The invention relates to a feature-based stereo image matching algorithm, which comprises the following steps: acquiring feature points of a first image and a second image, and respectively forming feature chains; selecting a new feature point from feature chains, which are not matched with each other, of the first image; searching a corresponding point whose similarity with the new feature pointis higher than a threshold value in a second image search range, obtaining the similarity between the next feature point of the feature chain where the new feature point is located and the next feature point of the feature chain where the second image corresponding point is located, and if the similarity is higher than the threshold value, matching the two feature chains; and if the new feature point is not higher than the threshold value, obtaining the similarity between the two rear feature points of the feature chain where the new feature point is located and the two feature points of the feature chain where the second image corresponding point is located, and if the new feature point is higher than the threshold value, matching the two feature chains. The algorithm provided by the invention is equivalent to the existing algorithm in efficiency, but is more reliable than the existing algorithm, especially for images with detected vertical feature number.