基于HDFS的大规模监控视频的细粒度检索

Hadoop Distributed File System (HDFS) is widely used to store massive surveillance video currently. However, there are no solutions for retrieving fine-grained surveillance video on HDFS. In this paper, we propose a novel method to retrieve surveillance video by second on HDFS. We draw lessons from small files storage problem in HDFS. Firstly we receive video segments from IP cameras. Then we merge them together and upload the merged files to HDFS. The merged file and the directory of HDFS are well designed to avoid extra in-memory index for file mapping. Experiments show that our approach can retrieve massive surveillance video on HDFS by second efficiently.