In this paper, a feature matching method for LiDAR based on Kalman fusion is proposed to address the problem of inaccurate localization using line segment features in existing LiDAR feature matching algorithms. Firstly, a frame of LiDAR data is scanned, and local map is generated by using an improved method for extracting line segment features. The rotation and translation parameters of the partial map are then determined, and the partial map is matched with the global map to obtain the matching result according to the relative deviation. Then, based on the Kalman filter, the IMU data is used to predict the estimation for the next moment, and the LiDAR matching result is used as the observation. Finally, the two results are fused to obtain the optimal estimation. The experimental results show that this method is more accurate in matching line segment features compared to the existing feature matching algorithms, which leads to better precision and robustness in localization and navigation.