论文部分内容阅读
针对移动机器人导航问题,采用视觉导航手段,通过辨识人工特征来获取环境观测基准.利用灰度值方差法检测二维图像特征点,并基于二维到三维的空间逆映射实现视觉特征点从相面坐标到世界坐标的转换,以此建立观测模型,并将其融入贝叶斯数据融合框架.为缓解模型线性化所引入的误差,提出迭代观测更新策略,通过持续优化滤波更新的初始点,提升系统联合后验概率估计的精度,进而改进对机器人位姿与环境基元的状态估计质量.使用搭载了机器视觉的机器人平台在真实环境中进行了轨迹总长为505 m实地实验,验证了本文所提出算法优于传统算法的性能.
In order to solve the problem of mobile robot navigation, the visual navigation method is adopted to obtain the environmental observation datum by recognizing the artificial characteristics.The gray value variance method is used to detect the two-dimensional image feature points and the visual feature points are obtained from the two-dimensional to three- Surface coordinates to the world coordinates in order to establish the observation model and integrate it into the Bayesian data fusion framework.In order to alleviate the error introduced by the model linearization, an iterative observation updating strategy is proposed. By continuously optimizing the initial points of the filtering update, Improve the accuracy of joint joint posteriori probability estimation, and then improve the state estimation quality of pose and environment primitives of robots.Using robotic platform equipped with machine vision, a field experiment with a total track length of 505 m was performed in real environment, The proposed algorithm outperforms the performance of traditional algorithms.