Research On Basketball Technology Action Recognition Based On Deep Learning

With the development of deep learning and artificial intelligence theory,the analysis based on various human behavior data sets has shown explosive development,and the motion recognition technology based on image frames is constantly being introduced and new,making video behavior motion recognition feasible..In basketball technical action videos,technical actions have obvious characteristics.Such videos on short video platforms have relatively fixed scenes and single scenes in sports videos.Therefore,intelligent classification of basketball technical action videos has a relatively prominent advantage.However,basketball technical action recognition also has many challenges,mainly including how to effectively use continuous image frames with strong correlation.In addition,there are many and complicated basketball techniques,and there are certain difficulties in how to filter out representative basketball techniques.In response to the above-mentioned problems,this paper first determines the concept of technical movements through physical literature research.By studying the practical application of technical actions in professional competitions and teaching videos from media authors on short video platforms,the basic elements of basketball technical action videos are analyzed.And through the basketball action data set related literature research,analyze the characteristics of the data set,refer to the badminton,table tennis technical action data set collection method based on the establishment of the basketball technical action data set.Finally,a dual-resolution 3D convolutional neural network architecture is proposed.Two image inputs of different resolution 3D convolutional neural networks are performed on the basketball technical action data set.Then feature vector extracted by the dual-resolution network are feature fused to perform SVM classification experiments.The experimental results show that the algorithm flow designed in this paper is effective in action recognition on the basketball technical action data set established in this paper.There are two main points of innovation in this paper: 1.The establishment of a video basketball technical action data set.2.Combined with target detection to generate low-resolution image input,dual-resolution 3D convolutional neural network architecture action recognition.