For the model reconstruction and pose estimation of non-cooperative rotating space targets with unknown model, the technology of graph-based optimization SLAM was applied to reduce the cumulative error in the pose tracking process by using 3D point clouds acquired through LiDAR. First, the relative pose between adjacent key frames was calculated by the Iterative Closest Point (ICP) algorithm, and the pose of the current key frame was obtained by the pose tracking method, thereby constructing the pose graph of the chaser spacecraft. Meanwhile, the global 3D signature GLAROT-3D (Geometric LAndmark relations ROTation-invariant 3D) was used to detect the loop closure, and adding the closed-loop constraint to the pose graph. Finally, the method based on pose graph optimization was used to adjust the pose and update the model point cloud. Experimental results show that in the simulation test, when the noise amplitude reaches 100mm, the attitude measurement error is less than 2°. In the field experiment, the attitude measurement error is less than 2.5°, and the target point cloud model is well reconstructed. Hence, the accuracy and the anti-noise ability of the proposed method can satisfy the mission requirements for the relative pose measurement of non-cooperative target.