YOLOv3 deep learning and solid-state LiDAR for teats detection

Real-time detection of teats is an important part of automatic cow medicine bath robot. However, the environment of dairy farms is complex, with changes in light, masking, shadows, teats shape color and so on. These make teats detection very challenging. The traditional detection method is laser-based and relies on the original edge detection technology, but the recognition error of this method is still very large. In this paper, we propose a combination of the YOLOV3 model and LiDAR to detect teats in complex environments. Firstly, the images of teats are initially collected. Then, the model effect is tested in the test set at the end of the training and on the fake teats model in the lab. The model can detect teats in 0.415s and the mAP@0.5 is 94.9%, which can provide real-time detection of teats in the milking room. Moreover, the YOLOV3 model can effectively provide teats detection under occlusion conditions and can be applied in the in the actual environment of ranch.