Pothole Detection

Access the research paper here

For this project, I designed and built an intelligent system capable of detecting road potholes in real time using drone imagery and deep learning. I developed a Convolutional Neural Network (CNN) that analyzed video frames captured by a custom-built FPV drone equipped with a GoPro camera. Using Python and TensorFlow, I combined image processing techniques - including grayscale conversion, edge and contour detection - with deep learning to accurately classify road conditions.  The model was trained on a dataset of 784 images and achieved 99.02% training accuracy and 99.12% testing accuracy, proving both its precision and scalability. One of the biggest challenges was differentiating potholes from manholes or puddles, which required extensive fine-tuning and data preprocessing. It also forced to expand my electronics skills, as I found the most energy efficient way to integrate and run the model on the <500g drone. Through this project, I strengthened my skills in machine learning, computer vision, Python programming, and hardware integration. It also taught me the value of experimentation and iterative testing. The final outcome demonstrated how AI-driven drone systems can make infrastructure monitoring safer, faster, and more cost-effective.