Video Tracking to Monitor Turkey Welfare

Although turkey production is important in the United States, few studies have focused on turkey welfare, partly because of the lack of non-invasive and automated techniques for detecting changes in turkey welfare. Disease can pose major threats to turkey welfare and human health. In this paper, we propose a novel approach for detecting and tracking turkeys in video as the first steps to monitor turkey welfare. A self-trained object detection model is used to identify turkeys in each frame of the video, and a modified object tracker is used to predict the location of each turkey in the next frame. Hand-crafted features are developed to better handle occlusion and to improve tracking accuracy. Our method demonstrates promising results when evaluated on a turkey video dataset in terms of precision, success, and size consistency.