below a predefined limit, a notification stating 'Insufficient Material' is sent via SMS to a specified contact number. The GSM module, connected to a Raspberry Pi microcontroller, facilitates this communication. A dataset of around 300 images with and without potholes, sourced from Kaggle, is used to train and test a convolutional neural network (CNN) model. Edge detection is applied during preprocessing to identify the pothole's contour. Image data processing is carried out using the OpenCV library in Python, integrated with Visual Studio. This system supports safe and efficient road repair, while enhancing productivity and the quality of road maintenance operations.