Calibration of the external parameters of cameras and LiDAR is a prerequisite for achieving high-precision fusion of multimodal data. Existing target-based calibration methods are found to compromise calibration accuracy due to the lack of distinct geometric features. Therefore, a joint external parameter calibration method for LiDAR and cameras, which is based on a centroid guidance strategy, was proposed. A multi-feature point calibration target with centroid constraints was designed, and the coordinate information of feature points was acquired from both point cloud and image data by integrating the geometric parameters of the calibration target. Furthermore, the Huber loss function was incorporated into the quasi-Newton method to further optimize the joint external parameters. Finally, 3D object detection was performed in real-world scenes to verify the effectiveness of the proposed method. The centroid guidance strategy was introduced. By improving the quasi-Newton algorithm for external parameter optimization, the average reprojection error is reduced by 8.4% compared with the four-feature point calibration method. Comparative experiments demonstrate that the improved quasi-Newton optimization algorithm has a beneficial effect on enhancing the accuracy of joint external parameters.