Abstract:

In terms of the problem of location distortion of mobile robot in SLAM（Simultaneous Localization And Mapping）, a method is proposed to correct incorrect location of the robot by building landmark database and establishing pose derivation model. The semantic laser is obtained by combining visual information and laser data in the process of mapping. Semantic labels is attached to landmarks and their location information on the map is recorded. Combining with various pose data and relative position relations, the actual position of the robot on the map can be calculated when location deviation occurs during navigation. The relocation can be completed. Experimental results show that this method overcomes the disadvantage of using laser or vision to locate or relocate the robot in the indoor environment, and it can effectively solve the problem of robot position drift. Even if the robot is hijacked from the current position to another position, it can be quickly relocated according to the proposed method, and the location accuracy is high.

Key words: Simultaneous Localization And Mapping（SLAM）, mobile robot, landmark, relocation, YOLOv3 algorithm

摘要：

针对移动机器人在SLAM（即时定位与地图构建）过程中出现的定位失真问题，提出一种通过搭建地标数据库和位姿推导模型，修正机器人错误定位的方法。建图过程中，融合视觉信息与激光数据，得到语义激光，赋予地标语义标签并记录其在地图上的位置信息。导航过程中，当产生定位偏差时，结合多种位姿数据和相对位置关系，推算出机器人在地图上的实际位置，完成重定位。通过实验测试可知，该方法克服了现有机器人在实际室内动态环境下，单一地采用激光或视觉进行定位或重定位技术的缺点和不足，能有效解决“机器人位置漂移问题”。将机器人从当前位置劫持到另一位置，也能根据提出的算法迅速重定位，且定位精度高。

关键词: 即时定位与地图构建（SLAM）, 移动机器人, 地标, 重定位, YOLOv3算法