【摘要】 点线特征融合可以使基于特征法的视觉同步定位与建图（SLAM）在不同的纹理场景下均能实现连续、准确的跟踪，但存在算法实时性较低的问题。为了提升计算速率，研究长度、线密度、线聚集度这3种评价指标与线特征精度的相关程度，通过理论分析和实验验证，表明线聚集度与线特征精度的相关程度较高。以此为基础，提出一种基于线聚集度进行线特征筛选的方法，限制线特征的最大提取数量，仅保留精度较高的线特征；同时，将全局暴力匹配改进为局部搜索匹配，以缩小待选线特征的范围、减少计算描述子间距的次数。实验结果表明，当线特征的最大提取数量为150时，算法的精度和实时性较高，综合性能较好；按照线聚集度进行筛选后，在不损失精度的基础上，算法实时性最大可提升32%；采用局部搜索匹配后，改进后相比改进前，在不损失精度的基础上，实时性最大可提升约35%。更多还原

【Abstract】 Point-line feature fusion enables feature-based visual Simultaneous Localization and Mapping(SLAM)to achieve continuous and accurate tracking across scenes with varying textures,but it suffers from limited real-time performance. To improve computational efficiency,we investigate the correlations between line length,line density,line aggregation degree,and line feature accuracy. Theoretical analysis and experiments demonstrate that line aggregation degree is strongly correlated with line feature accuracy. On this basis,we propose a line feature screening method based on line aggregation degree,which limits the maximum number of line features extracted and only retains line features with high accuracy. Additionally,global brute-force matching is replaced with local search matching,thereby narrowing the search space for candidate line features and reducing the number of descriptor distance calculations. Experimental results show that when the maximum number of line features is set to150,the algorithm achieves higher accuracy and better real-time performance,yielding the best overall performance. After screening based on line aggregation degree,real-time performance of the algorithm is improved by up to 32% without sacrificing accuracy. After adopting local search matching,real-time performance is improved by up to about 35% without sacrificing accuracy.更多还原