Research on the Design and Application of Intelligent Teaching and Training System for College Basketball Based on 3D Motion Recognition Technology

Abstract

In this paper, for the problems of traditional 2D convolutional networks in dealing with action recognition in video, a dual-resolution 3D-CNN action recognition network is induced to initialize 3D convolutional weight parameters by using 2D weight parameters of ImageNet, and the weight files are used as the parameters of the model to perform feature extraction on the sequence of technical action frames respectively. After feature extraction, collect the feature vectors that can effectively describe the human body movements, classify the feature vectors with the help of the kernel function in the Support Vector Machine (SVM), and finally determine the technical architecture of the basketball auxiliary training system, and complete the design of the intelligent teaching and training system for basketball in colleges and universities. After analyzing the method of the present paper, it can be seen that the recognition error of the basketball player’s running gait is 3.47%, which accurately reflects basketball sports training footwork trajectory to achieve the effect of intelligent guidance. In addition, the upper and lower limb movement recognition effect based on CNN+SVM is obvious, and the average accuracy rate of its upper and lower limb movements is 93.17% and 98.66%, respectively, which well meets the needs of basketball teaching in colleges and universities. In terms of students’ basketball skills, there is a significant difference between this paper’s system and traditional teaching (P<0.05), and this paper’s system has a higher priority for improving the quality of college basketball teaching than traditional teaching.

Keywords

• 3d-cnn
• svm
• action recognition technology
• basketball teaching system
• 97c70