Aiming at the problem of inaccurate localization of UAVs in weak texture scenarios, proposing a UAV localization method based on edge-based acceleration. Reference to the open source VINS-Fusion algorithmic architecture. Firstly, Harris corner point algorithm is used to extract the information of the corner points that need to be optimized, fusion of sub-pixel corner point algorithms for iterative and accuracy enhancement of extracted corner point information, provide good initial values for back-end optimization threads. Secondly, design a marginalization acceleration strategy, filtering visual residual information for optimization by sliding window method, using the Schur complement method to transform the filtered visual residual information into a priori information to be added to the optimization, splitting marginalized threads, and reconstruct the information matrix, indexing rows and columns of visual residual information, move the matrix block containing more information to the lower right corner of the information matrix, retain more a priori information. Finally, it is evaluated using the EuRoc dataset, and the experimental results show that compared with the open-source vision-inertial-guidance fusion SLAM system, our algorithm in this paper obtains a significant improvement in the positioning accuracy, while it can ensure a high computational efficiency and meet the real-time requirements of UAV positioning.