This paper presents a novel calibration method for multi-camera systems that effectively addresses the complexities and inaccuracies associated with traditional calibration techniques. The method relies on the sparse point cloud of a 3D object. By establishing correspondences between 2D points in the images and 3D points in the sparse point cloud, we compute the projection matrix of the camera to be calibrated. We sequentially place the calibration object at multiple positions within the coverage area of the multi-camera system’s field of view, capturing images simultaneously with all cameras. By constructing an undirected graph that reflects the positional relationships between the cameras and the calibration object, and solving for the shortest paths between the camera vertices, we complete the calibration of the multi-camera system. This method’s advantage is its ability to simultaneously obtain each camera intrinsic parameters and the relative extrinsic parameters between cameras in a single capture process.