Post Event Investigation of Multi-stream Video Data Utilizing Hadoop Cluster

Abstract

Rapid advancement in technology and in-expensive camera has raised the necessity of monitoring systems for surveillance applications. As a result data acquired from numerous cameras deployed for surveillance is tremendous. When an event is triggered then, manually investigating such a massive data is a complex task. Thus it is essential to explore an approach that, can store massive multi-stream video data as well as, process them to find useful information. To address the challenge of storing and processing multi-stream video data, we have used Hadoop, which has grown into a leading computing model for data intensive applications. In this paper we propose a novel technique for performing post event investigation on stored surveillance video data. Our algorithm stores video data in HDFS in such a way that it efficiently identifies the location of data from HDFS based on the time of occurrence of event and perform further processing. To prove efficiency of our proposed work, we have performed event detection in the video based on the time period provided by the user. In order to estimate the performance of our approach, we evaluated the storage and processing of video data by varying (i) pixel resolution of video frame (ii) size of video data (iii) number of reducers (workers) executing the task (iv) the number of nodes in the cluster. The proposed framework efficiently achieve speed up of 5.9 for large files of 1024X1024 pixel resolution video frames thus makes it appropriate for the feasible practical deployment in any applications.