基于虚拟现实的拖拉机双目视觉导航试验方法研究

发布时间：2018-02-11 22:02

本文关键词： 虚拟试验双目视觉物理引擎作物行识别导航控制　出处：《中国农业大学》2017年博士论文　论文类型：学位论文

【摘要】：农机视觉导航系统在复杂田间环境中的灵活性较好,如何设计图像处理算法以提高导航线识别精度和速度是当前研究热点之一。农机导航系统的传统试验方法以田间试验为主,存在试验成本较高、对作物生长时期依赖性较强、试验周期较长、试验过程易于对作物造成损伤等问题。为解决上述问题,本文以拖拉机为作业机械、苗期棉花为目标作物、棉田作物行环境为试验场景,在虚拟现实环境下,研究拖拉机双目视觉导航试验方法,主要研究内容如下:(1)虚拟导航试验场景几何建模方法研究。根据实际田间试验环境的几何特点,建立拖拉机外观模型、棉花单体模型、杂草模型、作物行模型和路面模型,形成虚拟试验场景的几何模型。该方法能够根据试验要求模拟多种棉田作物行场景,为研究作物行识别方法和路径跟踪控制方法提供丰富的试验环境。(2)拖拉机物理引擎建模方法研究。建立物理引擎的数学模型,包括整车模型及轮胎模型、传动系模型、制动系模型、转向系模型和解算模型。基于C++语言在Visual Studio 2008软件环境下开发物理引擎的软件系统。该方法能够根据实车参数和试验场景信息快速、正确地解算拖拉机的动力学参数,并在虚拟试验场景中实时渲染拖拉机的位姿状态。(3)基于双目视觉的田间导航路径识别方法研究。通过检测并统计作物行特征点的空间分布规律识别作物行中心线,有效减少了图像匹配点的数量,提高了识别精度和速度。在非地头环境下,作物行中心线的正确识别率不小于92.11%,平均偏差角度的绝对值不大于1.07°,偏差角度的标准差不大于2.52°;图像处理时间的平均值不大于202.90 ms、标准差不大于17.75 ms。通过比较作物行中心线与拖拉机行驶方位的相对位置规划导航路径,能够保证拖拉机稳定跟踪同一条目标作物行,目标路径规划的正确率为97.33%;导航路径规划时间的平均值为0.017 ms,标准差为 0.017 ms。(4)拖拉机路径跟踪控制方法研究。基于纯追踪方法建立前轮转向角计算模型。基于增量式PID算法设计转向控制方法,运用遗传算法优化PID控制器参数。设计路径跟踪控制策略以适应不同类型目标路径的跟踪精度和速度要求。虚拟试验结果表明,该方法设计的路径跟踪控制系统能够快速、稳定地跟踪目标路径,拖拉机的行驶轨迹相对于目标路径的超调量较小。(5)拖拉机虚拟导航系统验证试验。开展台阶障碍、随机路面和转向性能虚拟试验,测试物理引擎的有效性。分别运用虚拟棉田作物行图像和实际棉田作物行图像测试作物行识别方法的性能。开展平行阶跃直线路径和倾斜直线路径的跟踪虚拟试验,测试路径跟踪控制系统的性能。开展直线作物行和曲线作物行的跟踪虚拟试验,测试拖拉机虚拟导航试验系统的有效性。
[Abstract]:The flexibility of agricultural machinery visual navigation system in complex field environment is good. How to design image processing algorithm to improve the accuracy and speed of navigation line recognition is one of the current research hotspots. The traditional test method of agricultural machinery navigation system is mainly field experiment. In order to solve the above problems, such as high test cost, strong dependence on crop growth period, long test period and easy damage to crops, this paper takes tractor as working machine and seedling cotton as target crop. Under the virtual reality environment, the test method of binocular vision navigation for tractor is studied. The main research contents are as follows: (1) the geometric modeling method of virtual navigation test scene. According to the geometric characteristics of the actual field test environment, the tractor appearance model, cotton monomer model, weed model, crop row model and road surface model are established. The geometric model of virtual experiment scene is formed. In order to study crop row identification method and path tracking control method, this paper provides a rich test environment for tractor physical engine modeling, and establishes mathematical models of physical engine, including vehicle model, tire model, transmission system model, etc. Braking system model, steering system model and calculation model. Based on C language, the software system of physical engine is developed in Visual Studio 2008 software environment. Correctly calculate the dynamic parameters of the tractor, The field navigation path recognition method based on binocular vision is studied. By detecting and counting the spatial distribution of crop row feature points, the crop row centerline is identified. Effectively reduces the number of image matching points, improves the recognition accuracy and speed. The correct recognition rate of the crop line centerline is not less than 92.11, the absolute value of the average deviation angle is not more than 1.07 掳, the standard deviation of the deviation angle is not more than 2.52 掳, the average image processing time is not more than 202.90 Ms, and the standard deviation is not more than 17.75 Ms. Planning the navigation path of the relative position of the line centerline and the driving direction of the tractor, To ensure that the tractor keeps track of the same target crop row, The accuracy of target path planning is 97.33, the average time of navigation path planning is 0.017 ms, and the standard deviation is 0.017 ms.4) the method of tractor path tracking control is studied. Based on pure tracking method, the calculation model of front wheel steering angle is established. Quantitative PID algorithm is used to design steering control method. Genetic algorithm is used to optimize the parameters of PID controller. A path tracking control strategy is designed to meet the requirements of tracking accuracy and speed of different types of target paths. The virtual test results show that the path tracking control system designed by this method can be used quickly. Tracking the target path stably, the tractor track is smaller than the target path overshoot, the tractor virtual navigation system verification test. Step obstacle, random road surface and steering performance virtual test, To test the effectiveness of the physical engine, the performance of the crop row recognition method was tested by using the virtual crop row image and the actual cotton crop row image, respectively. The virtual experiment of parallel step straight path and inclined straight line path was carried out. To test the performance of the path tracking control system, to test the effectiveness of the tractor virtual navigation test system, the tracking virtual test of linear crop row and curve crop row was carried out.
【学位授予单位】：中国农业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.9;S219

【参考文献】