Optimization of SURF Algorithm for Image Matching of Parts

SURF algorithm, which is featured with high speed and good robustness, is a common algorithm for feature points detection. But this algorithm has such drawbacks as the separating capacity of its feature points descriptor is low and principal direction of feature points is not accurate, which can easily cause less image matching pairs and high mismatching ratio. Therefore, an improved SURF algorithm is proposed to increase matching pairs effectively and raise accuracy of matching. For the algorithm, two new-type feature sets are added, and 128-dimensional feature descriptor is established. The method of combining Euclidean distance with cosine similarity match is adopted to match feature points. Then RANSAC method is used to lower mismatching ratio, thus achieving robots’ parts-matching in the indoor environment. The experiment result shows that under the conditions of images rotating, zooming, blurring, lighting and view change, the improved algorithm has better robustness. Besides, its matching ratio and accuracy are higher than standard SURF algorithm.