Key Technology Research Of Intelligent Classroom Analysis For Surveillance Video Streams

Classroom teaching interactions reflect the nature of teaching and learning to some extent since they are a complicated process of social construction of meaning and cognitive information processing.Teaching interaction behavior is a socio-emotional reaction that takes place during teacher-student contact in the classroom.It involves both explicit actions and implicit learning states(such as cognitive,social,and emotional).In addition to identifying teaching patterns,it may be used to understand teaching events and difficulties.The precise,scientific identification and modeling of classroom interaction behaviors are a hot topic in the interdisciplinary field.This study incorporates theories and practices from the disciplines of education,psychology,and artificial intelligence,in opposition to conventional approaches and tactics.It looks at the spatiotemporal feature association method for micro-expression recognition in instructional films and implements the display and analysis of emotion in the classroom teaching process.At the same time,taking the teaching video as the basic unit,a deep learning behavior recognition method that integrates spatiotemporal semantics is proposed,and the multimodal data output by the attitude estimation model is experimentally analyzed to improve the recognition accuracy of students’ classroom behavior.(1)In this study,we combine computer vision expertise with Deep Convolutional Neural Network(Deep CNN)for identifying significant areas of faces and deep learning approach for extracting optical flow features to develop a dual-network microexpression recognition model based on optical flow features.The facial region of interest(ROI)is divided based on the key points of the face,and redundant features are removed using the improved optical flow direction histogram(HOOF),after which the optical flow characteristics in ROIs are statistically analyzed.The recognition of microexpressions is subsequently accomplished using a classification method.Experimental results employing MMEW,a video-based face micro-macro expression database,show that the dual network model created in this research improves the accuracy of micro-expression identification and can more successfully perform the micro-expression recognition task.(2)In this paper,we reconstruct a multi-object pose estimation model that includes spatio-temporal semantics for different sizes and video multi-object poses while taking into consideration the time dependency and coherence between video frames.The model uses an end-to-end detection framework in the first step to identify various video objects.In order to better localize human important points,it then uses temporal signals between video frames.Moreover,it creates modular components to increase posture information and effectively modify pose estimate findings.Last but not least,employing an improved algorithm for human skeletal behavior recognition based on position estimate,student classroom activity is identified for video streams.By empirically contrasting different classifiers,the efficiency of human skeletal activity recognition with multi-object pose estimate is successfully raised.The issues of teaching interactive behavior recognition accuracy and the interpretability of teaching phenomena can be resolved by the study topics suggested in this paper.The scientific concepts used are supported by technological methods like deep learning and data analysis.The application of this research’s findings is anticipated to resolve the modeling and computation of interactive behavior in the teaching of information-based courses,and to provide theoretical and technical support for artificial intelligence to encourage the high-quality development of education and teaching.