基于多径指纹信号子空间的室内回环检测算法研究

Indoor loop detection algorithm based on multipath fingerprint signal subspace

移动机器人的SLAM过程中,相机追踪得到的位姿会产生误差。针对随着误差累积导致后续得到的轨迹和地图在世界坐标系中产生尺度漂移的问题,提出一种基于信号子空间的多径指纹的室内单站定位算法与视觉里程计定位算法进行融合用于回环检测。用基于格拉斯曼判别投影分析的无线电指纹定位算法对位置信息特征进行辨识,并且引入格拉斯曼投影尺度函数构建代价函数,将各位置空间进行离散化进而降低类间特征的相似性。仿真实验结果表明,无线电指纹重定位算法可以改善位置模糊问题,提高位置指纹鲁棒性以及辨识正确率,从而判断出机器人是否运动到之前某一位置附近从而消除累积误差。

The position and posture traced by the camera will produce the error in the SLAM process of mobile robot,and with the accumulation of errors,the scale drift is generated in the obtained track and cartography in the world coordinate system.Therefore,an indoor single station location algorithm based on the multipath fingerprint signal subspace is fused with visual odometer location algorithm for loop detection.The radio fingerprint localization algorithm based on Glassman discriminant projection analysis is used to identify the location information feature.The Glassman projection scaling function is introduced to construct the cost function,and each location space is discretized to reduce the feature similarity between two classes.The simulation experiment results show that the radio fingerprint relocation algorithm can improve the location ambiguity,robustness of location fingerprint and corrected identification rate,determine whether the robot moves to the location nearby the previous location,and eliminate the cumulative errors.