Figure 2.
Complete steps of proposed methodology. The pig postures are identified and localized by using a deep-learning object detection model, then the simple online real-time tracking with deep association (Deep-SORT) algorithm tracks each pig by associating the corresponding detected posture, and, finally, the detected pigs’ postures with time information are analyzed to score the posture activity occurring in an hour. All the modules are integrated to provide an end-to-end posture scoring.