Efficient Segment-Anything Model for Automatic Mask Region Extraction in Livestock Monitoring
Su Myat Noe, Thi Thi Zin, Pyke Tin, Ikoub Kobyashi
Abstract
This paper presents an efficient segment-anything model for automatic mask region extraction in livestock tracking. The research focuses on developing and evaluating automatic mask segmentation models for tracking black cattle. The primary contribution is a tailored extraction segmentation model for automatically extracting cattle mask regions utilizing in the livestock tracking. The methodology utilizes Segment Anything Model (SAM), Grounded SAM, Grounding Dino, YOLOv8, and DeepOCSort algorithms for detection and tracking. Experimental results demonstrate the effectiveness of the proposed approach in extracting black cattle mask regions and improving livestock tracking. Integration of YOLOv8 and DeepOCSort ensures accurate association and tracking of mask regions across frames. The findings advance livestock tracking, with applications in precision agriculture. The proposed segment-anything model serves as a valuable tool for automatic mask region extraction in foreground-background separation.