To address the issue of limited feature point extraction and subsequent accuracy degradation in low-texture scenes within the ORB-SLAM3 algorithm, an improved SLAM algorithm based on feature fusion of points and lines in low-texture scenes is proposed within the ORB-SLAM3 framework. To ensure both the quantity and quality of feature extraction, an adaptive threshold based FAST algorithm is employed to increase the number of feature points extracted. A joint constraint utilizing the Hamming distance threshold and geometric motion consistency is applied to remove mismatched points. A dynamic adjustment of the short line rejection threshold is dynamically adjusted according to the extracted number of line segments. Additionally, a grouping fusion strategy is adopted to merge potential fragmented, intersecting, and overlapping lines. The experimental results on the MH_05_difficult dataset demonstrate that the proposed algorithm outperforms ORB-SLAM3, achieving a reduction of 25.77% in mean absolute trajectory error and a decrease of 27.88% in root mean square error.