THE 3D RECONSTRUCTION OF ROI BASED ON THE IMPROVED FEATURE FUSION AND MATCHING STRATEGY

The 3D reconstruction of the region of interest(ROI) has great applications, such as the virtual and augmented reality. The feature matching strategy is the key procedure in the 3D reconstruction for it determines the computational time cost and complexity. In order to speed up the acquisition of image feature points of pairs and get better reconstruction results, an improved matching strategy based on the complex shearlet transform is proposed for the 3D reconstruction. Firstly, the original images are clipped according to the masks to increase the proportion of ROI, and a fast image matching method based on the complex shearlet transform is proposed to accelerate and increase the robustness of the feature points of pairs; then, the mismatches are removed based on the Grid-Based Motion Statistics (GMS) and RANSAC; finally, the result can be obtained by the sparse and dense point cloud reconstruction. Experiments on four open datasets are implemented to show the accuracy and efficiency by comparing with the famous SIFT and SURF-based methods. It can be seen that more points of features in the ROI can be captured by the proposed method and thus better visual reconstruction results can be obtained.