基于立体相机和GPS的实时移动机器人混合导航研究

Real-time Mobile Robot Navigation Based on Stereo Vision and Low Cost GPS

目前大多数用于大规模户外运行的机器人导航系统是基于高端传感器实现,研发低成本导航系统具有实际应用价值,故设计一种仅使用立体相机和普通GPS接收器的低成本自主导航混合系统很有必要。本研究所提出的方法需包括视觉测距(VO),位姿估计,障碍物检测,局部路径规划和航路点跟随器。视觉测距不可避免地会出现漂移(误差累积),GPS提供用于校正VO漂移的绝对位置。使用扩展卡尔曼滤波器(EKF)融合来自VO和GPS的数据,实现更精确的定位。实验结果表明,移动机器人在该导航系统路径规划下到达目标终点,且精度较高。

At present, most of the robotic navigation systems for large-scale outdoor operation are based on high-end sensors and it is still challenging to develop low-cost automatic land vehicles. Therefore, it is necessary to design an autonomous navigation system only using stereo cameras and low-cost GPS receivers.The proposed methods include visual ranging(VO), pose estimation, obstacle detection, local path planning and point follower. VO calculates the relative pose between two pairs of stereo images. However, over the time, VO drift(error accumulation) inevitably occurs, but low-cost GPS can provide absolute position for VO drift correction. Extended Kalman filter(EKF) can be used to fuse data from VO and GPS to achieve more accurate localization locally and globally. At the same time, in order to detect obstacles, dense depth maps generated by stereo parallax estimation can be used and converted into 2 D raster maps. Finally, not only the local path planning calculates the temporary route points to avoid obstacles, but also the route point follower sets the robot as the target point at the same time. In order to verify the proposed scheme, the experimenter used the mobile robot platform in real-time experiments in outdoor environment and evaluated the proposed method. The experimental results showed that the mobile vision and control system can drive autonomously in this outdoor environment.