Automatic Cattle BCS Estimation Using Point Cloud Surface Features From a Single Top-View ToF RGB-D Camera

This is the authors' abstract. We don't add key points for this paper.

Body Condition Score (BCS) is a key indicator of nutritional status, productivity, and reproductive performance in dairy cows; however, manual assessment is subjective, labor-intensive, and unsuitable for continuous monitoring. This study proposes an automated and contactless BCS estimation framework using a single top-view time-of-flight (ToF) RGB-D camera and 3D dorsal region point cloud analysis. The cow back surface is reconstructed into a point cloud, and both edge-based geometric features and Fast Point Feature Histograms (FPFH) are integrated into a modified Dynamic Graph Convolutional Neural Network (DGCNN) for robust learning of fat-related surface patterns. By combining handcrafted 3D descriptors with graph-based deep feature learning, the proposed method significantly improves strict BCS matching performance. Experiments conducted on 95 Holstein cows across two commercial farms achieve accuracies of 69.2% at tolerance 0.0 and 94.85% at tolerance 0.25, outperforming existing 2D and depth-based approaches. The system enables scalable, objective, and non-invasive BCS monitoring, supporting proactive health interventions, optimizing feed efficiency, and enhancing transition-cow management in precision dairy farming.