Three-dimensional Reconstruction Based On Target Area Of Binocular Stereo Image

With the further development of virtual reality,artificial intelligence and other fields,3D reconstruction technology is playing an increasingly important role.The 3D reconstruction method based on binocular stereo vision is one of many reconstruction methods.Because of its low cost and it is not easily affected by environmental factors,it has attracted the attention of many researchers and institutions.In order to reconstruct a high-quality 3D model,this paper conducts indepth research on the issues related to the reconstruction of 3D models based on binocular stereo images.The main tasks are as follows:To reduce the influence of other non-target regions and improve the quality of the final 3D model,we use the target region of stereo image pairs to reconstruct 3D model.We focused on the study of how to obtain stereo images with target regions and the process of obtaining disparity maps through stereo matching.We verify the validity and accuracy of this method by experiments on public stereo image data.In the process of acquiring target regions of stereo images,region growing method based on matched feature points is proposed.We use A-KAZE feature detection and DAISY descriptor to extract and match feature points in stereo images.Firstly,the feature points can be detected in the non-linear scale space,and the DAISY descriptors can describe the feature points.Random Sample Consensus is used to ensure the accuracy of matching feature points.The experimental results on different stereo images show that this method has a higher match rate.After that,the matching feature points of the target area are used as the seed point pixels,and the region growing method is used to obtain stereo images with consistent target regions.In the process of calculating disparity maps based on stereo images,we use the gradient features of the pixels and the LBSP features to calculate the matching cost.The non-local cost aggregation method is used to aggregate the matching costs and calculate the initial disparity map,and in the disparity optimization step,the matched feature point's disparity information can be used to improve the accuracy of disparity map.The experimental results show that the method can obtain relatively accurate disparity map.Finally,the dense point clouds can be obtained through the disparity map.We can obtain a 3D model with texture,after the point clouds are denoised and meshed.