Advanced 3D Horse Reconstruction: Integrating Image-to-3D Modelling and Non-Rigid Registration for On-Barn Body Measurements.

Overview

• This research article presents an advanced method for creating precise 3D models of horses using consumer depth cameras combined with image-to-3D reconstruction and non-rigid registration techniques.
• The system aims to improve automatic horse body measurement in farm settings, facilitating better breeding management and health monitoring.

Background and Motivation

• Precision Livestock Farming (PLF) requires accurate 3D reconstructions of livestock for detailed body measurements.
• Traditional 3D reconstruction in barns is challenging due to environmental noise, hardware limitations, and complex horse anatomy.
• Variations in horse shapes can cause distortions in reconstructed models, reducing measurement accuracy.

Methodology

• Hardware Setup: Uses three consumer depth cameras positioned to capture multiple views of the horse.
• Image-to-3D Pipeline:
  • Extracts core morphological features such as torso outlines and limb ratios from images.
  • Generates an initial rough 3D model based on these core traits.
• Non-Rigid Registration:
  • Applies coarse-to-fine fitting optimization to refine the 3D model.
  • Incorporates dynamic surface feature matching weights to improve alignment with real horse anatomy.
  • Eliminates local geometric distortions caused by complex shapes or varying poses.
  • Recovers the accurate scale of the scene to match the horse’s actual size.

Evaluation and Results

• Tested on multiple equine datasets to validate performance.
• Compared with baseline and existing reconstruction algorithms.
• Key improvements include:
  • More than 50% reduction in core body measurement errors.
  • Similar reduction in Chamfer Distance (a metric quantifying surface reconstruction accuracy).
  • Over 20% increase in F-Score, representing better model fidelity.
• The refined models demonstrate higher integrity and precision in representing horse anatomical structures.

Significance and Applications

• Enables fully automatic phenotyping and body measurement of horses in natural farm environments.
• Supports detailed breeding management by providing accurate phenotypic data without manual measurement.
• Improved health surveillance through regular and precise monitoring of body condition.
• Provides a template for image-to-3D measurement pipelines that can be extended to other livestock species.
• Potential to assist large-scale, precise breeding programs by facilitating rapid, non-invasive data collection.

Conclusions

• The proposed integration of image-to-3D modelling with non-rigid registration results in high-fidelity, anatomically accurate 3D horse models.
• This method overcomes prior limitations related to environmental constraints and shape complexities.
• It reduces measurement errors substantially and enhances reconstruction quality, advancing the capabilities of precision livestock farming technologies.