Modeling of Optimal Bidirectional LSTM Based Human Motion Recognition for Virtual Reality Environment

Abstract

AbstractRecently, virtual reality (VR) has received significant attention among researchers to accomplish realistic interaction between humans and computers. Among the different application areas of VR, multimedia human computer interaction requires accurately collect and identify the motions of human body in real time. For enabling highly realistic and efficient communication among humans and computers, design of human motion recognition system becomes necessary to determine diverse complex and distinct human actions. Therefore, this study designs a new optimal bidirectional long short term memory (BiLSTM) with fully convolution network (FCN), called OBiLSTM-FCN model for human motion recognition in VR environment. The proposed OBiLSTM-FCN model comprises different processes namely feature extraction, classification, and hyperparameter optimization. Primarily, kernel based linear discriminant analysis (LDA) approach is employed as a feature extraction technique. In addition, the BiLSTM-FCN technique is applied as a recognition model to determine human motions. Finally, the Adam optimizer is applied to optimally tune the hyperparameters involved in the BiLSTM-FCN model. The performance validation of the OBiLSTM-FCN model take place and the resultant values portrayed the improved performance over the other compared methods.KeywordsVirtual realityDeep learningHuman motion recognitionHyperparameter tuningAdam optimizer