Research And Application Of Students Head Motion Recognition In Class

Head motion recognition is an important part in the fields of computer vision,artificial intelligence and pattern recognition.Its applications involve video surveillance,safe driving,and machine manipulation for people with mobility impairments.It is an important application of head motion recognition technology in classroom environment that realizing the recognition of students' head motion directly from the video data collected in the classroom for automatically judging the students' learning status.And it is of great significance to optimize teachers' teaching methods and improve students' learning quality by head motion recognition.With the rapid development of information technology and the popularization of intelligent hardware devices,the depth sensor provides a more accurate data source for the research of head motion recognition by virtue of its advantages of obtaining object space information and reducing the influence of illumination and other factors.Now,there are some problems of head motion recognition of students in the classroom environment.On the one hand,the illumination of the classroom environment are constantly changing during the day and the head motion recognition methods that extract facial features based on RGB face images lose spatial information,which is easily disturbed by environmental factors such as illumination,and is difficult to achieve ideal recognition effect.On the other hand,a large amount of student head motion data will be generated in a forty minutes classroom environment,which poses a challenge to the head motion recognition algorithm to improve recognition accuracy under large-scale data.Aiming at the above problems,this paper takes the head motion recognition algorithm of students in class as the research content,and researches and realizes the head motion recognition algorithm based on lie group feature and the head motion recognition algorithm based on Inception module.The main research work of this subject is as follows.(1)Aiming at the problem of using RGB facial images to extract facial features losing spatial information which is susceptible to illumination,this work proposes a head motion recognition algorithm based on lie group features,and uses lie group features to represent spatiotemporal context information of head motion.Among them,3D coordinates of facial key points are used as input,lie group features of each pair of key segments between adjacent frames are extracted to represent space-time information,and support vector machine of RBF kernel is introduced to identify head motion.The experiment shows that the head motion characteristics represented by liegroup can effectively represent the spatial-temporal context information,and the accuracy of the head motion recognition method based on lie group is improved by4.2% compared with that based on facial features without the introduction of depth information in the common dataset.In addition,it also got 73.63% accuracy in the classroom dataset,achieving a good performance.(2)Aiming at the problem that the traditional method is not accurate enough for the recognition of massive head motion data in the classroom environment,this work explores the effective fusion of deep learning and artificial features in the field of head motion recognition,and proposes an Inception module based on the lie group feature.Normalizing 3D data and video with inconsistent sequence lengths into vector patterns of the same dimension,and using the lie group feature set as the input of the convolutional neural network to design a deep learning network framework to more fully utilize the lie group features with higher refinement.Experiments show that this method can effectively express the head motion information under large-scale data to ensure the recognition accuracy rate.The accuracy rate of the head motion recognition method on the common dataset is increased by 4% ~ 10%.In addition,the accuracy rate of 76.10% was achieved on the classroom dataset.Because the size of the customized classroom dataset was relatively small compared with the size of the public dataset,the result was slightly lower than the comparison method.(3)Combined with the research results of this work,the prototype system of head motion recognition in classroom was designed and realized.Users can use Kinect to collect student head motion data in the classroom and store it in the database through the system.Besides,the user can independently select the test samples in the classroom and after the system recognizes the students' head motion the result is displayed on the page to verify the effectiveness of the proposed head motion recognition method.The paper has 31 figures,19 tables and 91 references.