TRAFFIC ACCIDENT EVENT PREDICTION IN TRAFFIC VIDEO USING DEEP LEARNING









Abstract

Road accidents are considered as one of the leading causes of human fatality in recent decade. Using machine vision in automobile sensors and through warning even a second before accidents occur, a human disaster can be prevented. In order to predict accident at the stage, it is needed to learn temporal and spatial characteristics. In this project, a primary real-time autonomous accident detection system has been proposed which is established on object detection deep learning in video prediction techniques. The events in video will be the accidents which are predicted several frames before the event. In this technique, 5000 accident images were labeled, and the model was classified using all the existing classes. Then, the special events were detected using object detection technique. The proposed model shows an accuracy of 92.98 and a speed of 200 images per second in video. The model can predict the event 0.92 second before accident. This model can be used in inter-city cameras with immediate analysis of meta data, and it has been piloted.


Modules


Algorithms


Software And Hardware