圈舍群养形式下基于视频的猪只行为智能监测算法研究

项目名称： 圈舍群养形式下基于视频的猪只行为智能监测算法研究

项目编号： No.61503187

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 李泊

作者单位： 南京农业大学

项目金额： 20万元

中文摘要： 基于监控视频实现猪只行为的智能监测将成为养猪业信息化发展的必然趋势。本项目旨在研究圈舍群养环境下猪只行为的视频监测算法，包括猪只检测、姿态估计、行为分类等研究内容，力图借助先进的计算机视觉技术提高规模化养猪的行为观测水平，实现猪只关键行为的自动观测和分类。首先，研究基于身体轮廓与头部纹理特征相融合的猪只检测器模型，克服现有猪只定位方法的局限性，能够充分利用目标的外观特性实现群体环境下猪只的准确定位。然后，研究基于部位组合的姿态估计方法，利用概率图模型描述猪只的结构特征，实现猪只部位组合结构的推断和识别。最后，基于图像中提取的猪只位置信息与关键部位结构信息，在视频中提取各个猪只的动物行为学指标特征，研究多种日常行为的分类模型。通过本项目的研究，拟建立一套完整的猪只行为视频监测体系，实现猪只行为观测的自动化和精确化。

中文关键词： 畜禽行为观测；目标检测；姿态估计；行为分类

英文摘要： Automatic monitoring of pig behavior by video surveillance systems will become the future direction of modern pig industry. In this project, we aim to study intelligent behavior monitoring algorithms for pigs in group housing environments, including pig detection, pig posture estimation, pig behavior classification and related contents. Advanced technologies in computer vision are introduced to improve the level of pig behavior monitoring and realize the automatic observation of critical behaviors. Firstly, we plan to implement the pig detection algorithm based on the body contour and head texture features. This algorithm will overcome limitations of existing pig detection methods. Essential pig appearances are fully exploited for accurate localization in group environments. Then, the part-based posture estimation algorithm is studied. Probabilistic graphical models are utilized to describe the structural feature of pigs. Thus, the optimal part configuration can be inferred to recognize pig postures. Finally, according to detected positions of the holistic pig and its parts, we extract movement indexes of pigs from the video. Thus, classification models for specific individual behaviors are built. Through this project, a whole intelligent behavior monitoring system for pigs will be constructed to realize automation and precision of pig behavior monitoring.

英文关键词： Animal behavior monitoring;object detection;posture estimation;behavior classification