Tiny Face Recognition In Information Educational Video Stream

With the progress of science and technology,artificial intelligence has been developing rapidly covering various aspects of our daily life.Meanwhile,face recognition has become one of the research hotspots in the relative field for its huge application prospects as a prominent representative.So far,face recognition has shown great advantages in the fields of finance,security and military but only few applications in educational research.As the development of information education is growing up gradually,educational video stream,one of valuable resources in class,is an important basis for analyzing students' activity and concentration and it also helps evaluate and improve the quality of the class.Attitude detection and emotion analysis can be conducted on educational videos for further evaluation while such computations must rely on the sequence of students' faces achieved from video streams.Obviously,face recognition is the foundation of all further research and occupies an important position in the analysis.Under such scenario,this paper extends the classical face recognition problem to real-time teaching environment.The main contribution of this paper are as follows:Based on laboratory projects,this paper collects a large quantity of class videos from multiple subjects and multiple schools.To apply face detection and recognition technologies to educational environment,this paper establishes a large-scale face database,which not only offers data samples in this research but also lays foundations for further emotion analysis.After detailed survey over latest face recognition technologies and deep study on relative deep learning progress,this paper proposes a specific face recognition algorithm targeting the teaching scenario where faces are tiny and highly occluded.This algorithm consists three steps:face detection,face tracking and face validation.This paper improves the network structure based on state-of-the-art deep learning technology to guarantee high accuracy and robustness.According to the characteristics of real classrooms,a tracking algorithm is proposed based on locations which greatly improves the efficiency of the algorithm.