温室盆花自动装载中的视觉定位系统关键技术研究

发布时间：2018-06-05 18:55

本文选题：盆花 + 图像处理　；参考：《天津理工大学》2015年硕士论文

【摘要】：随着传感技术、计算机技术、人工智能及其它相关学科的迅速发展,机器人正向具有自组织、自学习、自适应能力的智能化方向发展。对于视觉机器人而言,能够实现对周围环境物体的准确识别、定位、跟踪是代表其智能化能力的关键指标。机器人的导航方式有多种,其中,视觉导航系统的研究是当今机器人研究的热点。在机器人行为控制及视觉导航研究中,如何提高机器人在未知环境中行为的正确性和视觉系统图像识别的实时性和准确性,是研究的热点之一。针对这些问题,本文提出了基于DSP-FPGA的高速图像采集处理系统设计方案并将其运用在温室盆花装载中,该方案对采集到的图像进行边缘提取、锐化等图像处理,并根据FPGA并行处理的特点提出一种改进的中值滤波方法,且进行仿真实验,仿真结果表明:在FPGA进行图像处理中采用改进的中值滤波算法,不仅能够很好的对采集到的图片进行去噪,而且具有很快的运算速度。机器视觉定位系统解决机器人的目标定位和跟踪问题,是整个机器人系统的核心和关键。本课题采用跟踪物体目标的特征点的视觉定位算法为基础,从摄像机的标定、模板匹配、背景建模、前景目标分离到特征点提取、运动估计及卡尔曼滤波,最终实现了视觉机器人定位的过程及其效果。通过对传统的角点检测算法进行深入了解与分析,本文在其基础上提出了改进。引入一种基于改进Harris角点提取的准确跟踪方法,该方法在传统Harris特征点检测基础上,利用角点附近像素灰度值梯度的变化关系,利用简单的运算与分析,首先排除一些伪角点与非角点,接下来再对保留的点做进一步处理,得出正确特征点。通过编写本算法代码最终实现其检测效果,并与传统算法相比较,得出本算法能够再更短的时间内提取出更为精确的角点,为下一步盆花的准确跟踪打下的基础,体现了该算法的实用性。
[Abstract]:With the rapid development of sensing technology, computer technology, artificial intelligence and other related disciplines, robot is developing intelligently with self-organizing, self-learning and adaptive ability. For the visual robot, it is the key index to realize the accurate recognition, location and tracking of the surrounding objects. There are many navigation modes of robot, among which, the research of visual navigation system is the hot spot of robot research. In the research of robot behavior control and visual navigation, how to improve the correctness of robot behavior in unknown environment and the real-time and accuracy of visual system image recognition is one of the hot research topics. Aiming at these problems, this paper puts forward the design scheme of high speed image acquisition and processing system based on DSP-FPGA and applies it to the loading of greenhouse potted flowers. According to the characteristics of FPGA parallel processing, an improved median filtering method is proposed and simulated. The simulation results show that the improved median filtering algorithm is used in FPGA image processing. Not only can the collected images be de-noised very well, but also it has a fast computing speed. The machine vision positioning system is the core and key of the whole robot system to solve the problem of target location and tracking. Based on the visual localization algorithm for tracking the feature points of the object, the subject includes camera calibration, template matching, background modeling, extraction of feature points from foreground targets, motion estimation and Kalman filter. Finally, the process and effect of vision robot localization are realized. Based on the deep understanding and analysis of the traditional corner detection algorithm, this paper proposes some improvements. An accurate tracking method based on improved Harris corner extraction is introduced. On the basis of traditional Harris feature point detection, the change relation of pixel grayscale gradient near corner is used, and simple operation and analysis are used. Firstly, some pseudo-corner points and non-corner points are excluded, then the reserved points are further treated and the correct feature points are obtained. Compared with the traditional algorithm, the algorithm can extract more accurate corner points in a shorter time, and lay the foundation for the further accurate tracking of potted flowers. It reflects the practicability of the algorithm.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41

【参考文献】