With the development of artificial intelligence and the continuous deepening of visual surveillance applications, how to quickly and accurately locate targets in images and videos has become a hot research topic in the industry. This article proposes a two-dimensional feature point localization method based on a monocular vision ranging model to address the issues of low accuracy and speed in existing monocular vision feature point localization. This method fits the world coordinates of feature points by converting pixel coordinates into world coordinates and establishing a compensation model based on errors. The experiment shows that the average relative accuracy of world coordinates after error compensation reaches 98.590%, improving the accuracy of monocular visual positioning.