The autonomous positioning technology of a multi-camera system can restore the spatial position and pose of the system by using the multiple cameras to observe the feature points in space, which can overcome the influence of the complicated environment and improve the precision of measurement by using the large field of view of multiple cameras. Aiming at the problem of the complex structure of a multi-camera system, posture recovery is rather difficult and time-consuming, a self-positioning method based on the graphic optimization model is proposed. On the basis of solving the efficient perspective-n-point-positioning problem to obtain the approximate position and pose of a multi-camera system, a graph optimization framework is used to model the observation problem of the multi-camera system and the spatial control points. Thus, the problem of the position and pose recovery is equivalent to the problem of the minimization reprojection error nonlinear optimization. Using a control field build by a video-simultaneous triangulation and resection system (VSTARS) and a multi-camera system build by a linear guideway, the results of the measurement and simulation experiment show that this method possesses a high measurement accuracy and a fast running speed.