改进ORB特征提取环节的视觉SLAM算法

Visual SLAM algorithm for improving ORB feature extraction session

为了解决传统ORB-SLAM2算法尺度不变性较差和光照环境变化复杂导致定位跟踪不稳定的问题,提出了一种基于B-Spline图像金字塔的自适应阈值ORB特征点提取方法。首先采用B-Spline图像金字塔的方法,将图像层层划分,随后,通过计算图像周围的特征点的灰度值来设置自适应阈值,以便阈值随着光照变化而自动调整,从而实现图像特征点的有效提取。对改进部分分别实验验证,在光照环境发生较大变化时,改进方法在特征提取时重叠点降低且提取范围更加均匀,在图像尺度发生变化时,改进方法的特征匹配数量提升了近1倍,在轨迹追踪实验中,改进方法得到的估计轨迹误差降低了20%以上。改进的ORB-SLAM算法能够提高在复杂环境下机器人的定位精度。

In order to solve the problems of poor scale invariance of the traditional ORB-SLAM2 algorithm and the instability of localization tracking due to complex changes in the lighting environment,an adaptive thresholding ORB feature point extraction method based on the B-Spline image pyramid is proposed.First,the B-Spline image pyramid method is used to divide the image layer by layer,and subsequently,the adaptive threshold is set by calculating the gray value of the feature points around the image so that the threshold is automatically adjusted with the change of lighting,thus realizing the effective extraction of the image feature points.The improved aspects were experimentally verified,and the results revealed several significant enhancements.In scenarios involving drastic changes in illumination,the improved method significantly reduced the overlapping points in feature extraction and offered a more uniform extraction range.When encountering changes in the image scale,the number of feature matches using the improved method nearly doubled.In trajectory tracking experiments,the improved method achieved a reduction in estimated trajectory error of over 20%.The enhanced ORB-SLAM algorithm has the potential to significantly improve the localization accuracy of robots in complex environments.