Studies On Migrated Computing For Face Recognition In Metro AFC System

With the continuous development of the national economy and the rapid progress of urbanization,the urban rail transit system has achieved rapid development,and it is playing an increasingly important role in urban public transportation.In recent years,face recognition,as an important biometric identification method,has been commercially used in many scenarios on a large scale,and the urban rail transit automatic fare collection system is also vigorously developing the application of face recognition technology.Because the automatic fare collection system has high professional requirements for face recognition accuracy,real-time performance,interoperability,and reliability,its face recognition computing architecture needs to be specifically designed.To this end,the paper has carried out research on the facial recognition migration calculation technology of the urban rail transit ticketing system.The paper completed the work as follows(1)Propose and design a multi-engine migration calculation model for face recognition.In order to make the system compatible with the interoperability of multiple algorithm vendors and multiple face recognition engines,according to the deployment position of the face recognition function,the two typical computing architectures of "fat end + light cloud" and "thin end + heavy cloud" are targeted.Respectively designed migrating computing modes to achieve interoperability between multiple engines.Aiming at the functional division of cloud platforms and terminals,a cloud interaction protocol that supports interoperability and corresponding functional interfaces are designed,so that the face recognition engine on the cloud platform can support different types of terminal access and realize the migration calculation from the terminal to the cloud platform.And the migration calculation within the cloud platform.Related experiments verify the effectiveness of the program.(2)Designed a face recognition migration computing architecture that supports network degradation.In order to improve the reliability of the system,the end-side face recognition module can still provide a stable face recognition pass-through service under the coordination of cloud-free platform computing,and complete the migration of the face recognition function from the cloud platform to the terminal.In addition,according to the system cluster composed of the end-side terminal array,the migration calculation within the terminal is realized during the degraded use of the network.(3)Aiming at the ticket checking scene of rail transit gates,a prototype system of face recognition passing gates based on FL computing architecture is designed.The face recognition terminal is based on the Hi Silicon Hi3516DV300 development platform and neural network reasoning framework for software module development.The face recognition cloud platform is based on the Ubuntu system and Caffe framework for the development of corresponding software modules.Finally,the system functions and migration modes are simulated through the actual scene.The validity of the calculation was verified.Last,the dissertation is summarized and the future problems worth further study are prospected.