基于多特征融合的粒子滤波生猪采食行为跟踪

发布时间：2018-05-25 22:02

本文选题：采食 + 跟踪　；参考：《农业工程学报》2017年S1期

【摘要】：针对中国养猪业规模化、集约化迅猛发展过程中,人工观察监测记录生猪生长情况需损耗大量人力和物力,得到数据误差大的问题,该文提出将颜色特征与目标轮廓形心特征融合,基于粒子滤波算法实现生猪采食行为跟踪,当目标跟踪矩形框中心坐标和跟踪目标轮廓形心坐标之间的横坐标偏差大于跟踪目标轮廓横坐标方向的最大值与最小值的差的一半时,或其之间的纵坐标偏差大于跟踪目标轮廓纵坐标方向的最大值与最小值的差一半时,对基于颜色特征粒子滤波算法得到的跟踪矩形框的中心坐标进行二次修正,提高了目标生猪跟踪的可靠性和鲁棒性;通过对比试验,结果表明:该方法能够对目标生猪的采食行为进行自动跟踪、记录和分析,记录的目标生猪一天内的采食次数和采食时间与人工记录结果基本相同,有效跟踪平均精度为93.4%。
[Abstract]:In view of the rapid development of pig industry in China, manual observation and monitoring and recording of pig growth need a lot of manpower and material resources to get the problem of large data error. In this paper, the color feature and the centroid feature of the target contour are fused to track the feeding behavior of live pigs based on particle filter algorithm. When the deviation between the center coordinates of the target tracking rectangular frame and the centroid coordinates of the contour tracking object is greater than half of the difference between the maximum value and the minimum value in the direction of tracking the contour of the target, When the deviation of the longitudinal coordinates between them is greater than half of the maximum and minimum values of the longitudinal coordinates of the target contour, the center coordinates of the tracking rectangular frames based on the color feature particle filter algorithm are modified twice. The reliability and robustness of the target pig tracking are improved, and the results show that the method can automatically track, record and analyze the feeding behavior of the target live pig. The feed intake times and feeding time of the target pigs recorded in one day were basically the same as those recorded by manual records, and the average effective tracking accuracy was 93.4%.
【作者单位】：中国农业大学信息与电气工程学院;同方股份有限公司;
【基金】：国家高技术研究发展计划(863计划)资助项目(2013AA102306) “十二五”国家科技计划课题(2014BAD08B05) 山东省自主创新资助项目(2014XGA13054)
【分类号】：S828;TP391.41

【参考文献】