An extraction method in sub-pixel level was proposed to solve the high accuracy center point extraction problem of the light-spot image of the spacial light-emitting-spot. The gray level distribution mathematical model of the light-spot image of the spacial light-emitting-spot with Gauss energy distribution was established, as well as the proof was given for the perspective projection invariability of the spacial light-emitting-spot. Based on the analysis of the second-order directional derivatives and the self-correlation function of the light-spot image, the Hessian matrix was obtained which was used to derive the determinant terms for the pixel level center point. Furthermore, the Taylor expansion was performed on the image gray level within the sub-pixel neighborhood of the pixel level center point, and the maximum of the gray level curved surface of the light-spot image was taken as the wanted sub-pixel center point. Compared with the existing classical methods, the experiment results show that the proposed method has higher accuracy and better robustness which can be widely used in camera calibration and 3D shape measurement, etc.