In industrial farming, livestock well-being is becoming increasingly more important. Animal breeding companies are interested in enhancing the total merit index used in breeding programs. Pigs tracking and behaviour analysis plays a crucial role in breeding programs. To this end, we proposed a tracking-by-detection approach for detecting and tracking indoor farm animals for an extended period. We exploited a modified OpenPose model for the detection where the features from the input frames are extracted through EfficientNet, and the detected Keypoints are associated through a greedy optimization mechanism. Additionally, the attention mechanism is incorporated in the pose estimation framework to refine the input frames’ features maps. A bipartite graph is created for every two frames to track the animals over an extended period. The edge cost is defined by the spatial distance between the detected Keypoints of the animals in the temporal domain. We collected and annotated the customized dataset from the pig farm to train the model. The dataset and annotation will be made publicly available to help promote research in the farming industry. The proposed method is evaluated on $$AP^{OKS}$$APOKS and $$AR^{OKS}$$AROKS, and promising results are achieved.