Scalable Video Analysis Framework

Abstract

Nowadays, with the rapid development of the surveillance system and the Internet, a large amount of video data is collected and generated everyday. To handle such high volume video data, the distributed system and related programming paradigms like Apache Hadoop/Spark are a promising solution. Unfortunately, we will encounter many challenges when utilizing current big data framework to process videos. Firstly, there is no existing video IO interface. Second, existing video analytic applications are not compatible with Spark/Hadoop framework. Finally, current researches and papers are focusing on specific video applications, there lacks of a systematic guideline for users to port their program to spark. In this talk, we will address this issues and introduce our proposed framework and guidelines. Several design considerations and features will also be elaborated.