Research And Implementation For The Core Technology Of High-speed Railway Face Recognition Travel

With the continuous development and breakthroughs of some subjects like artificial intelligence,machine vision,machine learning and human-computer interaction,face recognition as an importantly non-contact biometric recognition technology has important scientific significance not only in theoretical research,but also in all areas of the national economy,especially national security,public safety,and finance,it has shown great practical value and has entered the practical stage.However,the two-dimensional face images are currently used in legal identity applications and contain a limited amount of information.In actual scenes,they are susceptible to factors such as lighting,face pose changes and face expressions,which make face recognition technology facing enormous challenge.Specific manifestation in current Chinese railway transportation,high-speed rail stations have adopted a mass of real-name verification equipment based on two-dimensional face recognition.However,due to the inherent problems of two-dimensional face recognition,it is difficult to further improve the recognition accuracy in actual scenarios,and the practical level is limited,requiring a lot of manual participation.Compared with the two-dimensional face recognition technology,the three-dimensional face recognition technology performs three-dimensional modeling on the face,and uses the three-dimensional structure information and texture information to improve the accuracy of face recognition and authentication,while also having huge advantages in terms of robustness and security,so it becomes the core of the next step in the promotion and application of face recognition.However,due to the inconvenience of using the 3D face acquisition device in the actual scene and the small amount of 3D face data collected,the 3D face recognition technology is not mature enough,and the registration cannot be performed without the 3D face basic library.Application is a pain point and difficulty in the development of 3D face recognition technology.Therefore,this paper focuses on the large-scale application of 3D face recognition.Based on the specific scenes and services of Chinese railway transportation,this paper researches large-scale 3D face data collection and database construction,3D face recognition algorithms for large-scale practical applications,large-scale 3D face data registration and other issues,so several new methods and application systems are put forward to meet practical applications.The specific results mainly include:(1)A set of technical solutions for “face swiping” travel on the railway based on highprecision real-time 3D face collection and recognition technology are proposed.Aiming at the urgent needs of major national projects and the insuperable difficulties existing in 2D face recognition technology,the whole set of technical solutions and system for“face swiping” travel on railway based on high-precision real-time 3D face collection and recognition technology was first proposed and implemented.The face recognition capability of the system has been verified by practical big data experiments,and its performance greatly exceeds many domestic artificial intelligence unicorn companies.At present,the system equipment has been installed and operated in the high-speed railway line of 2022 Beijing Olympic Winter Games(Beijing-Zhangjiakou high-speed railway),Beijing South Railway Station and other places,and has achieved good results,indicating the effectiveness of the scheme and the superiority of 3D face recognition.This is a first large-scale application of 3D face recognition in the world,and its success has laid a solid foundation for the next development of 3D face recognition.(2)A 3D face reconstruction algorithm based on speckle is proposed and a miniaturized and fast 3D face camera is designed and implemented,which solves the problem of real-time3 D face database construction.The traditional 3D face collection requires strong cooperation of collectors,which has a relatively slow collection speed and is not suitable for large-scale collection.Therefore,the construction of 3D face data is limited.Aiming at the on-site applicability of 3D face cameras,in areas with high traffic and uncontrolled environment under unconscious conditions,we propose and implement a 3D face reconstruction algorithm based on multi-speckle,and we design and realize a miniaturized 3D face acquisition camera based on this algorithm,which can quickly and effectively collect and reconstruct 3D face data while maintaining high acquisition accuracy.(3)The world's first large-scale 3D face database was established.In view of the lack of 3D face database,a non-sensory acquisition system and acquisition process are realized by combining with the travel scene design of Chinese railway traffic,which improves the success rate of 3D face data acquisition.Finally,the application system architecture of high speed and high precision 3D face recognition technology is constructed,and the application verification system of high speed and high precision 3D face recognition technology for super large groups is designed.The actual operation effect of high speed and high precision 3D face recognition technology for super large group is presented.After practical operation,the world's largest 3D face database of 1.8 million people has been built up.(4)3D face recognition algorithms for large-scale practical applications is proposed,which solve the bottleneck of face recognition.The purpose of the algorithms is to reduce the restrictions on the scenes application of 3D face recognition,which can effectively avoid the dependence of 3D face recognition on the cooperation between the acquisition equipment and the collectors,and enhance the practical application value.A 3D face data preprocessing method is proposed,which is simple and fast,and is beneficial to improve the speed of the 3D face recognition.It is proposed to design a deep convolutional residual neural network based on natural selection mechanism,which makes the algorithm has higher accuracy and faster speed.A fusion algorithm based on 3D face structure information recognition and near-infrared face texture recognition is proposed to improve the accuracy of the 3D face recognition algorithm.A algorithm of recognizing 2D nearinfrared faces by 3D face structure hybrid near-infrared texture,which realizes the recognition of the 2D near-infrared face by mixing the 3D facial structure with near-infrared texture.Finally,the proposed 3D face recognition algorithm achieved excellent results in actual field tests.