Study And System Implementation Of Affective Virtual Reality System And Its Multi-Perspectives Automatic Annotation

Emotion,as a subjective feeling,has a great influence on people's cognition and behavior,and plays a unique and important role in interpersonal communication.Due to the important correlation between emotions and social behaviors,emotional eliciting and evaluation have great potential value in many important directions,such as health monitoring,adjuvant therapy and game design.Therefore,this paper focuses on the study of the emotional eliciting based on virtual reality and its automatic annotation.There are several prominent problems in the field of emotional eliciting and recognition.First,traditional emotional inducing materials stay in low-dimensional space.There are also limitations of low emotional eliciting efficiency and high sensitivity to external disturbances,relating to these materials.Secondly,the current subjective self-rating scale of emotional eliciting materials has a large subjective error,and it is impossible to achieve continuous annotation.At the same time,in the field of emotion recognition,there is no machine learning classification model suitable for panorama data.Regarding these problems,the following three main tasks are carried out in this paper:(1)'Affective Virtual Reality System' is established.It contains 24 virtual reality emotion-inducing scenes,which are evenly distributed in the emotional space,and the emotional three-dimensional values and emotional classification of each scene.Through the psychological emotional assessment experiment,the verification system has reached the psychological standardization of the emotion eliciting system.Through the comparison experiment with the video eliciting,it is verified that the ‘Affective Virtual Reality System' has better emotional awakening effect.(2)A dual-supervised neural network model based on lightweight neural network is proposed.Considering recognition of emotional pictures as a hybrid problem of classification and regression,based on classification and regression double loss function constraint,an emotion classification algorithm is proposed.As a result of experiments,MobileNet is selected as the underlying network.After adding a dual-supervised mechanism to network,an accuracy of 91.429% was achieved,which is higher than the comparison algorithm.(3)Based on the dual-supervised emotion classification algorithm,a multi-view fusion mechanism is introduced to realize automatic annotation of the ‘Affective Virtual Reality System'.According to intelligently annotating the six VR test scenarios,the visual three-category emotional tags of the time dimension were obtained.The system accuracy reached 92.946%,which was higher than the comparison method.The validity of the annotation system is verified by experiments.Based on virtual reality technology,this paper studies emotion eliciting from a new perspective,and provides a more immersive and realistic way of audiovisual perception emotion eliciting.Furthermore,the dual-supervised neural network emotion recognition algorithm and multi-view fusion mechanism are proposed,which realize the intelligent annotation of the virtual reality emotional system of the visual domain in time dimension and spatial dimension.