【作者】 李哲；

【导师】 田建艳；

【作者基本信息】 太原理工大学 ， 控制科学与工程， 2015， 硕士

【摘要】 姿态检测是行为分析的重要组成部分，而行为分析又是福利养殖、健康养殖的重要研究方向。现代畜禽养殖业自动化研究从群体监控逐步转向个体行为特征的分析，本论文使用微惯性传感器作为研究手段，提出猪只姿态识别方法，组合利用多个传感器检测猪只姿态，设计硬件电路以及软件算法，并提出基于姿态检测的异常行为分析方法，有助于提高养殖场自动化水平，减少人力资源需求，辅助饲养人员更好的预防各种疾病以及疫情的发生。本论文首先介绍了当前国内外畜禽异常行为检测的研究现状和研究方法，在传感器技术、图像处理、声音检测这三种检测手段中选取传感器技术作为研究方向。分析猪只常见姿态寻找各姿态对应特征，提出姿态检测方法，研究常用微惯性传感器的检测原理，对常见坐标系与姿态角的三种表示方法做了详细阐述。其次进行硬件选型，设计猪只姿态检测系统，选用微惯性传感器MPU6050（加速度传感器+陀螺仪）与HMC5883L（地磁传感器）作为检测传感器，利用CC2530组建Zigbee通讯网络，在STM32上完成姿态解算，最后将数据传输到上位机进行异常分析。检测系统分为信息发射单元、信息接收单元、上位机三部分，针对前两部分设计原理图和所需扩展电路，制作PCB电路板。在系统软件设计中，分析各传感器误差产生原因与校准方法，研究单独使用陀螺仪以及加速度传感器+地磁传感器进行姿态解算的局限性，提出互补滤波和卡尔曼滤波两种数据融合算法，综合使用三种传感器数据进行姿态解算。系统使用时首先完成初始化，对各传感器进行校准，然后采集陀螺仪角速度数据，利用加速度传感器与地磁传感器进行数据修正，通过四元数进行姿态解算，综合各传感器特点得到姿态角，利用姿态角和速度变化设计针对猪只的姿态检测算法，在实验室环境下进行仿真模拟分析。最后根据检测到的姿态数据设计异常行为分析方法，通过将一天中被测个体各姿态的持续时间周围同种猪只各姿态持续时间、该猪只前10天各姿态持续时间分别进行对比得到异常等级，综合分析两种情况得到警告等级，评价猪只的异常状况，该方法能够适应多种环境，综合考虑各种因素判断猪只异常状况，为饲养人员进行进一步判断提供参考。更多还原

【Abstract】 Posture detection is an important part of behavior analysis， behavioranalysis play an important part in welfare cultivation and healthy breeding.Automation research of livestock breeding changes from group monitoring toindividual study. This paper uses micro inertial sensor as research tool, present amethod to identify posture of pigs, combine multiple sensors to detection pigpostures. This paper also presents hardware circuit and software algorithm, putsforward abnormal behavior analysis method based on the posture detection. Thismethod can help the livestock farm raising automation level and reducing therequirement of the human resources, help the breeder do better ondisease prevention and find the epidemic infectious disease ealier than before.Firstly, this paper introduces the current domestic and foreign researchstatus and research methods on abnormal behavior decections of livestock andpoultry. Among sensor technology, image processing technology and audiodetection technology, sensor technology is selected as research direction.Common postures of pigs are analyzed to find the feature of different postures. And detection method of postures is presented. The detection principle ofcommon sensors is studied. The common coordinate system and the euler anglewith three methods are introduced.Secondly, the hardware model should be selected to design the posturedetection system of pigs. This system chooses MPU6050and HMC5883L asdetection sensors. CC2530is used to build Zigbee communication network.Then, data is transmitted to STM32and the attitude algorithm is executed. Andthe abnormal behavior is judged by the upper computer with data fromSTM32.This detection system can be divided into information transmission unit,information receiving unit and host computer. According to the schematicdiagram and the required extension for receiving unit and transmission unit,PCB circuit board is made.In the design of software system, the detection error of sensors andadjustment are analyzed. It has restriction to complete the attitude algorithmonly using GYRO or acceleration sensor or geomagnetic sensor. Two data fusionalgorithms, complementary filter and Kalman filter, are proposed. In the system,the parameters should be initialized firstly, gyroscope’s angular velocity iscollected, acceleration sensor and geomagnetic sensor are used to correct thedata, the attitude algorithm is computed by quaternion. Attitude angle andcomponent of velocity are used to design posture detection method. Simulationswhich are got under the environment of laboratory are used to analyze.Finally, abnormal behavior can be analyzed by posture data. The durations of each posture are recorded. Compared with the average data of pigs aroundand the data of this pig during the last10days, the abnormal level could be got.Then the warning level could be got with the abnormal level through analysis.The abnormal level of the pig can be evaluated. This method can be used inmany environments, and various factors are taken to judge the abnormalsituation, which can provide references for breeders to make further judgment.更多还原