**2. System Overview**

The overall framework of this system is shown in the Figure 1. LIDAR and IMU measurements are the inputs for the system.

**Figure 1.** Overall framework of our LIO system.

The system can be divided into three parts.

First, *the preprocessing module*, the raw point clouds are de-skewed using gyroscope data and IMU pre-integration value. Current scan's point cloud is projected to the 2D image. The depth characteristic value is used to remove the outlier points. Image is used to segment the planar feature and cluster the rod-shaped feature information.

Then, *the LIO odometry module*, IMU pre-integration results are used to estimate motion pose. The scan-to-map between current frame and local map is performed. In the scan-tomap module, we introduce the graph optimization model which can enhance the speed and accuracy of the solution, and a sliding window-based way is applied to update and maintain the local map.

At last, *the global optimization module*, if the current frame is judged to be a keyframe, LIDAR scan-to-scan residuals, pre-integrated IMU residuals and loop residuals are optimized via the slide window optimization. Information of marginalization is used for prior constraints. Loop closure is detected and performed in an effective way, which is beneficial to reduce cumulative error.

According to this system, we get the 6-DOF pose estimation and a real-time updated global map. Exhaustive comparisons have been conducted to prove the superiority of our system.

We define notations and frame definitions throughout the article. (·) *<sup>W</sup>* is considered as world frame. In the LIO system, the origin of the world coordinate is identified as the first LIDAR frame (·) *<sup>B</sup>* is the body frame and (·) *<sup>L</sup>* is the LIDAR frame. Rotation is represented by rotation matrices *R* and quaternions *q*. So *R<sup>B</sup> <sup>W</sup>* and *<sup>q</sup><sup>B</sup> <sup>W</sup>* is the rotation from world frame to body frame, and *p<sup>B</sup> <sup>W</sup>* is the translation. ⊗ is defined as the multiplication between two quaternions.
