In response to the issues of drifting and tracking failures encountered in the Visual-Inertial SLAM (VIO-SLAM) algorithm under complex lighting conditions, an improved algorithm is proposed. The algorithm, called PVIO-SLAM, performs illumination discrimination on images and applies adaptive logarithmic correction and CLAHE to different illumination. Then, the adaptive weighted fusion strategy proposed in this paper is used to fuse the pre-processed images, the pre-integral is used to reduce the matching range and accelerate the feature point tracking and matching, and the loopback check is introduced to improve the robustness of the algorithm. Experimental results on a public dataset demonstrate that, compared with the original algorithm, when loopback check is not used, the proposed algorithm achieves an average reduction of 35.75% in RMSE, which is further reduced by 14.96% when enabling loopback check. Experimental results in real-world scenarios indicate that the proposed algorithm yields motion trajectories that are closer to the ground truth. In summary, the proposed algorithm effectively improves the accuracy and robustness of VIO-SLAM under complex lighting conditions.