32 / 2025-03-31 11:21:14

Leveraging keypoints for multi-camera tracking of Jersey cattle in calving pens

Computer Vision,Cattle Management,Multi Object Tracking,Precision Livestock Farming,Animal Welfare

3. 智慧牧场与动物福利 > （1）智慧养殖的利惠：提高动物福利与健康、环境、经济效益与劳动力

Accurate multi-camera tracking of individuals is essential to advance precision livestock management and automated behaviour monitoring, yet reliably matching indistinguishable animals such as Jersey cows across different viewpoints remains a challenge. Re-identification approaches often rely on visually distinctive features that Jersey cattle lack, leading to frequent ID mismatches and reduced tracking accuracy. Wide side-angled views introduce further challenges such as occlusion, scale variation, and perspective distortion. This study developed a multi-camera tracking framework capable of tracking nine cows across two cameras covering a 108 m² calving pen. We trained a multi-class YOLOv11-Pose model on 1035 images, distinct from the tracking dataset, that can extract four dorsal keypoints as well as classify standing and lying postures with a 98.2% accuracy. We extracted and annotated 1000 frames per camera for our tracking dataset. We utilised an improved ByteTrack, with an additional matching step incorporating keypoints, over a sliding window to yield tracklets per camera. For cross-camera matching, we computed a cost matrix by measuring the reprojection error from triangulated 3D keypoints back onto each 2D camera plane. The Hungarian algorithm matched tracklets across the two cameras, assigning each cow a consistent global ID regardless of viewpoint. Our framework achieved multi-object tracking accuracies (MOTA) of 92.43% in Camera 1 and 97.02% in Camera 2, illustrating reliable tracking for both cameras. This study underscores the viability of a multi-camera, keypoint-driven approach for tracking visually similar animals, which can be extended to larger-scale farms where minimising camera infrastructure is a critical necessity.