Design of running posture identification system based on CenterNet
Inorder to improve the current situation that the public running posture is generally not standardized,this paper proposed a running posture identification system based on CenterNet.Firstly,a dataset was created by taking screenshots and photos,and was cleaned,annotated,and analyzed the dataset to eliminate data-independent information and simplify the data.Secondly,the CenterNet algorithm model was improved by introducing multi-scale channel attention mechanism and adding cross star deformation convolution to transform the action image into digital information and feature vector.Then,KNN(K-nearest neighbors)algorithm was used to classify the running posture types.Finally,compared with the classical model scheme,the effectiveness of the improved CenterNet algorithm identification system was verified.The experimental results show that the accuracy and recall rates of the improved CenterNet model are improved,and the parameter quantity and calculation amounts are reduced.It can provide timely and accurate feedback on most bad postures,and can also effectively help running enthusiasts find problems,thereby improving running posture,improving exercise efficiency,and preventing injuries.
万方数据
维普
