【作者】 杨杰；

【导师】 闵华松；

【作者基本信息】 武汉科技大学 ， 电路与系统， 2014， 硕士

【摘要】 移动机器人的定位技术是实现机器人自主导航的关键技术之一。随着图像处理技术和计算机视觉技术的发展，基于视觉传感器的定位技术在机器人定位系统中得以广泛使用。Kinect传感器，是一款能够同时提供彩色图像和深度图像的视觉传感器。基于Kinect的视觉里程技术，同传统视觉里程计一样，通过对采集的图像进行特征提取与匹配，然后进行运动估计计算出姿态信息。不同的是，Kinect摄像机能同时获取到环境的深度信息，采用密集点云匹配技术对位姿进行优化，因此Kinect视觉里程计比传统视觉里程计的定位精度更高。但是由于对图像信息的依赖，在特殊场景下会存在匹配失误，同时由于仅依赖单个传感器，系统的累积误差无法修正。在机器人导航系统中，组合导航技术对提高系统的定位与导航性能具有很大的促进作用，因此本文提出将惯性导航技术与视觉导航技术相结合，研究多传感器融合对系统定位性能的改善。在机器人定位融合的过程中提出将姿态和位移分别融合的方案，对前者采用加权平均的方法，后者采用扩展卡尔曼滤波的方法。最后搭建三轮移动机器人平台和ROS软件测试平台，布置不同运动模式和不同局部场景两组测试环境，对提出的改进定位算法进行验证与分析。更多还原

【Abstract】 Mobile robot localization technology is one of the key technologies to realizeautonomous navigation. With the development of image processing and computer visiontechnology, localization technology which based on vision sensor is widely used in robotlocalization system. Kinect is a sensor which can provide RGBD image and depth RGBDimage. Visual odometry technology based on Kinect does feature extraction and matchingusing the collected image information, then does motion estimation to calculate the motionparameters of the camera, like tradition visual odometry technology does. The differenceis that Kinect visual odometry technology can optimize the position of the camera by pointcloud matching technology using the depth information of environment. So visualodometry technology based on Kinect has higher accuracy in localization than traditionvisual odometry technology. But owing to the dependence on image information,mismatching error will happen in special circumstances. Moreover the cumulative errorcannot be modified because of the only rely on a single sensor.Integrated navigation technology has played great role in promoting the localizationand navigation performance in the robot navigation system, so the combination of theinertial navigation technology and visual navigation technology is proposed in this paperto research the improvement of localization performance using multi-sensor fusiontechnology. In the process of robot localization fusion in the attitude and displacement,weighted average method is used for the attitude fusion, extended Kalman filter is used forthe displacement fusion. Finally, three wheeled mobile robot platform and ROS softwareplatform are built, and two groups of test environment of different movement patterns anddifferent local scene are layout to analyze and verify the improved localization algorithm.更多还原