Research On Intelligent Traffic Video Processing Framework Based On Deep Learning

The massive deployments of transportation surveillance are producing large-scale video data in a rapid speed,which demands high quality of surveillance videos.At the same time,the traditional video processing systems can hardly handle the new trend of massive and high-definition video.The traditional video management systems could provide video storage and management under a lower level,regardless of information about the relevance of video data and events,let alone the predict trends,making it difficult to support decisions based in video data of a higher level and offer dynamic scalability both in software and hardware system.In addition,the requirements of real-time large-scale video data analysis online and batch processing off-line show up.The growth of cloud computing and big data technology brings a new direction for video data parallel processing.Deep learning can make the calculation model with multiple processing levels to learn the representation of abstract data with multi levels,which guarantees the intelligent processing of big data.This paper presents a general cloud-based architecture and platform using deep learning as a method to provide solutions for intelligent analysis and storage of video data.We have implemented this architecture using both Hadoop platform and Storm platform,which are typical offline batch processing cloud platform and online real-time processing cloud platform,respectively.The proposed architecture can handle continual surveillance video data effectively,where real-time analysis,batch processing,distributed storage and cloud services are seamlessly integrated to meet the requirements of video data processing and management.The evaluations show that the proposed approach is efficient in terms of performance,storage and fault tolerance.