基于双目视觉的柑橘采摘机器人目标识别及定位技术研究

发布时间：2018-06-04 12:29

本文选题：双目视觉 + 柑橘果实　；参考：《重庆理工大学》2017年硕士论文

【摘要】：柑橘是我国广泛种植的水果之一,在柑橘生产作业中,33%-50%的劳动力用于采摘,采摘作业重复性强且对劳动力需求巨大。如果能在柑橘生产过程中引进智能采摘机器人,实现机械化、智能化采摘,可以解放有限的劳动力资源,提高生产率,不但具有重要现实意义,也将是农业现代化的必然选择。果实识别和定位,是采摘机器人的首要任务和设计难点,其准确性决定了采摘机器人的工作效果。因此,对柑橘采摘机器人目标识别和定位技术的研究,具有重要的意义。本论文基于双目视觉技术,对柑橘采摘机器人目标识别和定位技术进行了研究,实现了柑橘果实的有效识别和三维空间的定位,为柑橘采摘机器人提供了必要的信息。主要的研究工作与成果如下:(1)搭建了双目立体视觉系统,并进行了双目摄像机的标定,得到了摄像机的内部和外部参数。对左、右相机进行了立体标定,得到了描述两个相机空间位置关系的旋转矩阵和平移向量。(2)对柑橘图像的分割方法进行了研究,对比了三种不同的柑橘图像分割方法,选择K-means聚类算法与HSV颜色空间下阈值分割的分割方法作为本研究的柑橘图像分割方法,设计了图像预处理方法对分割后的图像进行进一步的处理。(3)对柑橘图像的识别方法进行了研究,分别对单个柑橘目标和重叠柑橘目标进行了识别,并提出一种基于凸壳及距离变换理论的重叠柑橘目标识别方法,实现了柑橘的采摘中心点定位和柑橘目标的还原。试验结果表明,对于单个柑橘的识别,平均识别误差为2.03%。对于重叠柑橘目标的识别,仿真试验中的采摘中心点定位误差为6.51%,真实重叠柑橘的采摘中心点定位试验中,本论文方法的定位误差为1.58%。在重叠柑橘图像的还原试验中,平均还原误差为13.78%,表明该算法能够较精确地识别柑橘。(4)对柑橘目标的三维空间定位方法进行了研究,论文使用SURF算法进行了特征点提取,利用RANSAC算法和极线约束进行了误匹配点对的剔除,并依此提出了一种基于柑橘图像相似度及极线约束的采摘中心点匹配方法,实现了柑橘目标采摘中心点的三维坐标计算。试验结果表明,柑橘的平均定位误差为1.824 mm,满足采摘机器人定位要求。
[Abstract]:Citrus is one of the fruits widely planted in China. In citrus production, 33% to 50% of the labor force is used for picking. If intelligent picking robot can be introduced into citrus production process, mechanization and intelligent picking can liberate limited labor resources and increase productivity, which is not only of great practical significance, but also an inevitable choice of agricultural modernization. Fruit recognition and location is the most important task and design difficulty of picking robot, and its accuracy determines the working effect of picking robot. Therefore, it is of great significance to study the target recognition and location technology of citrus picking robot. In this paper, based on binocular vision technology, the target recognition and location technology of citrus picking robot is studied, which can effectively recognize citrus fruit and locate in three dimensional space, which provides necessary information for citrus picking robot. The main research work and results are as follows: 1) the binocular stereo vision system is built, and the binocular camera is calibrated, and the internal and external parameters of the camera are obtained. The rotation matrix and translation vector of the two cameras are obtained to describe the spatial relationship between the two cameras. The segmentation methods of citrus images are studied, and three different methods of citrus image segmentation are compared. The K-means clustering algorithm and the threshold segmentation method based on HSV color space are selected as the citrus image segmentation methods in this paper. The image preprocessing method is designed to further process the segmented image. The recognition method of citrus image is studied, and the single citrus target and the overlapping citrus target are identified, respectively. An overlapping citrus target recognition method based on convex hull and distance transformation theory is proposed to locate the picking center point and restore the citrus target. The results show that the average recognition error for single citrus is 2.03. For the identification of overlapping citrus targets, the positioning error of picking center point in simulation experiment is 6.51, and that in real overlapping citrus picking center point positioning test is 1.58. In the experiment of the reduction of overlapping citrus images, the average reduction error is 13.78, which shows that the algorithm can accurately identify the citrus. The 3D spatial location method of citrus target is studied. SURF algorithm is used to extract the feature points. The RANSAC algorithm and polar line constraint are used to eliminate the mismatched point pairs, and a matching method of picking center point based on citrus image similarity and polar line constraint is proposed, which realizes the 3D coordinate calculation of the citrus target picking center point. The experimental results show that the average positioning error of citrus is 1.824 mm, which meets the requirements of picking robot.
【学位授予单位】：重庆理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【参考文献】