Skirting Line Annotation via Deformation Modelling

Publication Type:
Conference Proceeding
Citation:
2021
Issue Date:
2021-12-06
Full metadata record
Automating the process of wool handling has the potential to drastically improve the productivity of on-farm operations that would result in significant cost savings for wool growers. Towards this goal, we present a method to automatically extract the skirting line (\ie the separation between clean and contaminated wool) by comparing pre- and post-skirted RGB images of freshly shorn wool fleece. The intention is to provide annotation to support down-stream learning methods. Our approach detects feature correspondences then performs non-rigid outlier rejection to overcome the challenge of deformation when the wool is handled. The final alignment, and hence identification of the skirting line, is achieved through the use of a non-rigid deformation method. A controlled experiment shows, quantitatively, that our approach outperforms a rigid registration baseline. We then demonstrate the applicability to the real use case by presenting qualitative results on images of skirted fleeces collected from a wool shed.
Please use this identifier to cite or link to this item: