Accident Detection and Severity Analysis using Deep Learning

With rapid development in technology, much advancements have been made in automobile industry to ensure safety of the people. Despite, accidents have become unavoidable. In this paper, an image processing and deep learning based real time system is proposed for accident detection and classification of accidents based on severity thereby making cost estimation and timely medical assistance possible. 84.5% accuracy was obtained when manually collected images were trained and 230 images were tested via CNN Model. A database system is designed that contains information about the location, date, time, and image of the occurred accident. Further success on classifying accidents based on the severity with an accuracy of 96.8% was achieved when pre trained model VGG16 was used.