【摘要】 为了提高SLAM系统特征提取和匹配的稳定性,文章提出了在全景展开图像的多尺度空间上提取特征的方法。该方法首先对全景图像进行基于虚拟柱面的展开和矫正,然后在此基础上为展开图像建立多尺度空间,最后在多尺度空间内通过Harris特征跟踪器来选择候选特征点,并用Laplace公式选择出在尺度空间范围内具有局部最大值的特征点。这些特征点具有尺度、旋转和光照不变性,并且在有限视角变化下也能保持稳定性。实验证明,这种方法比在未展开图像或展开图像的单尺度空间内提取特征更具稳定性。更多还原

【Abstract】 In order to improve the stability of feature selecting and matching,a method is proposed to abstract features on the unwrapped panorama’s multi-scale space.At first,the panoramic image is unwrapped and rectified based on the virtual cylinder.Then multi-scale space is built for the unwrapped panoramic image.Finally the candidate features are selected by Harris dictator and by Laplace function that decides local maximum features in multi-scale space.Those features have the characteristics of scale invariance,rotating invariance as well as illuminating invariance,and will be stable under the conditions of limited changes of visual angles.Experiment proves that this method is more stable than abstracting features on wrapped panorama’s multi-scale space and unwrapped panorama’s single-scale space.更多还原