In recent years, sensor fusion technology has become increasingly important in fields such as autonomous driving. LiDAR and cameras are typically key components of multi-sensor systems. Calibration allows for accurate spatial alignment between them, enabling complementary information fusion, thereby enhancing the robustness and accuracy of the perception system. However, existing LiDAR-camera calibration algorithms are often limited in their applicable scenarios and lack sufficient precision, making it difficult to meet the demands of complex environments. Additionally, sensors may be subject to environmental vibrations that cause changes in pose, requiring additional human and material resources for re-calibration. To address these challenges, this paper investigates an automated extrinsic calibration method for lidar and cameras based on semantic constraints. By utilizing the MobileSAMv2 large model for image segmentation and leveraging the model’s zero-shot transfer capability in new scenes, the universality of the calibration scheme is improved. Since the large model itself lacks semantic information, pedestrian semantic information is introduced through the YOLOv5 object detection algorithm, and pedestrian masks are matched and filtered. Subsequently, 2D and 3D pedestrian semantic matching is conducted. High-level semantic matching is more robust compared to low-level feature point matching. Finally, semi-globally constrained online optimization is applied to successfully matched pedestrian targets, with the specific optimization goal of maximizing the number of 3D pedestrian projection points falling on pedestrian masks. This optimization method is more powerful and effective than some traditional methods. The automated extrinsic calibration method for LiDAR and cameras based on semantic constraints could achieve online, automated operation, exhibits high robustness, and yields satisfactory calibration results, meeting the demands of complex and variable scenarios.