摘要
为提高基于机器视觉的猪体体尺测量研究中的图像利用率和体尺测量效率,以长白猪和大白猪为研究对象,基于Kinect相机获取的猪体视频数据,提出了一种猪体理想姿态检测算法。该算法利用最小外接矩形法调整猪体为水平方向;利用投影法和差分法识别头部和尾部位置,通过头部边界标记法判断是否耳部缺失;利用骨骼化算法结合霍夫变换算法检测猪体头部是否歪斜。在此基础上,设计了猪体体尺测量算法。针对养殖场获取的103组视频数据、俯视和侧视各52 016帧图像,进行了理想姿态检测及体尺测量。结果表明,检测出理想姿态2 592帧、漏报432帧、误报0帧,误报率较低;每帧图像的体长偏差与本组体长均值小于2. 3%,组内理想姿态帧之间差异较小,一致性较好;体宽测量的平均精确度为95. 5%,体高测量的平均精确度为96. 3%,体长测量的平均精确度为97. 3%,测量的平均准确度较高。本研究成果应用于基于机器视觉的猪体体尺测量,可提高图像利用率和体尺测量效率。
In the research of pig body size measurement based on machine vision, the demand for posture of pig is high. Image frames of ideal posture need manual selection, which limits the application of body size measurement based on machine vision. To improve the image utilization rate and the efficiency of pig body measurement, pig species of Landrace and Large White were taken as the researches object. Kinect camera was used to obtain video data of pigs. An algorithm was proposed which can detect the posture of pig in the image. In this algorithm, the minimum external rectangles were computed to adjust the level of the pig’s body. Head and tail positions were identified by projection and difference methods. Boundary signature was used to determine whether part of the ears was missing. Image skeleton algorithm and Hough transform algorithm were applied to judge whether the pig body was skewed. On this basis, algorithms for measuring pig body size were designed. The top view and side view of video had 52 016 frames, respectively. These frames of 103 sets of video data were tested by the posture detection algorithm and body size measuring algorithm. And 2 592 frames of ideal posture frames were screened out. It produced high false negatives (432 frames) and very low false positives (0 frames). The results showed that the absolute deviation of body length was small. The body length deviation of each frame was less than 2.3%, and the consistency of the measurement results was high. The average accuracy of body width was 95.5%, the average accuracy of body height was 96.3%, and the average accuracy of body length was 97.3%. This research can be used to measure pig body size based on machine vision to improve measurement efficiency.
作者
司永胜
安露露
刘刚
李保成
SI Yongsheng;AN Lulu;LIU Gang;LI Baocheng(College of Information Science and Technology, Hebei Agricultural University, Baoding 071001, China;Key Laboratory of Modern Precision Agriculture System Integration Research, Ministry of Education,China Agricultural University, Beijing 100083, China;Key Laboratory of Agricultural Information Acquisition Technology, Ministry of Agriculture,China Agricultural University, Beijing 100083, China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2019年第1期58-65,共8页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家重点研发计划项目(2016YFD0700202)
河北农业大学理工基金项目(ZD201702)
