Most existing LiDAR-based SLAM methods rely on the assumption of static environments, which is often violated in real-world scenarios involving dynamic objects such as moving vehicles or pedestrians. To address this issue, this paper proposes a real-time LiDAR-Inertial Odometry framework with online dynamic object removal. The proposed method utilizes ground fitting results as a reference, selects points with distinct geometric features as seeds for region growing, and refines segmentation confidence through clustering to distinguish between dynamic and static elements. Extensive experiments on the KITTI dataset demonstrate the effectiveness of the approach, achieving competitive accuracy with an average absolute trajectory error of 0.51% and an absolute rotation error of 0.19 \(^\circ \) /100 m, respectively.