With emerging technologies such as autonomous driving and mobile robotic navigation, multi-sensor fusion that combines the Global Navigation Satellite System (GNSS), Inertial Measurement Unit (IMU), and Light Detection and Ranging (LiDAR) is of critical importance to achieve continuous and robust positioning. To improve the GNSS-IMU-LiDAR fusion positioning performance, a two-step LiDAR keyframe selection strategy was designed, which considers the consistency between normal distribution transform registration and IMU pre-integration and sequential changes of the Euler angle and Euclidean distance. This strategy is adaptive to the carrier motion state and can avoid missing keyframe selections. To address the insufficient accuracy in the back-end IMU pre-integration algorithm, a refined pre-integration model was proposed, which incorporates the Coriolis acceleration, changes in the Earth’s gravity, and centripetal acceleration due to carrier motion. A series of vehicle experiments were conducted and compared with LIO-SAM to evaluate the effectiveness of the proposed method. With more selected keyframes and reduced pre-integration error, the trajectory was smoother, and both positioning accuracy and yaw angle accuracy were improved. Our implementation is available at https://github.com/xdinav .