Real Time Pothole Detection in Video Streams Using AI-Driven Deep Learning Techniques

Abstract

Road potholes are a major cause of traffic accidents, vehicle damage, and increased maintenance costs, while traditional manual inspection methods are slow and inefficient. To address this problem, this project proposes a Real-Time Pothole Detection System in Video Streams Using AI-Driven Deep Learning Techniques. The system employs a YOLOv3-Tiny deep learning model, based on convolutional neural networks, for fast and accurate pothole detection. The trained model is integrated into a Flask-based web application using OpenCV’s Deep Neural Network (DNN) module. Live video input is captured from a mobile camera stream and processed frame by frame to identify potholes, which are displayed with bounding boxes and confidence scores along with an alert mechanism which produces beep sound when the pothole is detected. This system provides road safety and reduces the need for manual inspection.