基于粒子群神经网络的移动机器人门牌号码识别技术研究

【作者】 莫国梁；

【导师】 郭健；

【作者基本信息】 南京理工大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 科学技术的不断进步推动了机器人的发展，移动机器人的应用越来越广泛，对移动机器人的研究成为当今的热点之一。本文以AS-R移动机器人为平台，针对移动机器人视觉导航问题，研究了移动机器人门牌识别技术。利用CCD传感器来感知环境信息是移动机器人视觉中的一个关键环节。但由于传感器精度、稳定性等问题，往往使得检测到的数据变得不完全、不连续、不可靠，不能准确地、全面地实现对环境的描述，因此要实现移动机器人通过摄像头来检测与识别目标是一项艰巨的任务。本文在总结国内外移动机器人研究成果的基础上，给合本课题的实际情况，提出了一种新的门牌检测与识别方法。在检测方面，针对移动机器人在运动的过程中所拍摄到的图像的特点，提出一种基于HSI色彩空间模型的粗定位与基于边缘特征的细定位方法来实现门牌号码的定位，并利用门牌号码的轮廓特征来实现字符分割。在识别方面，考虑到门牌号码识别特殊性，提出了一种基于字符特殊节点与粒子群神经网络相结合的新方法。该方法是利用门牌数字的字符特殊节点，进行粗分类；再通过计算字符的不变矩，利用粒子群神经网络进行细分类。与传统的识别方法相比，该方法提高了对门牌识别的准确率，而且具有较好的鲁棒性。实验验证了方法的有效性。最后利用Visual C++集成开发环境，编程实现各模块功能，进行实际系统调试。实验结果说明移动机器人能完成任务要求，并在满足实时性的前提下，能准确检识别门牌号码，验证了所提方法的可行性和有效性。更多还原

【Abstract】 The Robot technique has been developing for the advancement of science. Nowadays, Robot has been widely used. And Mobile Robot research is one of the hottest researches in the world. This paper is based on the AS-R Robot platform to do a lot of work concerning about the Robot Visual Navigation, and the main work is how to let the Robot to recognize the door number quickly and precisely.The use of CCD sensor to detect environment information is the key progress of the Robot Vision. But for the sensor is not precise and stabile enough that result in the data from the sensor would be not complete, not continue or uncertain, and the robot can not get the precise and complete information about the environment. That makes the Robot to detect and recognize an object become a difficult task.In this paper, after summarize the current research on the mobile Robot and considering the specific of the problem, we put forward a new door number detection and recognition method. About the detection, considering the specific circumstance that during the movement of the mobile Robot, the image from the CCD camera has its particular character, we use the HSI color model to realize the rough location of the door number and use the edge character to realize the precise location of the door number. After the door number location, we use the contour character to make the door number segmentation come true. About the recognition, for the specialness of the door number, a new method combine with the number special node and neural network to recognize the number is presented. This method use the special node to make the rough classification; And calculates moment invariants as the neural network input to realize the precise classification. Compare with the traditional method, this method can reduce the recognition error and has a good adaptability.This paper implements the module function under Visual C++ integrated develop environment. The experiment result shows that the robot could recognize the door number precisely in set time and shows that this method is feasible and effective.更多还原