基于点云的生猪体尺参数自动提取方法

Automatic Extraction of Pig Body Size Parameters Based on Point Cloud

摘要:

家畜体尺参数在动物优选育种、健康监测及性能遗传研究等领域中具有至关重要的地位，体尺参数的自动测量是数字化农业的重要研究方向。本文依托双深度相机测量系统，采集生猪点云数据，开发一种能够自动识别和提取生猪特征点、特征面以及体尺参数的算法。鉴于生猪点云数据常伴有严重的噪声问题，与传统单一滤波方法相比，本文提出一种多滤波器叠加的滤波算法；为实现生猪特征点和特征面的自动检测，与单一视角计算特征点不同，提出一种基于侧视和俯视轮廓线凹凸性的特征点自动提取算法和基于躯干剖面积变化趋势的特征面自动提取算法；针对现有体尺参数种类有限且拟合精度不足的问题，成功地从生猪点云数据中自动提取17个体尺参数和12种常见形状因子。为验证算法有效性，对20头生猪的200组点云数据进行了详尽分析，结果表明：平均滤波总误差为3.84%，在原始点云精度达到0.036mm的条件下，生猪体尺参数的平均测量误差仅为2.46%。此外，对生猪的17个几何参数和12个形状因子进行了基于主成分的权重分析，探讨了不同体尺参数和形状因子在生猪几何形态分析中的权重。本研究为动物三维表型参数的高通量测量提供了一种有效方法，不仅提升了测量的精度和效率，还为动物优选育种和健康监测等领域的深入研究提供了有力支持。

Abstract:

The body dimension parameters of livestock play a crucial role in various fields such as animal selection breeding, health monitoring, and genetic studies of performance. The automatic measurement of these parameters represents an important research direction in digital agriculture. Relied on a dual-depth camera measurement system to collect point cloud data of live pigs and aimed to develop an algorithm capable of automatically identifying and extracting pig feature points, feature surfaces, and body dimension parameters. Given that pig point cloud data often contain significant noise issues, a filtering algorithm with multiple filters stacked was innovatively proposed, in contrast to traditional single-filter methods. To achieve automatic detection of pig feature points and surfaces, an algorithm for automatically extracting feature points was proposed based on the convexity and concavity of side and top-view contour lines, and an algorithm for automatically extracting feature surfaces based on the trend of trunk cross-sectional area changes, differing from approaches that calculated feature points from a single perspective. Addressing the limitations of existing body dimension parameters in terms of variety and fitting accuracy, totally 17 body dimension parameters and 12 common shape factors from pig point cloud data were successfully automatically extracted. To validate the effectiveness of the algorithm, a detailed analysis of 200 sets of point cloud data from 20 live pigs was conducted. Experimental results showed that the average total filtering error was 3.84%, and under the condition that the original point cloud accuracy reached 0.036mm, the average measurement error of pig body dimension parameters was only 2.46%. Additionally, a principal component-based weight analysis on the 17 geometric parameters and 12 shape factors of pigs was conducted, discussing the weights of different body dimension parameters and shape factors in pig geometric morphology analysis. In summary, the research result can provide an effective method for high-throughput measurement of animal three-dimensional phenotypic parameters, not only improving measurement accuracy and efficiency but also providing strong support for in-depth research in fields such as animal selection breeding and health monitoring.