Feature Point Matching Based on similar Attention Mechanism and Game Theory
Zhiyong Jiang · 2023
In the research of traditional multi-view 3D reconstruction based on RGB, The feature points in two views from different perspectives should be matched. Due to the high similarity between feature points of plant images, global matching will result in low efficiency and low accuracy. Therefore, a method for feature point matching of plants is proposed to reduce the search range of points and improve the matching efficiency. Firstly, a similarity attention mechanism is used to predict the predicted coordinates of adjacent feature points in the reference view corresponding to the matching point in the view to be matched based on the offset of the relative coordinates between the matched point pairs. Then, the feature points within a certain range centered on the predicted coordinates in the matching view will form the set of points to be matched. Finally, the optimal matching point for each feature point in the reference view is found in its corresponding set of matching points based on game theory. The experimental results show that the average matching accuracy of the proposed algorithm is 90.50%, and the number of correct matching points is more than those of the contrast algorithms, which proves that the proposed algorithm is effective in feature point matching of plant multi-view 3D reconstruction.