摘要
如何降低计算复杂度是视觉机器人同步定位与地图(SLAM)构建的热点问题。提出一种基于单目视觉的低计算复杂度的轮式机器人同步定位与地图构建算法。该算法在观测步通过图像处理与分析,识别特征点并进行定位,将轮式机器人的视觉投影与空间物体的几何关系转换为计算机器人相对特征点的距离和角度。整体算法步骤按照预测、观测、数据关联、更新、地图构建的递推算法进行同步定位与地图构建。提出的算法可识别环境目标,并进行平滑运动。在滤波观测步只处理单帧图像数据,和Active Vision和立体视觉方法相比,降低了算法的计算复杂度。
How to reduce computation complexity is a hot issue for visual robot simultaneous localization and map building. This paper proposes a low computation complexity method for wheel robot simultaneous localization and map building on the basis of monocular vision. In observation step, the proposed method can identify and locate landmarks through observations obtained from image processing and analysis. Geometrical relationship between visual image projection of the wheel robot and spatial objects is converted into the relative distance and angle between landmarks and the robot. The integral procedure allows the robot to conduct SLAM through a recursive algorithm, which predicts, observes, associates data, performs updates, and builds the map. The proposed algorithm is able to identify environmental objects and conduct smooth movement. Only one-frame image of the data is processed during filtering observation step. Compared with active vision algorithm and stereo vision algorithm, the proposed method is able to significantly reduce the computation complexity.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2010年第1期117-121,共5页
Chinese Journal of Scientific Instrument
基金
国家自然科学(60675048)
陕西省教育厅专项(08JK385)
西安市创新支撑计划(YF07021)资助项目
