IT4010 Research Project 4th Year, 1st Semester Final Report IT21166556 Nadeeshan I.U.N B.Sc. (Hons) Degree in Information Technology Sri Lanka Institute of Information Technology April 2025 1 DECLARATION Researchers affirm that this proposal is entirely this original work and does not contain any material previously submitted for a degree or diploma at any other university or institute of higher learning. Furthermore, researchers confirm that to the best of researcher’s knowledge, it does not include any previously published or written material by another person, except where appropriately cited in the text. Name Student ID Nadeeshan I.U.N IT21166556 Signature The above candidate is carrying out research for the undergraduate Dissertation under our supervision. Signature of the Supervisor Date 11/04/2025 ……………………………. ……………………………. 2 ABSTRACT The increasing number of road accidents in urban environments has highlighted the limitations of traditional traffic monitoring systems that rely on manual reporting or costly sensor-based mechanisms. This dissertation proposes a deep learning-based accident detection system that operates in real-time using existing CCTV infrastructure. The primary objective is to develop a cost-effective, scalable, and automated solution that enhances emergency response capabilities and improves public safety in smart cities. The methodology is centered around convolutional neural networks (CNNs), explicitly utilizing the MobileNetV2 architecture due to its proven performance in balancing accuracy and computational efficiency. A labeled CCTV image dataset was curated and divided into training, validation, and testing sets with balanced class distribution. Several deep learning models were trained and evaluated, with MobileNetV2 achieving the highest test accuracy of approximately 92% and low inference latency of 250 milliseconds. The trained model was integrated into a real-time processing pipeline using OpenCV and Flask, with ac