PCL Researchers Win ICME 2019 Best Paper Award
Date: 2019-08-13 Source:
The 2019 IEEE International Conference on Multimedia and Expo (ICME) was successfully held in Shanghai from July 8th to 12th, 2019. The paper entitled “Towards Digital Retina in Smart Cities: A Model Generation, Utilization and Communication Paradigm” by the research team under the leadership of Academician Wen Gao from AI Research Center of Peng Cheng Laboratory (PCL) and National Engineering Laboratory for Video Technology at the Peking University has won the Best Paper Award. ICME is the flagship annual conference of IEEE in the field of multimedia. It received more than 1000 papers from 36 countries or regions this year, with an acceptance rate of about 31%.
The team of Academician Wen Gao from PCL is dedicated to the research of digital retina computing architecture. They learn from the mechanism of human retina to realize real-time convergence of features, video retrieval on demand, definable front-end software and other emerging functions. Focusing on digital retina computing, the wining paper proposes a multi-stream collaborative computing model for the first time, which is composed of video stream, feature stream and model stream. The model improves the visual pathway of City Brain, effectively overcomes the problems such as model reuse and bottleneck in the transmission of "end-edge-cloud" collaborative computing, and provides new ideas for the large-scale convergence, storage and computing of video data.
The collaborative research was part of the Intelligent Transportation Project of Cloud Brain Open Source Platform and Intelligent Application Project at PCL, of which Prof. Lingyu Duan is the project investigator. Aiming at major national strategic demand, the project’s goal is to establish the computing theory, method and technical system of digital retina. They have carried out intensive study on the bottleneck of large-scale video features computing in intelligent traffic scenario; gradually implement on "cloud brain" platform of PCL the video monitoring with a scale of 10 billion, efficient indexing of traffic data, real-time detection of incidents and validation of causal inference technology. By taking advantage of the collective mechanism of video stream, feature stream and model stream, the team has developed the correlation understanding of video objects with a scale of 10 million and feature analysis engine of knowledge evolution, and performed the application demonstration in the intelligent transportation system of the key cities. Through the innovation of basic theoretical methods such as video, feature and model collaborative computing under the digital retina framework, this project enhances the international academic prestige and influence of China in basic theory of AI. Through the construction of the application demonstration system, the application of AI provides orderly guidance and causal inference for the smart travel and other application fields.