VR Child Care Simulation Teaching Based on Improved EfficientDet and 3DCNN BiLSTM

Abstract

Child care work requires extremely high professionalism, and nursing staff need to be proficient in operating skills to ensure the accuracy of nursing operations. However, traditional child care practical training is limited by issues such as a shortage of teaching resources, which hinders the cultivation and improvement of nursing skills. Therefore, this study constructs a virtual human child care simulation teaching platform based on virtual reality technology. This platform integrates action detection algorithms based on improved EfficientDet, as well as action recognition algorithms based on 3D convolutional neural networks and bidirectional long short-term memory networks. The results showed that on the ActiveNet1.3 dataset, the improved EfficientDet achieved a detection speed of up to 25.14 FPS. The detection speed of the temporal recursive network model was only 15.24 FPS. On the THUMOS14 dataset, the improved EfficientDet model achieved an average detection accuracy of 96.54%, which was significantly superior to other models. Meanwhile, in the confusion matrix analysis, the action recognition model used achieved a recognition accuracy of 98.17% for posture adjustment actions, which was superior to the comparison model. In addition, the nursing learning time and effectiveness using virtual human teaching platforms were significantly better than traditional teaching methods. This indicates that the developed platform can effectively improve the effectiveness and efficiency of child care teaching, and promote the development of medical education towards a more intelligent direction.