Research On Extraction And Recognition Technology Of Teacher Behavior With The Characteristics Of Teaching Behavior

Optimizing the teaching service form with artificial intelligence technology has become a new trend of "Artificial intelligence+Education".Feedback of teachers’ behavior,as a kind of service for teaching process evaluation,is an important method for teachers to improve their teaching ability.Teacher behavior analysis aims to mine the valuable information from the behavior data,and finally promotes teachers to optimize the teaching process,teaching results and teaching environment.Teacher behavior extraction and recognition is the first step of teacher behavior analysis.However,the videobased teacher behavior extraction and recognition based on video is still adopting semi-automatic mode,after the video scene is split at a fixed time,teacher behavior extraction and recognition based on video relies too much on the subjective experience of the analyst,which cannot guarantee the scientificity of teacher behavior identification and analysis.Furthermore,it is very complicated and time-consuming to manually extract and identify the teacher’s behavior from massive teaching videos.In terms of artificial intelligence technology,existing behavior extraction and recognition technologies in video scenes mainly use single features of images and sounds for behavior extraction and recognition,which have achieved good results in the database of daily life scenes,while without considering the complete teaching meaning of teachers’ behaviors in teaching scenes and videos,these methods have the problem of low robustness.Due to the massive teacher’s behaviors in the teaching scenes,these methods are not enough to meet the requirements of efficient and accurate extracting and identifying teachers’ behaviors in massive teaching scene videos.Focusing on these issues,this paper analyzed the explicit forms and characteristics of teaching behaviors in teaching scenarios,extracted the visual,auditory and semantic multi-dimensional features of the subject in the teaching scene,explored the meaningful teaching behavior units,combining with the teaching behavior rules,studied new methods of teacher behavior extraction and recognition in the teaching scene videos.The specific research on teacher behavior technology and extraction technology consists of four parts:The first is the analysis of the characteristics of teaching behavior by multi-dimensional feature fusion.Effective methods of extracting and identify ing teachers’ behaviors in teaching scenes inevitably depend on intensive study and understanding of the characteristics and ways of teachers’ and students’teaching behaviors in teaching scenes.This paper analyzed the characteristic information of teaching behavior in classroom teaching from multi-dimensional perspectives,designed a teaching behavior analysis model of multi-dimensional feature fusion which includes four core elements of "the analysis coding system of teaching behavior based on visual and auditory features,the auditory feature recognition of teaching behavior,the visual feature recognition of teaching behavior,and the visual display of the process data of teaching analysis".Among them,the coding system of teaching behavior analysis based on visual and auditory features extracts the corresponding visual and auditory features of the behavior subject under different behavior features for the first time,and at the same time increases the fine-grained classification of teaching behaviors.Furthermore,three practical approaches of "visual feature predominating,auditory feature predominating and fusion feature predominating" are proposed to clarify the relationship between intelligent analysis elements of teaching behavior.And taking the practical approach of "visual feature predominating" as an example,this paper preliminarily analyzed some complete classes and extracted the visual features of teaching behavior in time dimension.Secondly,in view of the problems that the current research on teacher behavior extraction relies too much on a single modal information and ignores the time attribute and subject characteristics of teacher behavior,according to the intelligent analysis model of teaching behavior based on multidimensional feature fusion,and combining the auditory features,semantic features and visual features of teacher’s behaviors,a method of extracting teachers’ behaviors from classroom videos based on multimodal feature fusion was proposed.Firstly,a time-series teacher behavior dataset is constructed,and then the proposed teacher-student scene fragments separation algorithm based on auditory features is used to obtain the teacher and student scene fragments in the teaching scene video and the teacher semantic text information in the corresponding time.Finally,the proposed teaching behavior extraction algorithm based on visual and auditory feature fusion is used to fuse the spatio-temporal features and text features of teaching videos,and extracts the corresponding start time,end time and teaching behavior types of teaching behaviors in classroom teaching.Thirdly,aiming at the problem that the existing behavior recognition algorithm cannot perceive the identity of the teacher in the teaching scene due to the complexity of the classroom environment and fine-grained recognition of the teacher’s behavior with the meaning of teaching function.Based on the fine-grained classification of teaching behaviors,this paper proposed a teacher behavior recognition method in video based on teacher sets.Firstly,by observing a large number of teaching videos,we summarized and put forward the educational rules which is named as "teacher set",that is,the spatial region of the video of the whole class where teachers should exist.Based on this,the algorithm of teacher set identification and extraction(Teacher-set IE algorithm)is studied to identify the teacher in the teaching video,and reduces the interference factors of classroom background.Then,a behavior recognition netw ork based on 3D bilinear pooling(3D BP-TBR)is designed,which fused the features of the last layer of 3D neural convolutional network and the cross layer of bilinear pooling.Therefore,the fine-grained identification of teacher behavior types in actual teaching scenarios is realized.Fourthly,the prototype system for extracting and recognizing teachers’ behaviors is developed,which integrates the characteristics of teaching behaviors.The system mainly includes four modules:teaching behavior analysis coding display module based on visual and auditory features,video preprocessing module of actual classroom teaching scenes,teacher behavior extraction module based on multi-modal fusion and teacher behavior recognition module based on teacher set,which can provide a better analysis tool for the activities of the intelligent analysis of teaching behavior such as "one excellent course for every teacher,one excellent teacher for every course","teacher training" and so on.