Model Creation Of Man-made Object From Depth Image

Devoting to assist the artist and naive users in achieving their ideas rapidly,3Dmodels creation is regarded as a fundamental topic in Computer Graphics all the time.However, it is difficult to build a moderate complex model directly, and one of theproper methods is to deform available models to fulfill our requirements, which seemsto be sensible and convenient comparatively.This paper is concentrating on the remarkable market of3D models’ applications,hoping to facilitate the man-made objects creation, and propose a new method takingadvantages of both scanning means and model-driven approaches for modeling. In thisway, our method could inherit the characteristics of keeping fidelity from scanningtechnique, as well as remaining excellent structural composition from the latter method.This combination has fully utilized their superiorities to avoid mutual weaknesses, andoffered a novel approach for3D modeling technique of man-made objects. Our workwill be introduced separately in four parts.At first step, we will introduce a novice depth camera, Kinect, for capturing scenedata, and then present our method for preprocessing of depth data about filling holesand smoothing noises.The second step is about preprocessing of color data of capturing scene, extractingour objective from scene and segmenting it into rational parts.The next step involves semantic analysis of man-made object model sets, includingconsistent segmentation, establishment of parts’ controllers and extraction of structuralpatterns.In the final part, we will focus on retrieval of candidate models, determiningcorrespondence between candidate model and scene data besides subsequent modelingprocess and final restructuring procedure. All above procedures will provide a completemodeling process which could approximately reproduce whatever we observe in dailylife, intriguing inspirations in3D modeling.As well as utilizing capturing scenes data as constraint for our3D model creation,we are also exploiting available model sets, so that we can maintain modeling results’inherited structures and keep them from being destroyed by structural artifacts, even ifthere are large deformation in model size, parts’ positions and scales. After examiningin different scenes, it can be proved that our method would work successfully, satisfyour objectives and produce fine results.