Underwater simultaneous localization and mapping (SLAM) constitutes a critical foundation for marine vehicle navigation. However, underwater SLAM faces challenges such as high motion distortion and low matching efficiency. In response to this, we introduce a visual-inertial odometry (VIO) and sonar fused SLAM framework with enhanced corner feature matching. Initially, we employ VIO to remove the motion distortion. Additionally, we design a localized corner feature matching mechanism to enhance iterative closest point (ICP) SLAM, aiming to increase both the matching frequency and precision. Experimental results demonstrate that our approach outperforms the compared methods in terms of performance.