In the domain of human activity recognition, deep learning has emerged as a powerful tool for analyzing and classifying complex sports movements and fitness activities, offering valuable insights into athlete performance. This research focuses on the application of deep learning for human activity recognition: cricket shot classification and yoga pose detection, which are analyzed independently within the study. By integrating Mediapipe for real-time pose estimation with a hybrid 3D CNN-LSTM model, the system captures both spatial and temporal features of human movements. Mediapipe extracts key body landmarks, enabling precise analysis of cricket batting techniques and static yoga asanas. These features are processed through a 3D CNN for spatial analysis, followed by an LSTM network to capture temporal dependencies, ensuring accurate classification of activities. The system offers real-time feedback, supporting performance improvement and injury prevention. The model’s accuracy and loss metrics will demonstrate its effectiveness in recognizing dynamic sports actions and static fitness poses. This study highlights the potential of combining deep learning and real-time pose estimation for advanced, data-driven tools to enhance performance in sports and fitness.