采摘机器人微扰目标与视觉定位的耦合机理

The coupling mechanism of the force and the visual position in picking target weak perturbation is researched for the problem of that the harvesting robots in complex environment are difficult to position precisely when disturbed.First the target perturbation state is analyzed quantitatively from mechanics angles, the relationship between the perturbation range and the force is researched to set up the relation model of the disturbing force, the identification parameters estimation are determined to realize the quantization and classification of the perturbation;then the relationship between the disturbing force and the visual image fuzzy degree is researched to realize the classification of fuzzy image. Further explore the fuzzy image processing methods based on image classification, through improving the discrete cosine transform method in the image restoration algorithm, effectively solve the regularized inverse of the image fuzzy operator, realize the image restoration algorithm of the edge blurring resulting from weak perturbation of litchi and its location algorithm; finally comprehensively analyze the visual position behavior, and the decoupling model and intelligent decision knowledge base in weak perturbation condition are built to realize the active identification for the target perturbation state of the visual system, and then to realize the intelligent decision-making and the position calculation. The homemade binocular vision picking robots physical prototype and quantitative simulation method are used to verify the visual position mechanism. This project aims to explore the visual position mechanism of the weak perturbation target, which can establish the theoretical and technical basis for robot vision precise position.

针对复杂环境下果蔬采摘机器人作业时，目标受到扰动难以精准定位的问题，研究采摘目标微扰动的受力与视觉定位的耦合机理。首先从力学角度定量分析目标扰动状态，研究扰动幅度与受力的关系，建立扰动力的关系模型，确定扰动状态的辨识参数，实现扰动状态的量化与分级；然后研究扰动力与视觉图像模糊程度的关系，进行模糊图像分类；进一步探索基于图像分类的模糊图像处理方法，通过改进图像复原算法中的离散余弦变换法，有效求取图像模糊算子的正则化逆，实现微扰动荔枝串边缘模糊的图像复原算法和果实的识别定位算法；最后对视觉定位行为进行系统综合分析，建立微扰动条件下视觉定位行为的解耦模型和智能决策知识库，实现视觉系统对目标扰动状态的主动辨识，并进行智能决策与定位计算。用自制的双目视觉采摘机器人物理样机和定量仿真方法,对视觉定位机制进行验证。本项目旨在探索微扰动目标的视觉定位机理，为机器人的视觉精准定位奠定理论和技术基础。

针对自然环境下采摘机器人目标扰动的视觉定位方法进行了研究。首先，对采摘目标的不同扰动状态，利用ANSYS有限元分析法进行了目标受力与扰动幅度大小的分析，建立了荔枝、柑橘扰动时摆动幅度与受力的数学模型，并确定果实质量与受力之间的函数关系；第二，设计了扰动荔枝、柑橘的视觉图像获取试验，确定了辨识三种扰动状态：静止、微扰动和大扰动的方法，对微扰动目标进行视频采集，然后再读取各帧图像，有效地避免了图像的模糊现象；然后利用模糊聚类分割法获得每帧图像中的果实和果梗，计算果梗上采摘点的平均值，从而确定扰动荔枝、柑橘的有效采摘点；第三，对自然环境不同光照、不同品质的荔枝进行了视觉识别与定位方法研究，利用探索性分析法实现了果实品质实时判断；采用Retinex图像增强算法结合双三次插值算法，以及改进模糊C均值（Fuzzy C-Mean）算法实现了不同光照下荔枝果实的识别。最后，对微扰动的采摘目标视觉精确定位机制进行智能化设计，构建了微扰动目标视觉定位的智能系统，并设计了真实环境和虚拟环境下扰动目标视觉定位的机器人采摘实验进行验证，研究为解决采摘机器人视觉的智能化提供了新的理论方法和途径。