Sports Video Data Classification using Yolov5 Model

The machine learning along with computer vision has helped widely in classification of sports videos. Deep learning techniques are also being used to perform research in this domain. The video frames are the most significant components of the sports classification system. There are many models that have been used to classify sports videos. The objective of this work is to develop a classifier for different sports activity recognition using video data with high accuracy and probability. Sports dataset has been used from Kaggle [10]. This framework has been created for applications related to sports, object detection, game identification, recognition, analysis, players’ tracking, and performance. This framework consists of various intermediary processes. At first, preprocessing has been carried out in this framework by converting input sports video into video frames. Then skeletonization has been carried out using computer vision. Finally, feature extraction and classification has been done using the Yolov5 model that has been trained with the dataset. Accuracy and loss graphs have been generated using prediction metrics for this model to evaluate the accuracy of this framework. The accuracy of the framework is more than 90% according to the results we have obtained.