Publication

Abstract : There have been many developments concerning road safety, but accidents continue to occur at constant intervals, and the major causes of these accidents are improper road conditions, such as potholes and speed breakers. This paper proposes a deep learning-based real-time detection system that identifies the anomalies above under various seasonal conditions. To recognize these anomalies, an InceptionNet V3 was trained on 12,390 images belonging to one of six categories: Potholes on Paved Roads during Rainy (PPR) and Summer (PPS) seasons, Potholes on Unpaved Roads during Rainy (PUR) and Summer (PUS) seasons, Speed Breakers, and Plain Roads. The dataset was preprocessed by performing normalization, resizing, and augmentation. Our model, based on this preprocessed dataset, was able to deliver a detection accuracy of 96%. The real-time detection was implemented using Raspberry Pi and a webcam; anomalies were detected, and alerts were issued to drivers to prevent mishaps. The InceptionNet V3 system was proved highly efficient along with edge computing on Raspberry Pi, lessening the latency and processing in real-time. The proposed solution can be used in autonomous vehicle navigation and road maintenance systems to enhance safety by providing real-time notifications and detecting anomalies. It addresses the limitations of other methods, such as scalability, cost, and the need for real-time responsiveness.