Automatic 3D Scanning System Via Depth Camera

Achieving 3D models of the object has always the top concern of fields such as computer graphics, computer vision, especially in animation, games, virutal real-ity(VR), it has a wide usage. In particular, three-dimensional data of the human body has the assessed value. The traditional ways to achieving 3D models use the structured light or laser scanner, the scanning system has a sophisticated calibration algorithm, and it has lower robust and higher complicated algorithm. Meanwhile it costs a much longer time to achieve data and does harm to the spread of 3D technology. In recent years, the depth camera have been much popular and their merits such as cheap, convenient to use, offer a wider prospect.Kinect Fusion and Dymaic Fusion are the real-time reconstruction system, and achieve the 3D models under the premise of zero interaction. Beacuse of the restriction of the principle, the algorithm are local and not able to get the data of the black region accurately. Meanwhile, Get the color information, we usually treat the average value of the multi-frame image color as the value of the points, but it will cause color blurry. When it comes to the human body scan,the algorithm we referred will be unable to get the information of black region, such as crown, we get vague color information of the body model, and it’s unacceptable for customers.Three important performance indicators are mainly reflected in three-dimensional reconstruction compatibility, the accuracy of the color and geometry of the model and automation. This paper presents a 3D self-scanning system with both compatibility and integrity, which can quickly and easily reconstruct the human body model with rich detail and texture information. This system’s main features are compatible with the existing depth of the scanner, and the model reconstruction just to complete the simple interaction. Compared to the previous practice, the main contribution of this paper is to propose a algorithm about the filling the model hole which is based on the distance field. Using the data-driven way, we can get the a better shape parameter. This algorithm makes it possible to fill out the shape of the head automatically without losing model details, and to avoid the defects brought by depth camera without being capable of scanning hair, and thus to scan more convenient and faster. Also a new automatic texture mapping algorithm is proposed to deal with depth camera, with data-driven of the depth data, it has the ability to calculate texture coordinates and balance the illumination difference between diversities pictures, thereby, the best texture effects have been achieved with satisfactory.