基于骨架特征的奶牛肢体分解方法研究

发布时间：2018-07-10 13:36

本文选题：奶牛 + 肢体分解　；参考：《西北农林科技大学》2017年硕士论文

【摘要】：我国养殖业逐步向规模化方向发展,规模化养殖对饲养管理方式提出更严格的要求,在养殖过程中广泛应用信息技术以提高养殖效率和健康管理水平已成为必然趋势。视频分析技术能够对动物的行为进行自动监测和理解,是提高养殖管理信息化水平的重要手段,且越来越多地应用于奶牛的精细养殖。国内外学者已经在这方面做了大量研究工作,但多是针对奶牛的整体进行分析,而奶牛作为一种多关节的大型动物,头部、脖子、前肢、后肢和尾巴等均是通过关节区分开的局部肢体,通过奶牛各个肢体部位可获取更加精准的奶牛运动细节信息,是奶牛姿态检测、行为分析和理解的基础。为实现奶牛头部、脖子、躯干、前肢、后肢和尾巴的精确分解,本文研究并提出一种基于骨架特征的奶牛肢体分解方法。通过在奶牛场布设Kinect传感器,获取奶牛的深度图像数据,研究基于深度图像的奶牛目标提取方法、奶牛骨架提取方法及基于骨架特征的奶牛肢体分解方法。本文主要工作和结论如下:(1)提出了综合利用深度阈值、图像形态学及中值滤波的奶牛目标提取方法。针对奶牛在养殖场中受复杂背景和光照影响而难于精确提取的问题,综合设备的布设环境和性价比等因素,选择Kinect获取奶牛的深度图像数据,将Kinect获取的深度数据转换为文本数据,利用深度图像中的深度阈值分割及图像形态学变换进行奶牛目标提取,并用中值滤波对图像进行去噪处理,从复杂背景中有效提取出目标奶牛。试验结果表明,本文提取的奶牛目标与人工提取奶牛的重叠率为95.62%。(2)借鉴Choi骨架宽度约束理论下的骨架点判定准则,提取出奶牛骨架并进行减枝处理。通过对目前主要骨架提取方式进行分析,综合考虑骨架的连通性、单像素性及高效性等因素,利用Choi定义的骨架宽度约束理论下的骨架点判定准则提取奶牛骨架,用离散曲线演化模型对骨架进行剪枝处理,简化后的骨架能反映奶牛完整的轮廓特征。(3)提出基于骨架特征的奶牛肢体分解方法。该方法提取奶牛骨架上含有重要位置信息的骨架分叉点,以骨架分叉点依据设定约束条件,生成奶牛肢体分解的分割线,并利用形状视觉显著度和分割线优先级准则对生成的分割线进行优化处理,实现了奶牛肢体的分解。试验结果表明,在显著性阈值取2.5时,奶牛各个肢体分解平均正确率为95.09%,且对较难分割的尾部正确率达95.51%;对仰头、正常行走、微低头和低头体态下的肢体分解平均正确率分别为95.18%、95.00%、94.85%和96.23%,可实现不同体态奶牛的高精度分解。
[Abstract]:The aquaculture industry in our country is gradually developing towards a large scale. It has become an inevitable trend to apply information technology widely to improve the efficiency and the level of health management of aquaculture. Video analysis technology can automatically monitor and understand the behavior of animals, which is an important means to improve the level of information management of breeding, and is more and more used in the fine breeding of cows. Scholars at home and abroad have done a lot of research in this area, but most of them are based on the analysis of the whole of the cow, and the cow, as a large animal with multiple joints, has a head, neck, forelimb, Hindlimb and tail are local limbs separated by joint, and more accurate details of cow motion can be obtained from each limb of cow, which is the basis of posture detection, behavior analysis and understanding. In order to decompose the head, neck, trunk, forelimb, hind limb and tail accurately, a decomposition method based on skeleton features is proposed in this paper. By using Kinect sensor in dairy farm to obtain the depth image data of dairy cow, the methods of dairy cow target extraction based on depth image, cow skeleton extraction method and cow limb decomposition method based on skeleton feature are studied. The main work and conclusions of this paper are as follows: (1) A method of dairy cow target extraction using depth threshold, image morphology and median filter is proposed. In order to solve the problem that it is difficult to extract accurately the dairy cattle under the influence of complex background and illumination, the Kinect is selected to obtain the depth image data of dairy cow, because of such factors as the setting environment of the equipment and the ratio of performance to price. The depth data obtained by Kinect are converted into text data, and the depth threshold segmentation and morphological transformation of the depth image are used to extract the dairy cow target, and the median filter is used to Denoise the image. The target cows were extracted from complex background. The experimental results show that the overlap rate between the objective extraction and artificial extraction is 95.622. (2) drawing lessons from Choi's skeleton width constraint theory, the cow skeleton is extracted and treated with branch reduction. Based on the analysis of the main skeleton extraction methods and considering the connectivity, single pixel and efficiency of skeleton, the skeleton of dairy cattle was extracted by using the criterion of skeleton point decision based on Choi's definition of skeleton width constraint theory. The discrete curve evolution model is used to prune the skeleton, and the simplified skeleton can reflect the complete contour features of dairy cattle. (3) A decomposition method based on skeleton feature is proposed. In this method, the skeleton bifurcation points with important position information are extracted from the dairy cow skeleton, and the splitting lines of the decomposition of the cow limbs are generated by the skeleton bifurcation points according to the constraint conditions set by the skeleton bifurcation points. The shape visual saliency and the priority criterion of the split-line are used to optimize the generated split-line, and the decomposition of the cow limb is realized. The results showed that at the significant threshold of 2.5, the average correct rate of decomposition was 95.09 for each limb, and 95.51 for the difficult tail, and 95.51 for the head up. The average correct rate of limb decomposition was 95.18% and 96.23%, respectively, which could be used to decompose dairy cows with different posture.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S823;TP391.41

【参考文献】