A Study On The Model Optimized Methods Of Polyhedral Visual Hull Based On Feature Points Matching

The 3D reconstruction based on the video image in real-time is an important problem of popular fields of computer, such as: computer vision, computer graphics. The accuracy of the model is even the center issue of these fields. After considering the popular polyhedral visual hull (PVH) reconstruction method, we find that there are no methods that make full use of the image feature information in order to make the model more accurate. Considering this problem above, this paper fullly considers the popular polyhedral visual hull reconstruction method and the optimizing solutions of it, and does some research about how to make full use of the image feature information to optimize the whole initial PVH model with the minimal impact on the rea-time reconstruction. The high-accuracy 3D model reconstructed can be used in interacting with objects between virtual world and reality, telemedicine, motion capturing in sports and peiformances and so on.1. This paper introduces a fast and accurate method of getting the corresponding image feature point pairs between the adjacent images. After getting the model using the PVH method, this paper extracts the feature points rapidly. Then calculate the corresponding epipolar of feature points in another camera. According to the relationship of spatial location, search the corresponding point of feature point on the polar by selecting the appropriate template parameter. By selecting the appropriate threshold, we determine whether this point is the corresponding point of the feature point and overcome the computational error.2. On the basis of analyzing various optimizing methods presently, we introduce a model optimizing method based on the image feature points matching. After getting the corresponding feature point pairs in the public visible areas of the adjacent images, considering the corresponding view points and using the principle of 3D geometry, we find the exact location of the model point that the feature point corresponds to. And triangulate the new point and the surface vertex that includes the point, then use the linear interpolation method to deal with the triangulated surfaces, and achieve optimizing the initial model of PVH.Finally, this optimizing method is verified through some experiments. And we compare the optimized model of PVH with the initial PVH model and the original object. The experimental results indicate that this method can optimize the original model of PVH well and make the optimized model more accurate.