Automated soccer event detection and highlight generation for short and long views

The computer vision field has wide applications in various areas, including sports. Almost all sports events have been exploiting the best features. Sports videos are structure-based, and due to this characteristic, these videos can be categorized into interesting and non-interesting events. Identifying the view of the video and separating important events from non-interesting events is challenging. However, correct view detection can lead to correct event detection. Various researchers have proposed many strategies for detecting events in sports videos. Significant research shows some gaps while generating highlights due to limited work available in the long or short view; there is still a need for some automated methods. The main purpose of this research work is to detect key events by normalizing the data, extracting features, fusing those features, and classifying them. In this research, events are detected by classification. A new dataset is created for research purposes. The benchmark dataset is divided into two subsets, which can be used separately or as part of a larger dataset. A proposed novel approach for event highlight generation in long and short view is presented using a fusion of AlexNet and VGGNet architectures to explore the model's efficiency in the context of accurate highlight generation deep learning models fused with handcrafted. AlexNet and VGG16 pre-trained deep CNN models are applied to extract deep prior features, which are then combined with HOG to improve the results. The proposed methodology undergoes evaluation on a dataset that we created, resulting in an accuracy of 99.6%.