Research on an environmental adaptive feature extraction algorithm

Abstract The visual simultaneous localization and mapping (SLAM) system is prone to tracking failure in complex and variable environments and lighting change environments. Therefore, an adaptive S-piecewise from an accelerated segment test (FAST) and boosted efficient binary local image descriptor or BEBLID (ASFB) feature extraction algorithm is proposed to obtain better matching results and more accurate poses. The algorithm designs a segment function adaptive thresholding FAST feature extraction algorithm based on the sigmoid function trajectory. It also calculates different thresholds for each pixel point in the image to complete the feature extraction, which adaptively calculates the threshold value for different luminance environments, to improve the quality of feature points, and uses BEBLID instead of binary robust independent elementary feature (BRIEF) to complete the description of feature points, which improves the accuracy of description information. The experiment selects KITTI and Euroc datasets, and the results show that the ASFB feature extraction algorithm performs better than the oriented FAST and rotated BRIEF algorithm in both the illumination changing environment and the normal environment. In addition, the bit pose accuracy is also improved. Moreover, it can meet the operational efficiency requirements of the SLAM algorithm.