基于点云边缘平滑和生物特征的鸡只体尺估测

吕恩利; 曾伯阳; 曾志雄; 彭玉平; 何欣源; 谢伯铭; 刘妍华

doi:10.7671/j.issn.1001-411X.202206041

基于点云边缘平滑和生物特征的鸡只体尺估测

1.
华南农业大学工程学院, 广东广州 510642
2.
岭南现代农业科学与技术广东省实验室茂名分中心, 广东茂名 525000

基金项目: 国家自然科学基金(31971806)；广东省重点领域研发计划(2019B020225001)；茂名实验室双聘团队科技研究专项(2021TDQ002)；广东省畜禽疫病防治研究重点实验室开放课题(H2019515)；广州市基础与应用基础研究项目(202102020870)；广州迦恩科技有限公司横向课题(H2020033)

详细信息

作者简介:
吕恩利，副教授，博士，主要从事智能农业技术与设备研究，E-mail: enlilv@scau.edu.cn

通讯作者:
刘妍华，副教授，博士，主要从事智能畜牧设备研究，E-mail: cynthial@scau.edu.cn

中图分类号: S828
计量
- 文章访问数: 125
- HTML全文浏览量: 6
- PDF下载量: 17
出版历程
- 收稿日期: 2022-06-26
- 网络出版日期: 2023-09-03
- 发布日期: 2023-05-23
- 刊出日期: 2023-07-09

Estimation of chicken body size based on point cloud edge smoothing and biometric features

1.
College of Engineering, South China Agricultural University, Guangzhou 510642, China
2.
Maoming Branch, Guandong Laboratory for Lingnan Modern Agriculture, Maoming 525000, China

摘要

摘要:
目的
针对使用深度相机的鸡只体尺估测中，鸡只点云边缘抖动、羽毛冗余、特征点提取难的问题，本文提出一种结合点云边缘平滑和基于生物特征的特征点提取方法用于鸡只多部位体尺估测。
方法
首先，通过直通滤波、统计滤波等方法对点云进行预处理，减少背景和噪点对目标的影响；其次，通过点云的空间变化约束边缘，采用连续多帧序列变化平滑边缘，减少边缘抖动对体尺测点提取的干扰；再次，对处理后的点云进行生物特征分析，结合基于邻域分析的边缘算法，融合RGB图像采用Canny边缘检测、霍夫变换等方法提取特征点；最后，依据特征点估测胸宽、半潜水长和胫长体尺。
结果
试验结果表明，估测的胸宽平均误差为6.64%，胫长平均误差为5.93%，半潜水长平均误差为3.34%，平均每帧图像计算体尺耗时8.8 s。
结论
本文算法可为鸡只体尺测量提供技术参考。
- 点云 /
- 鸡 /
- 体尺 /
- 边缘平滑 /
- 生物特征
Abstract:
Objective
To address the issues of edge jitter in chicken point clouds, feather redundancy and challenging feature point extraction in chicken body size estimation using depth cameras, this paper proposes a method combining point cloud edge smoothing and biometric-based feature point extraction for mult-position estimation of chicken body size.
Method
Firstly, the point cloud was preprocessed by direct filtering, statistical filtering and other methods to reduce the impact of background and noise on the target. Secondly, the edge was constrained by the spatial change of point cloud, and the edge was smoothed by continuous multi-frame sequence changes, so as to reduce the interference of edge jitter on the extraction of body measurement points. Thirdly, the biological characteristics of the processed point cloud were analyzed. Combined with the edge algorithm based on neighborhood analysis, the RGB image was fused and the feature points were extracted by Canny edge detection, Hough transform and other methods. Finally, the chest width, semi diving length and tibial length were estimated according to the feature points.
Result
The test results showed that the average error of estimated chest width was 6.64%, the average error of tibial length was 5.93%, and the average error of semi diving length was 3.34%. The average calculation time of body size per frame image was 8.8 s.
Conclusion
The algorithm of this paper can provide a technical reference for chicken body size measurement.
- Point cloud /
- Chicken /
- Body size /
- Edge smoothing /
- Biological characteristic

HTML全文

图 1 RealsenseD435实物示意图

1：左红外相机； 2：红外点射投射器； 3：右红外相机； 4：RGB相机

Figure 1. Physical sketch of RealsenseD435

1: Left infrared camera; 2: Infrared spot projector; 3: Right infrared camera; 4: RGB camera

下载: 全尺寸图片幻灯片

图 2 数据采集平台示意图

1：种鸡测定站；2：俯视相机；3：侧视相机；4：料筒

Figure 2. Schematic diagram of data acquisition platform

1: Breeding chicken testing station; 2: Overhead camera; 3: Side view camera; 4: Material barrel

下载: 全尺寸图片幻灯片

图 3 鸡只的3个体尺示意图

Figure 3. Schematic diagram of three body scales of chicken

下载: 全尺寸图片幻灯片

图 4 相机与点云变换原理示意图

m(u,v)为图像坐标系下的任意坐标点；O′为图像的中心点；M(X_w, Y_w, Z_w)为世界坐标系下的三维点云点

Figure 4. Schematic diagram of camera and point cloud transformation

m(u,v) is any coordinate point in the image coordinate system; O′ is the center point of the image; M(X_w, Y_w, Z_w) is a three-dimensional point cloud point in the world coordinate system

下载: 全尺寸图片幻灯片

图 5 预处理效果示意图

Figure 5. Schematic diagram of pretreatment effect

下载: 全尺寸图片幻灯片

图 6 连续帧边缘剪影示意图

Figure 6. Edge silhouette of continuous frames

下载: 全尺寸图片幻灯片

图 7 半潜水长人工测量点示意图

Figure 7. Schematic diagram of the manual measurement point for semi submersible length

下载: 全尺寸图片幻灯片

图 8 点云数统计图

Figure 8. Statistical chart of point cloud number

下载: 全尺寸图片幻灯片

图 9 点云拐点示意图

Figure 9. Schematic diagram of turning point of point cloud

下载: 全尺寸图片幻灯片

图 10 边缘算法原理图

Figure 10. Schematic diagram of edge algorithm

下载: 全尺寸图片幻灯片

图 11 剔除尾羽点云边缘示意图

Figure 11. Schematic diagram of point cloud edge with tail feather removed

下载: 全尺寸图片幻灯片

图 12 上轮廓点云图

Figure 12. Upper contour of point cloud

下载: 全尺寸图片幻灯片

图 13 胸宽人工测量示意图

Figure 13. Schematic diagram of manual measurement of chest width

下载: 全尺寸图片幻灯片

图 14 点云胸宽示意图

Figure 14. Schematic diagram of point cloud chest width

下载: 全尺寸图片幻灯片

图 15 胫长关系示意图

Figure 15. Schematic diagram of tibia length relationship

下载: 全尺寸图片幻灯片

图 16 点云边缘平滑

Figure 16. Point cloud edge smoothing

下载: 全尺寸图片幻灯片

图 17 不同边界算法效果

图a~c的边界点数分别为800、648和679

Figure 17. Results of different boundary algorithms

The boundary points of figure a−c are 800, 648 and 679, respectively

下载: 全尺寸图片幻灯片

表 1 相机指标数据

Table 1 Camera index data

目标距离/m Target distance	Z轴准确率/% Z axis accuracy	填充率/% Fill-rate	平面误差/% Plane fit error	平面拟合误差¹⁾/像素 Subpixel RMS error	健康评价²⁾ Health-check (HLC)
0.20	99.76	100	0.26	0.03	0.04
0.25	99.67	100	0.17	0.03	0.05
0.30	99.55	100	0.29	0.03	0.04
1) 平面拟合误差表示局部平面拟合到深度值的噪声占据的像素数，当其小于0.1时认为噪声不影响相机状态；2) HLC用于衡量相机深度性能，当HLC<0.25时，认为相机的深度性能处于较好状态　1) Sub-pixel RMS error indicates the number of pixels occupied by the noise fitting the local plane to the depth value, and if it is less than 0.1, it is considered that the noise does not affect the camera state; 2) The HLC is used to measure the depth performance of the camera, and if HLC is less than 0.25, the depth performance of the camera is considered to be in a good state

下载: 导出CSV

表 2 鸡只体尺数据

Table 2 Body size data of chicken

序号 No.	估测值/cm Estimated value			测量值/cm Measurement value			t/s
序号 No.	胸宽 Chest width	胫长 Shank length	半潜水长 Half-diving depth	胸宽 Chest width	胫长 Shank length	半潜水长 Half-diving depth	t/s
1	8.11	7.58	36.46	9.2	7.2	38.5	8.1
2	7.51	8.53	33.53	7.6	7.7	33.9	8.3
3	7.17	7.80	35.54	7.6	7.6	35.4	13.2
4	6.87	7.57	34.12	7.3	7.9	33.7	7.2
5	7.86	7.57	32.61	8.6	8.1	35.7	7.1

下载: 导出CSV

参考文献(24)

[1]	沈明霞, 丁奇安, 陈佳, 等. 信息感知技术在畜禽养殖中的研究进展[J]. 南京农业大学学报, 2022, 45(5): 1072-1085.
[2]	JOHNSTONA M, EDWARDSD S. Welfare implications of identification of cattle by ear tags[J]. The Veterinary Record, 1996, 138(25): 612-614. doi: 10.1136/vr.138.25.612
[3]	ZARAGOZA O, ENRIQUE L. Evaluation of the accuracy of simple body measurements for live weight prediction in growing-finishing pigs[D]. Champion: University of Illinois at Urban-Champagin, 2011: 1-2.
[4]	陈瑶, 张云伟, 雷金辉, 等. 基于视觉的四足动物骨架及行走步态特征提取方法[J]. 电子测量与仪器学报, 2022, 36(2): 68-77.
[5]	彭孝东, 时磊, 何静, 等. 消费级RGB-D相机在农业领域应用现状与发展趋势[J]. 中国农机化学报, 2022, 43(4): 206-215.
[6]	张婧婧, 李勇伟. 基于机器视觉的马体尺测量系统设计与研究[J]. 计算机测量与控制, 2017, 25(12): 17-20.
[7]	尹令, 蔡更元, 田绪红, 等. 多视角深度相机的猪体三维点云重构及体尺测量[J]. 农业工程学报, 2019, 35(23): 201-208.
[8]	刘同海, 滕光辉, 付为森, 等. 基于机器视觉的猪体体尺测点提取算法与应用[J]. 农业工程学报, 2013, 29(2): 161-168.
[9]	马学磊, 薛河儒, 周艳青, 等. 基于改进区域生长法的羊体点云分割及体尺参数测量[J]. 中国农业大学学报, 2020, 25(3): 99-105.
[10]	赵建敏, 赵成, 夏海光. 基于Kinect v4的牛体尺测量方法[J]. 计算机应用, 2022, 42(5): 1598-1606.
[11]	戚超. 基于深度相机和机器视觉技术的鸡胴体质量在线分级系统[D]. 南京: 南京农业大学, 2019.
[12]	AYDIN A. Using 3D vision camera system to automatically assess the level of inactivity in broiler chickens[J]. Computers and Electronics in Agriculture, 2017, 135: 4-10. doi: 10.1016/j.compag.2017.01.024
[13]	MORTENSEN A K, LISOUSKI P, AHRENDT P. Weight prediction of broiler chickens using 3D computer vision[J]. Computers and Electronics in Agriculture, 2016, 123: 319-326. doi: 10.1016/j.compag.2016.03.011
[14]	赵振华, 黎寿丰, 黄华云, 等. 2个品系肉鸡体尺性状和屠宰性状的典型相关分析[J]. 福建农林大学学报(自然科学版), 2012, 41(2): 166-169.
[15]	郭宏杰, 马德新. 计算机视觉技术在农业领域的应用[J]. 乡村科技, 2021, 12(14): 119-120.
[16]	董静超, 陈家雪, 李华, 等. 清远麻鸡慢羽系屠宰性状与体尺性状的主成分分析[J]. 江苏农业科学, 2022, 50(5): 151-155.
[17]	姜宏正, 杨德智, 马中华, 等. 儋州鸡体尺性状全基因组关联分析[J]. 中国畜牧兽医, 2022, 49(2): 598-607.
[18]	文杰. 肉鸡种业的昨天、今天和明天[J]. 中国畜牧业, 2021(17): 27-30.
[19]	XU R, TYASI T L, QIN N, et al. Assessment of relationship between body weight and body measurement traits of indigenous Chinese Dagu chickens using path analysis[J]. Indian Journal of Animal Research, 2017: 588-593.
[20]	FU J, WANG S, LU Y, et al. Kinect-like depth denoising[C]//2012 IEEE International Symposium on Circuits and Systems (ISCAS). Seoul: IEEE, 2012: 512-515.
[21]	ZHANG Y Z. A flexible new technique for camera calibration[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(11): 1330-1334. doi: 10.1109/34.888718
[22]	中华人民共和国农业部. 家禽生产性能名词术语和度量统计方法: NY/T 823—2004[S]. 北京: 中国农业出版社, 2009: 3-4.
[23]	LIG X, LIUX L, MAY F, et al. Body size measurement and live body weight estimation for pigs basedon back surface point clouds[J]. Biosystems Engineering, 2022, 218: 10-22. doi: 10.1016/j.biosystemseng.2022.03.014
[24]	XU G, RUAN P. Point cloud boundary detection in preprocessor of optical-mechanical integrated simulation[J]. Infrared and Laser Engineering, 2016, 45(4): 0428001. doi: 10.3788/irla201645.0428001