Research On Dynamic Transfer Performance And Load Identification Of High-speed Wheel Set

Image-based realistic 3D model reconstruction, which is mainly about how to recovery a digital model with a realistic texture from image sequence of different viewpoint, now become one of the most popular research areas in computer vision. It plays an active role in such fields as cultural relic protection, digital media entertainment like computer game and animation movie. Realistic 3D model reconstruction include tow steps:the first one is reconstructing the precise geometrical outline, and the second one is recoverying the realistic texture of model.According to this twe subjects, in this article we collcect and analyse the existing methods, and propose a complete soluction to solve the first subject. Then we improve the texture mapping algorithm to avoid the texture discontinuity and illumination discrepancy problem. The study of this article could be summarized as follows:1. We collect and analyse the existing methods of 3D reconstruction.Comparing the effect of these methods, we adapt patch-based multi-view stereosis algorithm to reconstruct the point cloud of the model. And then the suface reconstruction of point cloud can be done by using the poission reconstruction algorithm, achieving to recovery an accurate geometrical mesh. With the mesh model, we use the least-squre conformal maps algorithm to build a one to one map between the 3D space and 2D space, preparing to generate a texture map for the next step.2. The following two parts is key improvement of this article. After geometrical reconstruction and surface parameterizion, based on the Markov Random Field model, we reduce the texture discontinuity by optimizating the energy function. The energy function is composed from two terms. We use the angle between the viewing direction of the corresponding view and the face normal as the first energy term; the second energy term corresponds to the color continuity. And an efficient solution to this problem is Graph Cuts algorithm.3. To eliminate the texture area color difference caused by illumination discrepancy, in this paper, we introduce a color transfer methods and improve it as our global color correction methods. First of all, we transform the image from RGB color space to the lap space which can separate the luminance channel from the color channel better than other spaces. Then we choose a basic image as a target image, and normalize the other images'luminance to the basic one. We can get the transformation matrix through calculating the luminance difference between these images. And finally reversed the previous steps we can complete the color correction.By these jobs done, we can not only reconstruct a geometrical mesh model, but also recovery its texture map. Compined these two data, we can finallu render a realistic 3D model, and use it to produce an animation movie.