Analysis And Application Of Volleyball Technical Movements Based On Deep Learning

Video-based human action recognition has become one of the hot topics in the field of artificial intelligence vision applications in recent years.With the development of deep learning and artificial intelligence theory,the classification of different human behavior and motion datasets is also emerging one after another.The frame recognition technology is also constantly innovating and improving,which makes the video behavior dynamic recognition technology have a certain possibility.In the volleyball basic skill action video,the basic skill action includes a total of six major items,and each has its own distinctive characteristics;due to the relatively abundant data resources of the volleyball technical video on the Internet platform,and the target characters in the original video are relatively stable,the action scenes are relatively simple,so using artificial intelligence technology to distinguish the categories of sports actions in volleyball videos has obvious advantages.However,the current UCF video datasets in the field of volleyball are limited to spiking actions,with a single action category.Other volleyball basic technical action datasets have not yet been released or lack corresponding normative samples,and there is currently a lack of professional assistance in China.The commercial volleyball video analysis system for training cannot meet the needs of schools and amateur training.Aiming at the above problems,this paper studies volleyball action recognition and system design based on the application of deep learning to sports.The main work and innovations of this paper are as follows:(1)This paper is based on the basic technical movements that often appear in volleyball training and competitions,combined with the characteristics of each movement,and refers to the collection methods of technical movement data sets such as badminton and basketball.Before intercepting these four volleyball basic technical action data sets,the data set includes 500 training sets and 500 test sets,with a total of 17653 experimental data,the video resolution is roughly determined to be 640*480,and the number of frames per second is continuous interception.16 frames.After experimental verification,the data set is suitable for the algorithm used in this paper and has a good matching effect.(2)The article proposes an optimization innovation for the training model of a neural network(CNN-LSTM)mixed with a convolutional neural network(CNN)and a long short-term memory network(LSTM).By removing the fully connected layer of the convolutional neural network,the convolutional layer is directly connected to the LSTM network after mean pooling,and the VOLO target detection algorithm is used in the convolutional layer of the CNN network instead of the sliding window to extract the video features,effectively The training efficiency of the model is improved;then the average value of the features obtained after the extraction of the convolutional layer is input into the corresponding LSTM network,and the LSTM network structure of three hidden layers is designed to obtain the time series signal of the video and count different time points.The average value of the LSTM output is used to transfer the mixed characteristics to the Softmax classifier for final classification.By comparing the recognition results with other deep learning training models,the accuracy of model recognition is effectively improved while ensuring the training efficiency.The experimental results show that the optimized training recognition algorithm designed in this paper is effective for the action recognition of the volleyball technical action dataset established in this paper.(3)According to the design and optimization of the action recognition algorithm and the experimental exploration results,this paper designs a volleyball video dynamic recognition auxiliary training analysis system based on deep learning.The design and development goals of the system,the feasibility of the system implementation and the system The architecture and each part of the functional modules are introduced in detail.The system can identify the basic technical movements in volleyball and then apply it to technical specifications and auxiliary player training.The experimental verification and data investigation show that the feasibility of the system design and the actual selection value are in line with the goals of the final research.The technical innovations of this paper mainly include the following three points:1.Create different types of video volleyball technical action data sets,and combine the target detection algorithm to extract action features to generate a cropped frame-based volleyball technical action data set.2.The optimization and improvement of the human video action recognition method based on CNN and LSTM network is proposed.While realizing multi-scale feature extraction,the optimized hybrid neural network has more accurate recognition performance than the current mainstream recognition network,and The training efficiency of the model is greatly improved.3.Propose a design scheme of a volleyball video action recognition training assistance system based on the algorithm in this paper.The system combines the functions of algorithm analysis and data analysis,and can provide users with services such as standardizing volleyball basic actions,training action interpretation and analysis,and the system The design cost is low,the operation is simple,and the operation efficiency is high.