| In the last several decades, the ability of computers retrieving, processing and communicating information have been evolving dramatically. Human are visual species and incorporate visual ability into computer. Techniques of processing visual information has also been developed and used all around the world. Video compression techniques has allow human transmit much more video data via network than ever before. The advent of CPU has made processing video data in real time possible. Industries surrounding video has also evolved dramatically, including video surveillance, film making and even personal video sharing.This paper analyzes the procedures of traditional video analytics including video receiving, video segmentation, video decoding and feature extraction and their inherent parallelism. Then this paper analyzes the requirement of this multi-way video analytics system. Then this paper analyzes the current popular distributed frameworks. Then the system architecture is analyzed in details. Finally, this paper analyzes in details the design and implementation of the front-end video feature extraction sub-cluster, central real-time alert sub-cluster and video object storage sub-cluster. Although video vendors have different standards, their general architecture is the same. This platform tries to tackle and hide the differences and provide a unified interface for vendors to implement plugins for this platform so that they will implement video big data products quickly. |
PDF Full Text Request