Segmentation methods for a group-housed pig monitoring system

Abstract

The analysis of individual pig behavior in group-housed pigs is important for pig management. In this study, we propose two low-level segmentation methods for group-housed pigs to facilitate the video-based high-level analysis of pig behavior. In a 24-hour pig room monitoring environment where no pig is allowed to enter/leave the room during the monitored period, the previous video frame has sufficient information for separating touching-pigs in the current video frame. In this paper, we propose two methods to separate touching-pigs using the information of the previous video frame and a hybrid method for combining the segmentation results of each method. According to experimental results with the Korean pig farm data, the proposed segmentation methods based on the labeled outline/region information can provide more accurate results than widely used methods. © 2005 - Ongoing JATIT & LLS.