基于线特征的单目SLAM中的迭代数据关联算法

针对基于线特征的单目SLAM（同时定位与地图构建）中的数据关联问题，提出了一种基于线段端点Patch确认的迭代数据关联算法。算法依据近似共线和端点近似重合两个指标来获取线特征的最近邻关联对，使用基于线段端点Patch的定向匹配确认机制来剔除最近邻关联对中的错误项，同时以迭代处理的方式提高数据关联的准确性，综合利用了线特征间的几何约束和图像相似性约束。上述算法在公开数据集上进行了测试，与现有线特征数据关联算法的对比实验结果表明，该算法在满足系统实时性的前提下，在线特征关联对数目和关联准确率上均达到了较好效果。

Iterative data association algorithm for line-based monocular SLAM

Aiming at the data association problem of line-based monocular SLAM, this paper proposed a iterative data association algorithm based on confirmation of endpoint patch. The algorithm obtained closest feature line pairs according to approximate collinearity and approximate coincidence of endpoints, then used directional search of endpoint patches to remove the error pairs. Iterative approach is used to improve accuracy. The algorithm comprehensive used geometric constraint and image similarity of feature line pairs. The experiments are carried out on a public dataset. By comparison with other similar methods, it proved that the proposed data association approach not only satisfy real-time requirement, but also has a good performance on match amount and matching accuracy.