A Study on the 3D Reconstruction Strategy of a Sheep Body Based on a Kinect v2 Depth Camera Array

Area of Science:

• Agricultural Engineering
• Computer Vision
• Animal Science

Background:

• Manual measurement of sheep body dimensions causes stress and is prone to errors.
• Existing non-contact methods face challenges with environmental factors and data noise, limiting practical application in livestock management.

Purpose of the Study:

• To develop a robust non-contact data acquisition system and a 3D point cloud reconstruction method for sheep bodies.
• To provide reliable reference data for sheep breeding, fattening, and related research.

Main Methods:

• A data acquisition system using Kinect v2 cameras, a sheep passage, and a restraining pen was designed for multi-perspective data collection.
• A 3D point cloud reconstruction method was implemented using C++ and the Point Cloud Library (PCL), incorporating noise filtering (pass-through, statistical, RANSAC) and downsampling (conditional voxel filtering).
• Coarse and fine registration were performed using RANSAC and Iterative Closest Point (ICP) algorithms to enhance accuracy and robustness.

Main Results:

• The system successfully collected 135 sets of point cloud data from 20 sheep.
• The 3D reconstruction method achieved a low body length reconstruction error of 0.79% compared to actual measurements.
• The developed method demonstrated robustness and accuracy in reconstructing sheep bodies.

Conclusions:

• The proposed non-contact 3D reconstruction system and method effectively address the limitations of manual measurement and noisy data collection.
• This technology offers a reliable solution for obtaining precise sheep body data, supporting advancements in animal breeding and management.
• The study validates the potential of 3D point cloud reconstruction for practical applications in the livestock industry.