Research And Application Of Fitness Action Evaluation Based On Posture Estimation

With the completion of a well-off society in an all-round way,people’s material and cultural aspects have been satisfied to a certain extent,and fitness exercise has become a way for people to achieve spiritual satisfaction.Proper fitness exercises can help people get rid of the physical burden brought by work,strengthen people’s physical quality,and lay a solid foundation for people to enjoy a better life.At present,most fitness motion analysis systems need to wear wearable sensors in key parts of the human body to realize the purpose of fitness motion monitoring.This method costs too much and will cause certain impact on users during exercise.In order to realize low-cost and efficient fitness monitoring,this paper proposes a fitness motion evaluation method based on human posture estimation,and designs a fitness motion evaluation system based on human posture estimation.Experiments show that the fitness motion evaluation method proposed in this paper not only has good performance in the prediction accuracy of human posture estimation,but also has excellent real-time performance.The main research contents of this paper are as follows:(1)A lightweight human posture estimation network model based on high resolution network is proposed.This model uses the method of lightweight network model design,referring to the high resolution network structure design,using two lightweight modules Ghost module and Sandglass module instead of conventional convolution,this way to redesign the network basic module can greatly reduce the number of network model parameters and computational complexity;In addition,the lightweight ECA attention module is integrated to ensure better prediction accuracy of the network model.Subsequently,COCO open data set and MPII open data set were used for training and verification,and the proposed network model was compared with the current popular models in human posture estimation.The results show that the improved model proposed in this paper has a higher prediction accuracy for key points of people,while reducing the number of parameters and computational complexity of the network model.The purpose of lightweight network model is realized.(2)A fitness motion similarity evaluation method based on lightweight human pose estimation algorithm is proposed.Firstly,the dynamic time normalization algorithm was introduced to align the frame between the standard fitness action video and the fitness action video to be evaluated,in order to reduce the negative impact of time series difference on the fitness action similarity.Then,a fitness action data set suitable for fitness analysis is constructed,and the proposed lightweight human posture estimation algorithm is used to retrain on the data set,and the trained model is processed to get the fitness action sequence.Finally,the weighted average method of Euclidean distance and cosine similarity is used to evaluate the obtained movement sequence and the standard movement sequence to judge whether the fitness movement is standardized.(3)Design a fitness movement evaluation system based on posture estimation.The system has realized the application of the lightweight human posture estimation network model in the actual scene.Users can choose the video to practice.After the completion of the practice,the system can generate the action score report for users to view,so that users can correct the fitness movements.The system can achieve low cost and high efficiency fitness exercise,help users to achieve independent fitness exercise.The final experimental results show that the research content and the work done in this paper meet the current social needs to some extent.The fitness movement evaluation system can also be applied to family fitness and campus exercise and other scenes,providing a convenient and effective auxiliary training platform for people who love fitness and exercise.