【作者】 李永强；

【导师】 肖英奎；

【作者基本信息】 吉林大学 ， 农业生物环境与能源工程， 2018， 硕士

【摘要】 在果蔬生产过程中,采摘工作费时费力,劳动力需求较大,生产成本较高。随着人口老龄化以及城镇化的发展,农业劳动力日渐匮乏,采摘工作所需成本大幅度增长。目前,我国农业生产向着智能化、精准化、多样化、规模化的方向发展,研究智能化程度较高的农业机械装备和农业机器人可以有效解决农业劳动力不足、生产成本增加、工作效率低等问题,还能够改善工作环境,避免长时间工作以及农药、化肥的使用等对人体的损伤,具有现实意义。本文以品种为“吉杂4号”的黄瓜为研讨对象,在对黄瓜的物理特性了解和分析后,制作了4自由度黄瓜采摘机械手样机,并对目标果实的抓取特征点空间坐标值的获取和采摘机械手的总体控制系统进行了开发研究。主要研究内容及结果如下:(1)根据实际采摘环境对黄瓜采摘机械手的整体机械结构进行设计,主要包括:设计并制作4自由度机械臂,该机械臂的运动既灵活又稳定;设计制作摄像机支架,使摄像机跟随机械臂的腰关节运动,获得当前实时画面;根据黄瓜的物理特性对末端执行器进行了设计制造,根据黄瓜果实长度和黄瓜果柄长度之间的关系,将果实的抓取特征点设定在距离黄瓜果实顶端的1/4处,切割刀片距手指垂直距离为5.2 cm,实验表明该末端执行器的切割成功率达到95%,验证了该设计方法的可行性。(2)对如何通过图像处理获得特征点空间坐标值进行了研究。通过对比常用颜色模型,设定HSV颜色模型中的H和S分量阈值,将目标果实分割出来。应用MATLAB处理分割后的彩色图像得到目标果实的最小外接矩形,标记出形心像素坐标并转换为特征点对应的像素坐标值。针对黄瓜果实的识别情况对20组样本进行了实验,识别成功率达到80%,并且得到的坐标值在Y方向的误差可以满足采摘条件。介绍了特征点空间坐标值的获取方法和原理。针对Kinect传感器获得的深度距离进行误差实验,在20组样本数据中,误差的范围基本在0~2 mm之间。由实验验证:获取的深度值产生的误差对实际摘取结果产生的影响较小。(3)针对采摘机械手识别与控制系统的设计进行研究。针对机械臂的总体传动方式,选择了合适的驱动器和驱动结构。设计了基于PCI总线控制的采摘机械臂控制系统、基于Kinect for Windows SDK和MATLAB开发的特征点识别与获取系统以及基于STC89C52单片机开发的下位机控制系统。对这三部分进行了硬件和软件的设计与开发,设计制作了伺服控制电路、末端执行器控制电路、图像处理界面以及机械臂的控制界面等,初步完成了控制系统的软件与硬件的开发。(4)针对设计的采摘机械手样机,在实验室条件下,进行了黄瓜的定位与采摘模拟实验。首先,对机械臂的性能进行测试实验。实验表明:机械臂具备较好的位置控制精度和速度控制精度,能够满足机械臂的运动要求。其次,在实验室条件下进行了特征点空间坐标值误差的模拟实验。实验表明:图像处理获得的特征点空间坐标值与实际测量的空间坐标值之间在Y轴方向上产生的误差主要集中在3mm~5mm之间,在Z轴方向上产生的误差主要集中在2mm~5mm之间,两个方向产生的误差不会对采摘产生较大的影响,可以满足抓取要求。该模拟实验从原理上验证了采用的特征点空间坐标值获取方法的可行性。最后,在实验室搭建的模拟采摘环境下对机械臂和末端执行器配合采摘效果进行了模拟采摘实验。通过对40组模拟采摘实验结果进行分析可知:机械臂将末端执行器送到指定特征点处的定位成功率达到92.5%;末端执行器对目标果实的采摘成功率达到82.5%;两者配合作业完成目标果实采摘的整个过程所用的平均时间约为15.28 s。该采摘机械手运行稳定,能与控制系统良好的配合工作,实现果实的完整采摘。更多还原

【Abstract】 In the production of fruits and vegetables production,the picking work is time-consuming and laborious,requires a lot of labor,and costs more.With the aging population and the development of urbanization,The agricultural labor force is becoming less and less,and the cost of picking jobs has increased greatly.At present,the agricultural production in our country is moving towards the direction of intelligence,precision,diversification and scale.Studying agricultural machinery with high intelligence and agricultural robots can effectively solve the problems of insufficient agricultural labor force,increase in production cost and improve work efficiency.It can also improve the working environment,prevent long-term work,pesticides,fertilizers and other injuries to the human body,which is of practical significance.In this paper,the cucumber variety "Ji Za-4" was studied.After understanding and analyzing the cucumber’s physical characteristics,a 4-DOF cucumber picking robot prototype was made,and research on the acquisition of the coordinate value of the target feature point,the overall control system of the picking manipulator was carried out.The main research contents and results are as follows:(1)According to the actual picking environment,the overall mechanical structure of cucumber picking robot was designed,including: designed and made a 4-DOF manipulator to ensure that the motion of the manipulator is flexible and stable.,the camera bracket was designed and manufactured so that the camera can follow the movement of the waist joint of the manipulator to obtain the current picture.According to the cucumber’s physical characteristics,the end effector had been designed and made,according to the relationship between the length of the cucumber fruit and the length of the cucumber fruit stem,the grasping feature point of the fruit was set at 1/4 distance from the top of the cucumber fruit,and the cutting blade distance from finger is 5.2 cm,experiments show that the end effector cutting success rate was 95%,showed the feasibility of the design method.(2)How to get the coordinate value of feature point through image processing was studied.By comparing commonly used color models,the H and S component thresholds in the HSV color model were set,to segment the target fruit.Apply MATLAB to process the segmented color image to get the minimum bounding rectangle of the target fruit,mark the centroid pixel coordinates and convert to the pixel coordinate values corresponding to the feature points.According to the identification of cucumber fruit,20 groups of samples were tested,the recognition success rate was 80%,and the error of the obtained coordinate value in Y direction could meet the picking conditions.The method and principle of obtaining the spatial coordinates of feature points are introduced in detail.Error experiments were performed on the depth distances obtained with the Kinect sensor.In the 20 sets of sample data,the error range is basically between 0-2 mm.It is verified by experiments that the error generated by the obtained depth value has little effect on the actual extraction result.(3)Research on the design of picking robot identification and control system was performed.For Manipulator’s overall transmission,selected the appropriate drive and drive structure.The control system of picking manipulator based on PCI bus control,the feature point recognition and acquisition system based on Kinect for Windows SDK and MATLAB,and the lower computer control system developed by STC89C52 microcontroller were designed.For these three parts,the hardware and software are designed,design and manufacture of servo control circuit,the end of the actuator control circuit,image processing interface and the manipulator control interface.(4)Under laboratory conditions,for the design of picking robot prototype,the simulation experiment of positioning and picking were carried out.First,test the performance of the manipulator,experiments show that manipulator has better position control accuracy and speed control accuracy,it can meet the requirements of the manipulator movement.Secondly,in the laboratory conditions,the simulation experiment of the eigen-point space coordinate error was carried out,experiments show that the errors in the Y-axis direction between the spatial coordinate values of the feature points obtained by the image processing and the actual measured spatial coordinate values are mainly concentrated in the range of 3mm-5mm;the errors generated in the Z-axis direction are mainly concentrated in the range of 2mm-5mm.The errors generated in both directions will not have a great impact on the picking and can meet the requirements of grasping.The simulation experiment verifies the feasibility of using the method of obtaining the spatial coordinates of feature points.Finally,the simulation picking environment was set up in the laboratory and the simulation picking experiment was carried out according to the picking effect of the manipulator and the end effector.By analyzing the results of 40 groups of simulated picking experiments,we can see that the success rate of the manipulator positioning the end effector to the designated feature point was 92.5%,the success rate of the end effector picking the target fruit was 82.5%,the average time taken to complete the whole process of picking the target fruit with the two parts is about 15.28 seconds;the picking robot runs stably and can work well with the control system to achieve the complete picking of the fruits.更多还原