In order to eliminate the negative influence of positioning errors of industrial robots on measurement results in 3D robotic scanners, a multiocular vision measurement system is set up to locate the 3D scanning probe. In the multiocular vision measurement system, due to the large change in the distance between the camera and the marker, the reprojection error of the marker is difficult to evaluate the spatial positioning error. In addition, due to the influence of the camera calibration accuracy, the correction of the 3D coordinate by the bundle adjustment is not obvious. The bundle adjustment method is improved in this paper to improve the 3D reconstruction accuracy of the system. The image scale is introduced into the error equation of the traditional bundle adjustment method. The optimization of overall parameters is implemented in two steps based on the new error equation. Firstly, the intrinsic and extrinsic parameters of the camera are corrected, then the 3D coordinates of the measured point are corrected. The experimental results show that, in the measurement experiment of the length of the Creaform scale bar, the deviation between the measurement result and the reference value is less than 0.300 mm, and the standard deviation is not bigger than 0.110 mm, which fulfill the application requirements of the 3D robotic scanner.