ABNORMAL EVENT SUMMARIZATION IN VIDEO SURVEILLANCE USING HIERARCHICAL RECURRENT NEURAL NETWORK

Abstract

Roadways play a vital role among the various transportation modes in the world for people. Even in developing countries, death occurs every four minutes cause of road accidents which is a deep concern to humanity. Intelligent Transportation system, Vehicle behaviour analysis, road accident detection, and event summarization are the great interest fields of research. From the surveillance videos, providing the summarization of abnormal events is quite complex. To solve this issue, a technique called conceptual video summarization is presented which enables the speed of accident events visualizing from the videos in a stack. In this proposal, the video summarization is carried out using a Hierarchical-Recurrent Neural Network which is used to summarize the abnormal events in the video surveillance. The primary aim of the suggested video resuming method is to summarise and display the anomalous event portions of the video to the user. This work exploits the dependency of long temporal among the frames and the computations are less when compared to the traditional Recurrent Neural Network. The experimental results show that the proposed system provides higher accuracy when compared with state-of-the-art techniques.