Research On Recognition Algorithms Of Rehabilitation Posture And Action Based On Multi-Rules Leaning

With the acceleration of population aging and the increase of social pressure,the number of patients with limb dysfunction is increasing.Exercise therapy as a traditional and effective rehabilitation technology,has the characteristics of low cost and reliable effect.However,the existing rehabilitation methods need to be carried out under the guidance of professional doctors.The training process is dull and tedious.It is often difficult for patients to adhere to it for a long time,which makes it difficult for many patients to get the best therapeutic effect.In view of the above questions,this paper studies the technology of automatic rehabilitation training assisted by computer program,and deeply studies the automatic recognition of posture and action in rehabilitation training.The specific research contents are as follows:First of all,put forward rehabilitation posture recognition algorithm based on learning rule.The definition of two kinds of features to describe human posture,the combination of Bagging algorithm and random subspace algorithm increases the diversity of samples and features,improve the superiority of each sub classifier classification of different samples.The rule-based learning method is used to generate classification rule set,which can be used as a discriminant rule to distinguish different rehabilitation postures in the training system.The posture data set for rehabilitation training is designed and applied to the verification of the algorithm and the implementation of the system.Then,an action recognition algorithm based on multi-rule constraints is proposed.According to the motion data obtained by Kinect,proposed rules method of positioning,the automatic continuous action is divided into several short segment action data;At the same time,the DTW(Dynamic Time Warping)algorithm is used to analyze the obtained segmentation action data,and the final matching action label is obtained,which realizes the action recognition and achieves better recognition result on the established rehabilitation action data set.Finally,the rehabilitation training system is built by combining rehabilitation posture and action recognition algorithm.Through the analysis of rehabilitation needs,training scenarios are designed.The system integrates posture and action recognition algorithm.It can automatically count the number of execution postures and movements,output the accuracy of execution actions and other evaluation indicators.It can also save the user's joint action data,and provide a variety of reference for the training process.