A Scalable Target Indexing and Retrieval System for Massive Video Data Processing based on Elasticsearch and Hadoop
Qiaojin Guo, Jie Hu, Zhong‐Yan Liang · 2024
With the rapidly development of video surveillance techniques, more and more high resolution video files are generated. However, traditional video surveillance system cannot utilize the massive data efficiently. This paper proposes a scalable video processing and target retrieval system based on Hadoop and Elasticsearch. The imported massive videos can be parallel processed with transcode, target detection and feature extraction modules. These video processing modules are packaged and provide RPC services through thrift. With the design of decoupling, other detection and feature extraction algorithms can be easily integrated in our application to improve the speed and performance. We also build an easy-to-use application for users to manage the processing procedure, monitor the processing status, retrieve detected targets with defined rules and analysis the relationships of targets to discover underlying connections between different targets with time and geo-spatial interactions.