The current mainstream visual-inertial SLAM systems primarily rely on point features derived from the visual front end to perform localization and mapping tasks. However, in environments with weak texture, reliance solely on point features can result in errors in matching and localization. To address this issue, we propose a fast visual-inertial odometry based on point-line features in the Manhattan World. We integrate an adaptive matrix and an edge suppression factor with the Edge Drawing algorithm to enable fast and adaptive line feature extraction. We then employ the Mean Structural Similarity metric to match inter-frame features, eliminating false matches and enhancing positioning accuracy. Comparative experiments with the classic visual-inertial algorithms Vins-mono and PL-VIO were conducted on the EuRoC public dataset, demonstrating a 14.54% improvement in global positioning accuracy, which shows the accuracy and adaptability of the proposed algorithm.