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.