Abstract: Objectives: The ranging performance of a LiDAR system is positively correlated with its echo reception efficiency. To extend the maximum detection range without modifying the hardware configuration, the relative position between the optical axis of the echo signal and the photosensitive surface of the photoelectric detector can be optimized to enhance the efficiency of echo energy reception. In current engineering applications, this alignment is predominantly performed manually—a process highly dependent on operator skill, resulting in high positioning errors, low efficiency, and poor repeatability. Automatic optical axis calibration offers a promising solution to improve assembly efficiency. However, existing calibration generally requires separate radial spot centroid detection and axial focus alignment steps, lacking an integrated framework. Moreover, their reliance on charge-coupled device (CCD) cameras for spot imaging increases system complexity and limits registration accuracy due to finite CCD resolution. To address these limitations, a calibration approach that directly utilizes the LiDAR's built-in photoelectric detector to acquire spot information is introduced, thus eliminating the need for external imaging components. Based on the three-dimensional intensity distribution characteristics of the echo signal, the adaptive three-dimensional centroid search process is proposed to achieve automatic calibration of the LiDAR echo optical axis. This approach both reduces hardware complexity and improves calibration accuracy. Methods: The photoelectric detector is mounted on a high-precision three-axis stage to scan spatial positions while recording echo intensity. Based on the three-dimensional intensity profile of the echo signal, an adaptive three-dimensional centroid search process is proposed to achieve automatic optical axis alignment. Additionally, to handle multi-peak intensity distributions on defocused planes, a two-dimensional multi-peak Gaussian spot model is established, along with an adaptive centroid detection method that enables high-precision extraction of centroid positions under complex intensity distributions. Results: Simulation experiments compared the proposed adaptive multi-peak centroid detection method against the scanning line method, particle swarm optimization, and gradient ascent algorithms. Results show that the proposed method accurately reconstructs the spot intensity profile under both unimodal and multi-peak distributions, while reducing detection errors through parameter optimization. It achieved a centroid localization root mean square error (RMSE) of less than 10 μm and a standard deviation under 5 μm, outperforming existing methods by more than 30-fold in accuracy and 10-fold in stability. Experimental validation confirmed that the adaptive three-dimensional centroid search process reduces alignment time from approximately one hour (manual) to under five minutes, with no degradation in ranging performance. Conclusions: The adaptive threedimensional centroid search methodology offers a novel and efficient solution for LiDAR optical axis calibration, demonstrating significant potential for practical deployment in LiDAR assembly processes.