【作者】 丁刚；

【导师】 陈宗海； 王智灵；

【作者基本信息】 中国科学技术大学 ， 模式识别与智能系统， 2009， 硕士

【摘要】 移动机器人利用多超声波传感器识别周围环境是移动机器人在未知环境中自定位、地图创建以及导航的重要组成部分,是移动机器人导航领域的关键技术之一。视觉传感器是移动机器人的重要传感器,移动机器人主要通过视觉传感器来感知环境、识别物体。研究基于视觉传感器的移动机器人物体识别,有利于移动机器人智能化的提高。本文针对未知环境下移动机器人的目标识别问题,分别基于多超声波传感器和视觉传感器研究了移动机器人的环境感知和物体识别方法,提出了一种基于支持向量机(SVM)的移动机器人物体识别方法。主要的研究工作和贡献如下:(1)对基于多超声波传感器的移动机器人环境建模和识别进行了研究。针对未知环境下移动机器人的环境理解与识别问题,提出了一种基于支持向量机的环境识别算法。该算法在对移动机器人室内外特征环境分析和建模的基础上,通过机器人配置的多超声波传感器获取环境的距离信息,直接作为环境的特征,按照从左到右的顺序组成表征环境轮廓的六维特征向量,送入支持向量机训练并用于特征环境的识别。该算法克服了多超声波传感器测量数据的不确定性对分类结果准确度的影响,实现了移动机器人对室内外特征环境的正确识别。(2)对基于视觉的移动机器人物体图像综合特征的提取进行了研究。在图像形状特征提取方面,提出了一种基于面积比的仿射不变量的构造,该构造方法简单可靠,在物体图像形状发生仿射变换时具有很强的稳定性,在光照影响、噪声干扰情况下,也具有一定的鲁棒性;在图像纹理特征提取方面,采用了一种基于灰度共生矩阵(GLCM)的纹理特征提取方法,选取了对比度、相关性、角二阶矩、熵、局部平稳5种纹理描述能力最强的特征参数组合;在图像颜色特征提取方面,提出了一种基于加权分块颜色直方图的颜色特征提取方法,该方法既能够反映颜色的空间分布信息,又保证了对图像旋转、平移和尺度变化的适应能力。(3)对基于支持向量机的移动机器人物体识别方法进行了研究。针对移动机器人的物体识别问题,提出了一种基于支持向量机的移动机器人物体识别方法,并从物体图像数据库中选择了20个物体的图像用于实验分析。通过仿真实验,比较了基于图像综合特征方法与单一特征方法的分类性能,基于支持向量机的物体识别方法与其他分类算法的性能,并分析了本文所提出的方法在光照干扰下的性能。更多还原

【Abstract】 Environment recognition for mobile robot utilizing multi-ultrasonic sensors is an important part of localization, mapping and navigation for mobile robot in unknown environment, which is also one of the key technologies in mobile robot navigation. Vision sensor is an important sensor for mobile robot, which is used to perceive environment and indentify object. Research on object recognition for mobile robot based on vision sensor can improve the intelligence of mobile robot.Aimed at object recognition for mobile robot in unknown environment, environment perception and object recognition for mobile robot respectively based on multi-ultrasonic sensors and vision sensor are investigated. A kind of object recognition arithmetic based on support vector machine (SVM) is proposed. The main contributions of this thesis are:(1) Research on environment modeling and indentifying for mobile robot based on multi-ultrasonic sensors. Aimed at environment understanding and recognition for mobile robot in uncertain environment, a kind of environment recognition arithmetic based on support vector machine is proposed. On the basis of analyzing and modeling the indoor and outdoor environment with special features for mobile robot, the distance information of environment is measured by multi-ultrasonic sensors in mobile robot to be used as the feature of environment directly. Afterword, these features are combined into a six-dimensional eigenvector for environment profile from left to right, which is sent to support vector machine for training and recognition. The method overcomes the effects of data uncertainties from multi-ultrasonic sensors on accuracy of classification results and implements the accurate recognition of the indoor and outdoor environment with special features for mobile robot.(2) Research on comprehensive feature extraction of object image for mobile robot based on vision. In the aspect of shape feature extraction of image, a kind of method of constructing affine invariant based on area ratio is proposed. The method is simple and reliable, which shows very strong stability when affine transformation happens in the shape of object image and shows some robustness when illumination affecting and noise disturbing. In the aspect of texture feature extraction of image, the method of texture feature extraction based on gray level co-occurrence matrix (GLCM) is applied. The method selects five feature parameters including contrast, correlation, energy, entropy and locally stationary, whose ability in the texture description is the strongest. In the aspect of color feature extraction, a kind of method of color feature extraction based on weighted block color histogram is proposed. The method can reflect the spatial distribution of color and show the adaptability to image rotation, translation and scale variation.(3) Research on the arithmetic of object recognition for mobile robot based on support vector machine. Aimed at object recognition for mobile robot, a kind of method of object recognition for mobile robot based on support vector machine is proposed. And then the images of twenty objects from object image database are used to test. The experiments compare the classification performance based on comprehensive image feature to that based on single image feature and compare the performance of object recognition based on support vector machine to others. What’s more, the performance of the method presented in this paper when illumination disturbing is analyzed.更多还原