Multiclass Cricket Shot Classification using CNN

An acknowledgement of context-based advertising for the audience watching cricket can involve identifying the game of cricket based on the various batting shots, creating sensor-based commentary systems, and creating coaching assistants. This research identifies different cricket shot problems from a vast-scale cricket shot video dataset. The study uses a large dataset of annotated cricket shots that spans a variety of game formats and captures different shot types and player styles. Some cricket shots are the action we find most attractive in cricket videos. Those shots are often captured by cameras installed at the cricket ground's stands at both ends of the cricket pitch. To identify 4 different cricket batting shots from offline footage, this research paper presents a novel approach to classify cricket shots based on deep learning techniques. In this research, we propose a multi-stage classification pipeline for cricket shots, which combines dimensionality reduction, feature extraction, and classification techniques of cricket shots. Before classification, we have preprocessed the dataset that we prepared. The image data was extracted from the video containing around 3000 image frames. The dataset has four classes: i.e. pull shot, square drive, straight drive, and uppercut. The preprocessing steps are normalization, resize, and histogram equalization. After that a scratch model is designed and the model is trained based on the architecture of the scratch model. Finally, we have found the model's accuracy for our prepared dataset is 98.18 percent.