基于头尾定位的群养猪运动轨迹追踪

发布时间：2018-05-15 21:43

本文选题：算法 + 图像识别　；参考：《农业工程学报》2017年02期

【摘要】：猪的头/尾位置直观反映了猪的进食、饮水、争斗、追逐等日常活动。从群养猪俯视视频中有效分割粘连的猪个体,找出猪的头/尾部,并以头/尾坐标实现较精准的运动轨迹追踪有着较大的难度。该研究采用改进分水岭分割算法分割视频图像帧中的粘连猪个体;对分割后的猪体提取头/尾轮廓,分别用类Hough聚类和圆度识别算法识别每头猪的头/尾,用运动趋势算法修正头/尾识别的误差,生成以头/尾部为定位坐标的运动轨迹。运算结果和人工标记对比证明类Hough聚类和圆度识别算法的头尾识别正确率分别为71.79%和79.67%;经过运动趋势修正后,以头部为定位坐标生成的运动轨迹与人工标记生成运动轨迹吻合良好;对比头/尾轨迹和质心轨迹可以发现,头/尾轨迹能够更多获取猪个体和群体活动、运动信息。该研究对于实现自动记录和分析猪个体和群体的活动行为提供新的思路和方法。
[Abstract]:The head / tail position of the pig directly reflects the daily activities of the pig, such as feeding, drinking water, fighting and chasing. It is difficult to find out the head / tail of the pig and track the track accurately by looking down the video from the video of the herd pig to segment the conglutinated pig effectively and to find out the head / tail of the pig. In this study, an improved watershed segmentation algorithm is used to segment the conglutinated pig in the video frame, and the head / tail contour is extracted from the segmented pig body, and the head / tail of each pig is identified by Hough clustering and roundness recognition, respectively. The motion trend algorithm is used to correct the error of the head / tail recognition and to generate the moving track with the head / tail as the positioning coordinate. The results show that the correct rates of Hough clustering and roundness recognition algorithms are 71.79% and 79.67%, respectively. By comparing the head / tail trajectory and the centroid trajectory, it can be found that the head / tail trajectory can obtain more information about the movement of individual and colony. This study provides a new way to automatically record and analyze the behavior of pig individuals and populations.
【作者单位】：华中农业大学工学院;生猪健康养殖协同创新中心;华中农业大学动物科技学院动物医学院;
【基金】：“十三五”国家重点研发计划项目(2016YFD0500506) 湖北省自然科学基金(2014CFB317) 现代农业技术体系(CARS-36)
【分类号】：S828;TP391.41

【相似文献】