Deep Learning Based Sports Video Classification Research

Abstract

In this paper, we study sports video classification technology based on a deep learning algorithm, using a convolutional neural network and deep learning gradient descent algorithm as the main research method to classify and regress the image features of sports video output. A sports video classification model is built to clarify the full research idea, and the results of extracted data are verified one by one with OTB target tracking dataset and specifically analyzed the accuracy and check-all rate of various types of videos, and then conclusions are drawn. The data show that the accuracy of the sports video classification test set with deep learning algorithm reaches 98.3%, and the detection rate of tennis, badminton, and table tennis all reach 92%. Soccer and basketball have a lower accuracy rate but can reach 83%. The accuracy rate of capturing detailed actions is 97.6%. The sports video classification based on deep learning accurately and effectively captures faces and actions accurately, which is important for viewers to find interesting sports video categories quickly.

Keywords

• deep learning
• sports video
• convolutional neural network
• motion capture
• sports classification.
• 68m01