Research And Implementation Of Kinect Based3D Reconstruction

3D reconstruction is the process of using special device to digitalize the object in the real word. With this technique, users can play with the real objects in the digital world. Using the product of Microsoft’s Kinect as the depth sensor, this paper introduced a cheap, fast and easy to use way for3D reconstruction. Through the process of denoising, smoothing, aligning and view dependent texture mapping introduced in this paper, users can finally get the textured3D objects.In the completion of this paper, I independently completed the following tasks: Denosing the raw data obtained from the Kinect, align the depth images between RGB color images, segment the object from the whole image, convert the point cloud to3D model that can be displayed for Direct3D and view dependent texture mapping.Participated in the following tasks:Using Scale-invariant feature transform algorithm to calculate the transformation matrix between two mashes, using the iterative cloud point algorithm for multiple meshes alignments and using the surface reconstruction algorithm to merge aligned meshes to a single mash.With the help of my supervisor and literatures, I tried and modified a variety of algorithms to find the best way for our needs. The testing results will be shown in the following parts.The work in this paper can help users to reconstruct the objects without using the expensive equipments. They just need a very cheap Kinect and using this to capture a sequence of images. After that using the software introduced in this paper it can generate a usable3D model. As we can see in this paper, the final result is not accurate enough due to the equipment limitation. But the general outline is clear enough; it basically meets the daily application needs.