SwiftWeigh: a rapid calf weighing system using top-down 2D point cloud data

Context Accurate calf weight measurement is essential for effective health management and production optimisation in dairy farming. Existing non-contact methods include 2D image-based approaches, which are easy to deploy but often suffer from low precision, and 3D point cloud-based methods, which offer higher accuracy but are computationally intensive and less practical for real-time applications. Aims This study aims to develop a lightweight, accurate, and efficient calf weighing system that balances speed and precision by leveraging top-down 2D point cloud data. Methods We propose SwiftWeigh, a system that captures overhead point cloud frames using a stereo camera and projects them into 2D images while preserving essential spatial features. A convolutional neural network (CNN) is employed to detect key points and derive body measurements such as shoulder width, abdominal width, height, area, and a newly defined rump width (distance between the hip and buttock regions). These measurements are input into an ensemble regression model comprising Gradient Boosting, ElasticNet, and Ridge regression for predicting calf body weight. Key results SwiftWeigh was validated on a dataset of 100 Holstein calves, each with five frames. It achieved a root mean square error (RMSE) of 4.51 kg and a mean absolute error (MAE) of 4.04 kg, surpassing several existing methods in prediction accuracy. The system demonstrated efficient performance with minimal computational overhead. Conclusions The proposed method delivers state-of-the-art accuracy while ensuring rapid processing and low system requirements, making it suitable for real-world dairy farm applications. Implications SwiftWeigh provides a non-invasive, accurate, and scalable solution for dairy calf weight measurement. Its low-stress, real-time capability enhances on-farm monitoring, supporting better decision-making in modern dairy management systems.