Abstract: Objectives: Visual-inertial SLAM typically outperforms pure visual SLAM in indoor scenes characterized by low or sparse textures and varying lighting conditions. Nonetheless, most existing visual-inertial SLAM encounter challenges in detecting and tracking sufficient feature points. Moreover, the prior pose measurement information from the Inertial Measurement Unit is often underutilized, resulting in reduced pose estimation accuracy and limited robustness. Methods: An adaptive point detection approach has been developed to enhance the robustness of feature point detection in images. Additionally, the LSD line feature algorithm makes it easy to detect short lines and broken line features, and the performance of the algorithm is affected by the change of illumination, resulting in "over-extraction" or "wrong-extraction" of line features. Accordingly, an adaptive algorithm for extracting line features was introduced, utilizing edge-detected binary images and incorporating the removal of erroneous lines based on the geometry characteristics of the vanishing point. Following this, the algorithm integrates the visual measurements from point-line features with the pre-integration measurement of IMU to yield reliable outcomes for front-end pose estimation and initialization parameters in a loosely coupled manner. In the back-end section of our proposed SLAM method, a unified nonlinear minimization residual function is established for visual and IMU measurements through tight coupling, optimizing for obtaining precise pose of the image or camera. Results: Our SLAM method has been validated and tested on publicly available benchmarks, showcasing its performance through ablation experiments and qualitative as well as quantitative comparative analyses against several state-of-the-art visual-inertial SLAM algorithms. Conclusions: The results indicate that our algorithm improves average localization accuracy by at least 12% and displays significant robustness.