In this paper, we provide a precision prediction method for high-definition map elements extraction based on sequential images taken by crowd-sourcing cameras. Generating the updated elements through crowd-sourcing images using multiple view geometry technique is preferred because of low manufacture cost and easy access of image data. Constrained by the angles of images taken, the reconstructed location precision of elements by vision are inferior to lidar, and could not satisfy the requirements of high-definition map updating. Hence, we propose the concept of position dilution of precision which reveals the theoretical triangulation error of 3D point triggered by 2D pixel error. As long as the predicated precision is accurate, even if the location precision of updating element not so precisely, we could still update the high-definition map by providing the confidence level of the element. Simulations and field images data collected are provided for the validation of our theoretical analysis. It is shown that the predicated variances are almost aligned with the actual measured results.