Studies On 3D Human Reconstruction From Multi-images

Recently, 3D modeling is an important and interesting problem in Computer Graphics, Computer Vision and Virtual Reality. The methods for 3D modeling are described as follows: the traditional geometry modeling techniques, the 3D scanning techniques and the image based modeling techniques. The traditional geometry modeling techniques create the 3D models with the geometry editing software by hand, whose reconstruction process is a painful and time-consuming task, and the users need to be highly skilled to the software, moreover, the reconstructed models may be quite different from the real objects. The 3D scanning techniques are more widely used with the 3D scanning devices which are very expensive, and the algorithms are quite complex to solve a mass of data from those devices. The image based modeling techniques are receiving more research focuses in the past few years. Image based modeling techniques use the optical system to scan the scenes and reconstruct the 3D models automatically by processing the photographs.Generally speaking, Image based modeling techniques reconstruct 3D models of objects directly from single image, multi-images from different views or videos, and the whole modeling procedure can be fully automated or facilitated by human interaction. The image based modeling techniques contain the reconstruction process of 3D geometries, surface textures, reflectance characteristics, lighting conditions and so on. Recently, image based modeling techniques are receiving more research focuses because they are of low cost , easy to handle and can generate models with very high precision. In this paper, we focus on the reconstruction process of 3D geometries, some algorithms have been implemented and the contributions are described as follows:We first carry the pretreatment to the image sequences and use a kind of algorithm with the combination of background difference and frame difference. Background model was obtained effectively and noises were removed using the methods in morphology.A survey on the recent image based modeling techniques is given in the view of Computer Vision. We summarize the fundamental principles of those methods and classify for those algorithms.We implement the algorithm based on visual hulls. This algorithm converts the traditional 3D cones intersection problem to the 2D silhouettes intersection problem. Compared with the other methods, our approach simplifies the necessary interpolation process in the algorithm and the result models are also ensured to be watertight.The feature points of the model shape can be extracted automatically. Then we can get the 3D skeleton of the human model by combining the discrete penalty function from the system called Pinocchio,while the problem confusing the topology of the human model can be avoided.