【作者】 李俭；

【导师】 朱晓蕊；

【作者基本信息】 哈尔滨工业大学 ， 控制工程（专业学位）， 2017， 硕士

【摘要】 新型传感器技术的发展推动着机器人研究方向从传统的里程计信息定位制图转向现在广泛的视觉定位、激光定位等。目前机器人视觉研究领域大都集中在如何快速提取、描述图像特征与图像之间如何做到较好的鲁棒性匹配,但在工程应用中存在图像处理计算量大造成实时性不好、旋转导致图像变化快定位不准等问题。激光也存在不可忽略的应用限制,比如要求在同一水平面扫描、旋转速度不能太快、每帧特征点不能太稀疏等。本文在前人的研究基础上,主要研究低成本的清洁机器人器人定位问题,并根据定位信息返回充电站。本文根据里程计、陀螺仪信息与天花板的激光信息建立状态方程、观测方程,通过自适应扩展卡尔曼滤波算法对信息进行融合实现机器人定位,并根据机器人的位置信息返回充电站。在摄像头标定时,本文探讨了不同距离、不同角度对摄像头标定结果的影响,从多组标定中选取误差最小的标定结果。为了减少光照对激光的影响,本文在摄像头处添加滤光片滤掉可见光,并对激光十字进行特征提取,设置多个阈值条件正确识别两个不对称的激光十字。提取到激光十字后,由十字交点的像素坐标利用三角几何关系得到其相机坐标,通过平移旋转得到观测模型需要的机器人位置信息与角度信息。从激光十字中选取作为坐标转换的特征点时,通过实验对比发现激光特征点应选取激光线段交叉点,端点处由于衍射原因以及图像膨胀腐蚀等处理使得像素坐标不稳定,继而根据不确定的像素坐标求得的机器人世界坐标必然不对,偏差很大。静止时机器人通过激光图像得到的位姿信息比较准确,但是旋转时由于图像变化很快,得到的机器人位姿就变得不可靠了。本文对建立的观测模型,根据可观性的微分几何特性和矩阵的秩利用李导数给出了可观性证明。为了配合返回充电站,机器人在直线运动时采用了输入饱和约束下非完整移动机器人轨迹跟踪的方法,在转弯时应用简单的PID控制。本文在最后部分通过实验结果验证了本文提出的激光十字、普通摄像头实现低成本清洁机器人定位、返回充电站方法的可行性与有效性。更多还原

【Abstract】 Nowadays,the development of new sensor technology has promoted the development of robotics which get location and map based on the traditional odometer positioning information to the extensive visual and laser positioning.Most of the visual research areas focus on how to extract and describe the feature fast and the better robust match between the images.But there are problems such as poor real-time performance due to the large amount of image calculation,inaccurate positioning owing to rotation quickly in the engineering application.Some limitations exist on the application of laser like horizontal scanning and rotating speed low.Above the previous research,mainly research is the low-cost sweeping robot location and back to the charging station.The dissertation establishes the state and observation equations based on the information of the odometer and laser information of the ceiling.Through the AEKF,integrated information can realize the robot localization and help return the charging station according to the position information.When camera calibration,this dissertation discusses the influence of the different distances and angles on the calibration results in order to get a better calibration results.To reduce the influence of light on the laser,a filter is added to screen out the visible light.After extracting the feature of the laser crosses,we adopt many threshold settings to improve the correct recognition rate of the laser cross.Then it can obtain the pixel coordinates of the cross point by using the geometric relations.The position and angle information of the robot are obtained by way of the simplified camera model.In this dissertation,we will make a detailed experimental study on laser cross-point selection and can find that the cross point of the laser is stable.The endpoints of the laser lines are unstable due to the diffraction and the world coordinates of the robot have larger errors determined by unstable feature points.When stopping,the robot pose information obtained by the image is more accurate.However,the robot position is unstable because of the image changing rapidly when rotating.This dissertation proves the observability of system according to the differential geometric properties of the observability and the rank of the matrix.In order to return the charging station,the dissertation adopts the method of trajectory tracking of nonholonomic mobile robot with input saturation constraint when straight.In the last part of this dissertation,the experimental environment is set up.The experimental results verify the feasibility and effectiveness of the laser cross,ordinary camera to realize the low cost robot localization and return the charging station.更多还原