The main goal of this project is to develop a system that can automatically detect potholes using computer vision techniques, reducing the need for manual road inspections and accelerating road maintenance efforts. Potholes are a major contributor to road accidents and vehicular damage, particularly in urban areas. The project combines image processing with deep learning especially Convolutional Neural Networks (CNNs) to identify potholes from images or video streams captured by dashcams, surveillance cameras, or drones. By integrating GPS, the system can also record the exact location of the detected potholes and transmit alerts to municipal authorities for repair. By the end of the project, students will deliver a complete prototype that processes visual data, detects road anomalies, and provides geolocated alerts, contributing to smarter urban infrastructure solutions.
The project spans a twelve-week timeline with progressive development milestones. In the initial weeks, students will explore machine learning fundamentals, particularly deep learning models like CNNs, and set up the development environment using platforms such as Anaconda or Google Colab. They will also begin collecting and preparing a dataset of road images with labeled potholes.
During the middle phase, students will implement the core detection model, train it on varied datasets, and test it against new inputs to validate accuracy. Libraries such as TensorFlow, OpenCV, and Keras will be central to the development process. As the model matures, students will work on improving its precision and integrating GPS functionality for real-time location tagging. In the final weeks, full project integration, system testing, performance tuning, and documentation will be completed. Deliverables will include a complete working model, codebase, results, and a team presentation. The project focuses solely on pothole detection, with other types of road anomalies considered optional.
