多目鱼眼视觉SLAM特征点误匹配的剔除优化

Mismatched feature point elimination optimization for multi-fisheye visual SLAM system

为了解决由于鱼眼图像显著失真和明显的视角差异而导致的定位精度下降的问题,提出了一种去除多目鱼眼视觉SLAM系统中误匹配特征点的策略。该策略采用从粗到细的误匹配特征点去除过程,在粗匹配阶段,使用阈值为0.6的匹配算法获得粗匹配阶段的初始点集。在精细去除阶段,采用预校准的多摄像机模型将匹配的特征点转换为同一坐标系,并将其作为承载向量投影到单位球上,利用球上的外极约束去除不匹配的特征点。在去除不匹配的特征点后,在投影球面上建立重投影误差函数,对初始姿态点和映射点进行非线性优化。该策略在24 000张鱼眼图像上进行了测试,结果表明,该剔除策略显著降低了失配率,提高了初始化和定位精度,有效地提高了系统的定位精度和性能效率。

A strategy to remove the mismatching feature points in multi-fisheye SLAM system is proposed to solve the problem of decreased positioning accuracy caused by significant distortion and apparent viewing angle differences in fisheye images.This strategy uses a coarse-to-fine feature point removal process,using the initial set of points in the coarse matching algorithm with a threshold of 0.6.In the fine removal phase,a pre-calibrated multi-camera model was used to convert the matched feature points into the same coordinate system and projected onto the unit sphere as a carrying vector,removing the mismatch using the outer pole constraints on the sphere.After removing mismatched feature points,a reprojection error function was established to optimize the initial pose and mapping points.This strategy was tested on 24000 fisheye images.The results showed that the elimination strategy significantly reduced the mismatch rate,improved the initialization and localization accuracy,and effectively improved the positioning accuracy and performance efficiency of the system.