The Simultaneous Localization and Mapping (SLAM) method based on monocular vision is one of the most significant research directions for the robot autonomous walking, and the image feature matching technology is the key technology of this method.The SLAM matching method based on Scale Invariant Feature Transform (SIFT) algorithm has the advantage of abundant and stable feature points, but it also has some disadvantages in speed and accuracy.Therefore, considering that the SIFT descriptor has high dimensions and long matching time, an improved SIFT algorithm is proposed.The original 128-dimensional feature descriptor is reduced to a 24-dimensional feature descriptor with a rectangular inner side and round outer side.In the matching process, trilinear interpolation and RANSAC algorithm, etc.are applied to eliminate the mismatching of matching results.The experimental results finally show that the improved SIFT algorithm not only has good robustness to the changes of angle and illumination conditions and improves the matching speed and accuracy, but also meets the needs of SLAM synchronous map construction.