Camera calibration is often a necessary requirement for advanced computer vision-based sports analytics, such as player tracking, augmented reality, or movement-based performance analysis. In this paper, a novel end-to-end framework for automatic camera calibration is proposed. The specific novel aspect concerns the intentional design for sports with minimal to no pre-labelled camera calibration data. The proposed framework is tailored for sports with strictly defined rectangular sports fields, i.e. soccer, badminton, rugby, or volleyball, and delineates the design of deep learning-based algorithms to solve camera calibration from single-feed broadcast footage. The deep learning paradigms described in this framework are search-based and prediction-based camera calibration. Prediction-based camera calibration, however, is preferred due to the ease of data labelling and scalability of the solution. The application of the framework is verified by an implementation thereof for badminton, a sport without a public calibration dataset. For ease of use, the popular pose estimation algorithm by Ultralytics is utilised, namely YOLO11x-pose. The selected algorithm is validated independently with respect to two of the established benchmark datasets for soccer, namely the World Cup 2014 dataset and the time-sequence World Cup dataset. Comparative results were achieved with median intersection over union accuracies.