基于惯性导航与立体视觉的风管清扫机器人同时定位与地图创建方法

风管清扫机器人是用于完成中央空调通风管道系统清扫工作的一种自动化设备，为提高其在庞大且封闭的未知通风管道网络中的自主导航能力，提出一种低成本的机器人同时定位与地图创建解决方案。该方案结合惯性测量单元的高速动态性能及双目立体视觉传感器良好的环境感知能力，通过采用Rao-Blackwellized粒子滤波方法融合两种传感器的测量信息，可以有效地抑制惯性传感器的漂移，快速地估计机器人三维的位置、姿态和速度，并且获取稳定的环境视觉路标信息，能够满足风管清扫机器人同时定位和地图创建的要求。同时，为确保视觉路标关联的鲁棒性，提出一种双向的基于几何相容性及路标视觉特征相结合的数据关联方法。试验结果证明所提的同时定位与地图创建方案和数据关联法的有效性。

Simultaneous Localization and Mapping Research for Air-duct Cleaning Robot Based on Inertial Navigation and Stereo Vision

The air-duct cleaning robot(ADCR) is a kind of automation equipment to clean the ventilation-duct of the cental air-conditioning system. In order to improve its capability of autonomous navigation in an unknown, enormous and enclosed air-duct network, a low-cost robot simultaneous localization and mapping(SLAM) scheme is proposed. By combing an inertial measurement unit(IMU) with a stereo camera and fusing measurements of both sensors with the Rao-Blackwellized paticle filter, the proposed scheme can restrain drift of IMU, and provide fast dynamic full six-dimensional ego-motion information (including position, attitude and velocity) of the ADCR and reliable three-dimensional visual information inside the ventilation-duct. Furthermore, a bidirectional data association method based on the geometry compatibility and landmark’s visual appearance is proposed to guarantee robustness of visual landmarks. The performance of the proposed scheme is proved to be effective with experiments.