Álvaro García
Cattle body weight (BW) plays a pivotal role in beef cattle production, serving as a critical parameter for growth monitoring and informed management decisions. Traditional weighing scales have been the go-to method for BW measurement, but this approach has inherent challenges. It subjects animals to stress, involves substantial costs, and demands labor-intensive efforts. Fortunately, an innovative alternative has emerged—computer vision systems capable of extracting biometric measurements from three-dimensional (3D) images of cattle. These biometric measurements can then be integrated into complex algorithms in a cloud-based environment, enabling immediate deployment of results to predict BW, offering several advantages, including real-time monitoring, reduced stress on animals, risks to the operator, cost savings, and automation. The quality of BW prediction using computer vision hinges on factors such as the choice of biometric measurements and modeling techniques.
One of the key components of the system is to be able to accurately identify individual animals. In recent years, Radio-Frequency Identification (RFID) devices have gained traction, facilitating individual identification and traceability of livestock products. These systems, consist of RFID tags, a communication channel, a tag reader, an RFID network, and the RFID back-end, which through radio waves transmit animal information wirelessly in the form of distinctive electronic codes. However, the implementation and management of RFID systems require skilled personnel. Additionally, concerns regarding data security, including tag-content alterations and system spoofing, have limited the broader adoption of RFID technology. Conventional ear-tag-based cattle identification systems, while widely used, present challenges related to cost, manual labor for device attachment, and reliability in harsh environmental conditions. Recent innovations in individual animal recognition include methods based on cattle muzzle, retinal and iris patterns, facial recognition, and coat pattern analysis. Drone platforms equipped with deep learning capabilities now autonomously locate and identify individual cattle within geo-fenced farm areas furthering adding to the complexity and precision of these systems particularly outdoors.
Weight Estimation
The estimation of cattle weight is pivotal to optimize growth, farm income, and animal well-being. It has a profound impact on various aspects, including lactation, growth, pregnancy, fertility, and dietary calculations. Direct weighing, which involves placing each animal on an electronic or mechanical scale, offers unparalleled accuracy, it is however time-consuming and stressful for cattle. Moreover, automatic scales, while highly accurate, are expensive and not easily deployable in confined spaces or open farmland.
In contrast, indirect cattle weight estimation methods rely on assessing morphological traits like wither height, heart girth, body length, and hip width using 2D or 3D sensors. Data analysis establishes relationships between these traits and weight. While heart girth and hip width exhibit strong correlations with body weight, other characteristics such as heart girth, wither height, and body length can also contribute to accurate weight predictions. Camera-based methods, combined with automatic image analysis, offer cost-effective and efficient alternatives for cattle weight estimation. Algorithms linking body measurements features to weight are developed using image analysis and machine learning techniques. For example, Yan et al. (2019) used 2D cameras to measure withers height, body diagonal length, and body side area from images. Applying these measurements through multiple linear regression, they predicted weight with RMSE values ranging from 7.5 kg to 13.4 kg. However, it’s crucial to note that 2D camera-based systems can be influenced by camera angles and cattle posture.
3D-Based Cattle Weight Estimation
Several studies have tested the use of 3D Cameras to predict body weight in cattle. Yamashita et al. (2017) introduced a novel method that modeled calf shape using 3D data from stereo images. The technique of these researchers achieved an error rate of approximately 21.46% (around 20 kg) by averaging volumes derived from the slide division method, considering adjoining circles. Gomes et al. (2016) employed 3D cameras to measure cattle body features and revealed a strong correlation (R2= 0.967) between heart girth and body weight. Martins et al. (2020) utilized the Microsoft Kinect 3D camera to estimate body weight from lateral and dorsal perspectives. This approach achieved an R2 of 0.89 (RMSE=49.20 kg) for the lateral view and 0.96 (RMSE=26.89 kg) for the dorsal view.
Beyond live weight estimation, 3D cameras have also demonstrated precision in carcass weight prediction. Miller et al. (2019) harnessed 3D imaging technology and machine learning, particularly artificial neural networks, to forecast both live animal and carcass characteristics. Their method incorporated sixty morphological traits, including lengths, heights, widths, areas, volumes, and ratios, achieving RMSE values of 42 kg (R2 = 0.7) for live weight and 14 kg (R2 = 0.88) for carcass weight.
Recently, 3D Light Detection and Ranging (LiDAR) technology has gained prominence in remote sensing for cattle. This technology employs laser light to measure distances and create detailed three-dimensional representations of objects and environments. Huang et al. (2019) explored automatic cattle measurements, incorporating transfer learning from LiDAR sensing. They utilized LiDAR-sensed cattle point cloud datasets to extract cattle silhouettes for body measurements, with experimental results indicating a comprehensive error of body dimensions close to 2%. Sousa et al. (2018) developed a LiDAR sensor platform for estimating cattle live weight in feedlots, achieving an R2 of 0.85 and an RMSE of 8.93 kg by measuring beef cattle rump height and back view area and feeding it into an artificial neural network-based model.
Numerous studies have explored the use of body linear measurements to estimate cattle weight, with a predominant focus on morphological traits such as body width and length. However, our understanding of how live weight correlates with various factors and traits, including height, body condition, genetics, and genotype, remains somewhat elusive. Further research is needed to enhance our understanding of the accuracy and implications of live weight estimation, particularly in commercial and research contexts. Additionally, while 2D and 3D sensor-based approaches offer non-invasive means for weight estimation, addressing challenges related to environmental conditions, lighting, and cattle motion is crucial for developing practical and reliable weight estimation systems tailored to farm environments. These systems should exhibit adaptability to uneven illumination environments and consistently capture cattle contours, ultimately maximizing precision and repeatability in cattle weight estimation.
© 2025 Dellait Knowledge Center. All Rights Reserved.
