Cricket Shot Classification from Video Using Body Pose Landmarks

The primary focus of this study is to use deep learning techniques to classify cricket shots. Data acquisition begins by gathering video clips of different cricket shots and manually annotating them. To address imbalance in class, generated synthetic examples are used to ensure each shot type has equal representation. Advanced techniques for handling missing data points using combined object detection and 3D landmark extraction to track the batsman's move-ments are employed for efficiency. To enhance accuracy of the analysis, complex methods are used to deal with missing data. By integrating object detection methods with 3D land-mark extraction, effective tracking of batsman movements is captured. Tests with multiple machine learning models such as simple neural networks and more complicated models which could capture movement sequence over time are tested on the data. The most successful results were achieved with systems that could process video frames in both forward and reverse directions, leading to a more precise identification of cricket shots. This research illustrates the potential of machine learning techniques and how advanced models can be used to analyze and classify sports activities in video data. We contribute to the growing community and knowledge base of automated sports analytics, paving the way for future innovations in real-time action recognition and commentary generation.