The invention relates to a remote sensing image road extracting method based on stereo constraint, which comprises the following steps: extracting and refining the edge of the left image of a stereo pair, thereby obtaining a binary image of the single pixel edge; carrying out shape constraint and vectorization on the binary image of the single pixel edge by using an improved phase grouping method; carrying out edge line node matching on the basis of grey scale correlations, thereby determining the homonymous image point of the edge line node on the right image; resolving node elevation on the basis of an image pair RPC model parameter or interior/exterior orientation element forward intersection method; and determining the road extraction result on the basis of the road edge line constrained by the node elevation. Starting with the stereo image, the invention extracts the elevation information of the road edge line, which is obtained after edge extraction, refining, shape constraint and vectorization, and establishes a model by which the branch nodes of the candidate road edge lines can be matched and the elevation information can be acquired by carrying out forward intersection on the homonymous point information; and according to the elevation characteristic of the road, the invention establishes elevation constraint conditions, carries out elevation constraint, and extracts satisfactory road edge lines from the candidate road edge lines, thereby realizing quick automatic extraction of road thematic information on the remote sensing image.