Estimating body weight of pigs from posture analysis using a depth camera.

Area of Science:

• Agricultural Engineering
• Animal Science
• Computer Vision

Background:

• Accurate estimation of pig body weight (BW) is crucial for livestock management and economic assessment.
• Traditional methods for BW estimation can be invasive or labor-intensive.
• Noninvasive methods using computer vision show promise but can be affected by animal posture.

Purpose of the Study:

• To develop and validate a noninvasive method for estimating pig BW using a low-cost depth camera.
• To investigate the influence of pig posture on BW estimation accuracy.
• To propose an adjusted volume calculation that accounts for posture variations.

Main Methods:

• Utilized a Kinect v2 depth camera to capture 738 point clouds from 150 pigs.
• Calculated pig 'volume' from point clouds and assessed its correlation with BW.
• Quantitatively evaluated pig posture using seven calculated posture angles based on the extracted 'spine'.
• Developed an 'adjusted volume' by correcting for posture-induced estimation errors.

Main Results:

• A high correlation was found between calculated pig volume and BW.
• Posture angles, particularly head position height, correlated with BW estimation errors.
• The proposed adjusted volume model achieved a Mean Absolute Percentage Error (MAPE) of 4.87% and a Root Mean Square Percentage Error (RMSPE) of 6.13%.
• The accuracy was comparable to models using only data with ideal pig postures.

Conclusions:

• A noninvasive, posture-aware method for estimating pig BW using depth imaging is feasible.
• Accounting for pig posture significantly improves the accuracy of volume-based BW estimation.
• This method offers a cost-effective and efficient alternative for livestock monitoring.