| With the development of artificial intelligence technology,virtual rehabilitation has become an effective exercise rehabilitation therapy,and is widely accepted by medical institutions and pension institutions.However,the existing virtual rehabilitation methods have the problems of poor naturalness in human-computer interaction and unfriendly training process,which reduce the rehabilitation experience of patients and make it difficult for many patients to get the best rehabilitation treatment effect.Aiming at the above problems,this paper studies the pose recognition algorithm for virtual rehabilitation,and develops the corresponding training system.Firstly,a set of rehabilitation postures based on the combination of upper and lower limbs and trunk local postures is designed to meet the patients’ various rehabilitation needs,and the software and hardware platform of postural data acquisition is built to realize the acquisition,construction and marking of rehabilitation postural data set.Secondly,a rehabilitation posture recognition algorithm based on multi feature combination and multi-layer learning is studied.The local posture category features and skeleton geometry features are defined to describe human posture,and different features are fused.The training model is generated by using multi-layer learning method to judge different rehabilitation posture categories.The effectiveness of the algorithm is verified by using self built data sets and open data sets.Then,an attitude recognition algorithm based on depth forest is proposed.By taking multiple features as deep forest input and using a small number of samples for adaptive training,and the attitude recognition is realized,and good recognition effect is achieved.Finally,a virtual rehabilitation posture training system based on natural interaction is constructed.The system can realize automatic data acquisition of posture and movement,feature extraction and preservation,and training of defined rehabilitation posture combined with virtual reality task. |