Automated human body pose estimation and activity recognition in videos is still one of the challenging problems in computer vision. Generally, it is becomes a significantly difficult task in real world applications due to camera motion, cluttered background, occlusion, and scale/viewpoint/perspective variations. Moreover, the same action performed by two persons can appear to be very different. In addition, clothing, illumination and background changes can increase this dissimilarity. This project is about learning good features for automated human pose estimation and activity recognition using the broadcast video cameras in the context of sport videos. Therefore, this project contributes to constructing an automated and robust vision base human activity recognition and body pose estimation that works in real-time with respect to the current hardware resources.