Assess the theoretical benefits of keypoint-based matching

Determine whether keypoint-skeleton-based matching or IoU matching supplemented with pig-orientation information can improve BMCTrack-d when accurate ground-truth keypoint skeletons are available.

Background

BMCTrack-d uses IoU-based matching to associate pig detections across the frame history used by its temporal prediction consistency check. The paper also investigates two alternatives: matching keypoint skeletons with object keypoint similarity and supplementing IoU matching with orientation derived from keypoints.

Neither alternative improves performance with the pose estimates used in the study, partly because occasional pose inversions disrupt matching. The authors note that assessing the theoretical value of these approaches would require ground-truth keypoint skeletons, which were unavailable, leaving their potential benefit unresolved.

References

To study the theoretical benefits of these methods, ground truth keypoint skeletons are required, which were not available for the data in this study. This evaluation is left for future work.

— BMCTrack-d: Pig re-identification and tracking via back marks in challenging camera settings  (2609.03463 - Brunner et al., 3 Sep 2026) in Appendix B, Section “Extended optimal configuration study”