In order to solve the problem of the ORB-SLAM2 algorithm being unable to extract feature points stably and accurately under changes in environmental brightness, as well as the problem of constructing large visual maps on long and straight road sections due to the selection of a large number of redundant keyframes, an improved visual map construction method based on the ORB-SLAM2 algorithm is proposed. By using an adaptive threshold feature point extraction algorithm and utilizing the degree of dispersion of image grayscale values to adaptively threshold feature points, the system's ability to resist changes in environmental brightness is enhanced; Using the inter frame yaw angle and distance as the criteria for keyframe filtering, dynamically changing the keyframe filtering threshold through the number of inliers, removing redundant keyframes, and lightweighting the visual map; Conduct experimental verification based on the KITTI dataset and intelligent vehicle experimental platform. The experimental results show that this method can effectively improve the robustness and accuracy of visual map construction, reducing map storage capacity by more than 50%.