视觉导引的物料搬运AGV关键技术研究

发布时间：2018-01-17 00:09

本文关键词：视觉导引的物料搬运AGV关键技术研究　出处：《华南理工大学》2015年博士论文　论文类型：学位论文

【摘要】：制造业是国家的基础性、前沿性、支柱性与战略性的产业。随着精益生产、柔性制造、智能制造、计算机集成制造、绿色制造、准时制生产等先进制造模式的涌现及应用,制造企业对制造车间的制造装备和物料搬运系统提出了更高的要求。自动导引小车(AGV)作为现代工业物流系统的关键设备之一,如何进一步提升其智能化、柔性化是当今及今后一个时期的重要研究方向。本文在回顾国内外物料搬运AGV、视觉导引AGV研究应用现状,以及机器视觉关键技术的基础上,以视觉导引AGV为研究对象,对目标跟踪、目标识别、视觉定位以及目标导航等关键技术开展了四个方面的深入研究。1、提出了一种基于多特征融合稀疏表征的目标跟踪算法。针对目标跟踪算法在复杂条件下跟踪鲁棒性不强的问题,在粒子滤波框架下,引入稀疏表示理论与仿射变换理论,以减少目标状态描述所需的特征数量,提高了算法的运算速度,且具有对目标平移、旋转和尺度变化等因素不敏感的特性。算法中引入了快速粒子筛选法及基于灰度与LBP特征的特征选择方法,前者剔除了与目标相似度低的粒子,可进一步减少计算量,后者通过结合?2稀疏表示原理对目标特征具有更强的区分力;融入了APG稀疏方程求解方法,使算法运算速度和鲁棒性再次提升。2、研究了样本学习分类机制,提出了一种基于级联分类器的目标识别算法与基于朴素贝叶斯分类的目标跟踪算法的目标跟踪识别算法。在基于结构稀疏表示的目标跟踪算法基础上,利用贝叶斯分类器对候选样本进行分类,确定目标所在区域,提高了跟踪算法的稳健性。在跟踪算法基础上,融入基于学习分类机制的目标识别算法,使跟踪算法能够适应目标形变与消失的情况。3、提出了一种具有较好性能的改进随机抽样一致性估计方法。在基本RANSAC算法框架下,引入预检验技术提升基础矩阵估计运算效率;设计评估函数优化内外点筛选判定机制,基于点特征分布密度分析和对极距离分析提出了一种最终内点集优选策略,进一步提高了最终内点集数据质量;并结合最终内点集优选策略,提出候选样本中的错误最后利用M估计器进行最终基础矩阵估计,进一步提高了整个估计算法的精度及相对于图像噪音的质量。4、在智能视觉导引技术基础上,结合工业物料搬运需求,研制了一款基于视觉导引的智能型物料搬运AGV。通过分布式控制架构、高级控制、智能控制与智能导引技术,使本文研制的AGV能够实现自主物料运输功能。
[Abstract]:Manufacturing industry is a basic, leading, pillar and strategic industry. With lean production, flexible manufacturing, intelligent manufacturing, computer integrated manufacturing, green manufacturing. The emergence and application of advanced manufacturing models such as just-in-time production. Manufacturing enterprises have put forward higher requirements for manufacturing equipment and material handling system in manufacturing workshop. Automatic guided vehicle (AGV) is one of the key equipment of modern industrial logistics system. How to further improve its intelligence and flexibility is an important research direction at present and in the future. This paper reviews the domestic and foreign material handling AGV, visual guidance AGV research and application status. On the basis of the key technology of machine vision, taking visual guided AGV as the research object, four key technologies, such as target tracking, target recognition, visual positioning and target navigation, are studied deeply. A target tracking algorithm based on sparse representation of multi-feature fusion is proposed in this paper. The sparse representation theory and the affine transformation theory are introduced to reduce the number of features needed for the description of the target state and to improve the speed of the algorithm and the translation of the target. The fast particle selection method and the feature selection method based on gray and LBP features are introduced in the algorithm. The former removes the particles with low similarity to the target. Can the computation be further reduced by combining the latter? (2) sparse representation principle has stronger distinguishing power to target feature; APG sparse equation solution method is integrated to improve the speed and robustness of the algorithm again. 2. The classification mechanism of sample learning is studied. A target tracking algorithm based on concatenated classifier and naive Bayesian classification is proposed. Bayesian classifier is used to classify candidate samples, which determines the target region and improves the robustness of the tracking algorithm. Based on the tracking algorithm, the target recognition algorithm based on learning classification mechanism is incorporated. So that the tracking algorithm can adapt to the deformation and disappearance of the target. 3. An improved random sampling consistency estimation method with better performance is proposed under the framework of the basic RANSAC algorithm. The pretest technique is introduced to improve the efficiency of the basic matrix estimation. The evaluation function is designed to optimize the selection mechanism of inner and outer points. Based on point characteristic distribution density analysis and polar distance analysis, an optimal selection strategy is proposed to improve the data quality of the final interior point set. Combined with the optimal selection strategy of the final interior point set, the errors in the candidate samples are proposed. Finally, the M estimator is used to estimate the final basic matrix. Further improve the accuracy of the whole estimation algorithm and the image noise quality of .4. on the basis of intelligent visual guidance technology, combined with the industrial material handling requirements. An intelligent material handling AGV based on visual guidance is developed, which is based on distributed control architecture, advanced control, intelligent control and intelligent guidance technology. The AGV developed in this paper can realize the function of independent material transportation.
【学位授予单位】：华南理工大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TP391.41

【相似文献】