Visual-inertial SLAM improves the accuracy of map construction and positioning by integrating visual and inertial constraints, but the degradation problem of feature extraction in the environment of changing illumination leads to tracking failure. A monocular depth estimation visual-inertial SLAM algorithm based on VINS-Mono with improved adaptive image enhancement is proposed. Firstly, adaptive gamma correction is implemented to adaptively enhance the texture details and brightness of low-light images; secondly, a lightweight LapDepth depth estimation network is constructed to directly construct the pose through depth information and PNP to improve the pose accuracy; finally, the adaptive image enhancement algorithm is verified on the public dataset LOL, and the real-time and accuracy of the depth estimation network are evaluated on the KITTI and NYU Depth V2 datasets. The positioning accuracy of the SLAM algorithm is tested on the EuRoc light change dataset, and the absolute trajectory error (ATE) has achieved the highest improvement of 49.4% and 55.0% in both RMSE and STD. Finally, the real-time performance of the system is analyzed.