基于特征稀疏策略的室内机器人SLAM研究

Research on indoor SLAM based on strategy of sparse features

针对室内环境下的移动机器人的定位问题,研究了一种能够适应空间狭小、特征点密集的复杂环境的双目视觉SLAM方法。该方法以EKF-SLAM方法为基础,引入了一种特征点稀疏性控制机制,该机制对地图库中的特征点同时在位置空间和特征空间进行分布密度限制,克服了因特征点快速上升而导致的庞大的计算量问题,提高了数据关联的准确性。此外,降低了SIFT矢量的维数以降低计算量,且在左右试图的特征匹配中,引入了多个约束条件,以提高匹配的准确性。最终实验结果表明,方法能够实现特征点分布的均匀性,且在定位精度满足要求的前提下缩短定位时间。

Aimed at the problem of mobile robot localization in indoor environment,a binocular vision method of SLAM is researched,which can be applied to the incomprehensive and complicated environment with dense feature points.Based on EKF-SLAM method,a sparse control mechanism of feature points,which is characterized by restricting the distributed-density of the feature points on the global map in terms of the position space and the feature space,is proposed and introduced,thereby enormous computation cost caused by the fast increase in the number of feature points is avoided,and the data relation accuracy get improved.Moreover,the dimensions of SIFT feature vector is decreased to reduce computational complexity,and multiple conditional constraints is added into the process of matching such feature points in the left and right images to enhance the precision of the stereo matching.The experimental results indicate that,utilizing the methods proposed,the distributed-uniformity of the feature points can be realized and the cost of localization time is condensed on the premise of satisfying the requirements of localization accuracy.