Stereo Image Dense Matching Based on SGM Constrained by Feature Matching

Dense image matching plays a important role in stereo vision and remote sensing. Semi-global matching (SGM) is a widely used dense image matching framwork, which uses dynamic programming calculation strategy to optimize the global energy function. SGM has the problem of wrong disparity accumulation along the path. To solve the problem, this paper proposes a new dense image matching method based on feature matching and SGM. First, the feature matching is used to extract the reliable reference points. And then, reference points are used as additional constraints in matching cost aggregation to cut off the wrong disparity delivering along the path. In order to evaluate our method, qualitative and quantitative experiments are carried out on three stereo image pairs. The experimental results show that our method can achieve high matching accuracy by effectively reducing the error propagation.