Optimization on stereo correspondence based on local feature algorithm

Stereo correspondence is one of the most important steps in binocular stereovision. It consists feature point extraction and image matching. In order to solve the problems of bad anti-noise performance and low accuracy of image matching in Scale Invariant Feature Transform (SIFT) algorithm, an optimized matching method based on local feature algorithm with Speeded-up Robust Feature (SURF) is proposed in this paper. In terms of feature extraction, SURF feature descriptor has a good anti-noise performance, which is extended from 64 dimensions to 128 dimensions makes the descriptor more specific, and the matching method is improved. The average value of the feature distance is used to replace the second neatest distance of the original matching algorithm, and Random Sample Consensus (RANSAC) algorithm is used to eliminate the wrong matching pairs. Test results indicate that the change of SURF feature points numbers in Gaussian noise is no more than positive or negative 15%, while the change of SIFT is more than 50%. In addition, the matching accuracy of the proposed method is increased by 20.5% compared to the original method of the shortest Euclidean distance between two feature vectors. Based on such result analysis, SURF algorithm with optimization matching method makes the matching accuracy more effective and has a practical value.