ing at the problem of the complexity of the existing global calibration method for multi-camera system with non-overlapping fields, an accurate calibration method with non-overlapping fields of view (FOV) cameras based on spatial constraints is proposed. Firstly, spliced small targets distributed in the field of respective camera are fixed together to make up a large calibration planar. A large-field camera is used to measure the spliced small targets, and the obtained relationship between the calibration plates is used as the correlated parameter between cameras. Then, small targets in the fields of view are obtained by non-overlapping FOV cameras, the reprojection error functions of cameras can be obtained based on spatial constraints, and the transformation matrix between non-overlapping FOV cameras is nonlinearly optimized by the Levenberg-Marquardt algorithm. The experimental results show that the reprojection errors of the global calibration method are 0.33 mm and 0.57 mm for the X-axes and Y-axes, respectively. It achieves global calibration of non-overlapping FOV cameras with high precision and stability.