Interpretable AI architecture of machine learning algorithm for intelligent video surveillance based on fog and edge computing

Abstract

In order to address the application scenarios and technical requirements of video monitoring, the centralized data processing model represented by cloud computing has a large cost in terms of resource requirements, relies excessively on the network bandwidth of the cloud computing center, and is difficult to meet the needs of video processing in real-time and other aspects, the author proposes an interpretable AI architecture of intelligent video monitoring machine learning algorithm based on fog and edge computing. The author proposes a edge computing model suitable for video monitoring scenarios. From the three main resources of computing, network bandwidth and storage as the entry point, the system architecture is designed. After using the computing power of edge nodes to complete video preprocessing, the Docker container platform is built, and hierarchical scheduling strategy is adopted to reduce network congestion. The results showed that compared to video surveillance with added motion detection function, analyzing data from week 1 to week 3 compared to video surveillance without added motion detection function, the number of video files stored was about 1176 records per week, saving about 41.56% of storage space. This model can effectively reduce the computational, storage, and network transmission costs in video surveillance scenarios.