Research On 3D Reconstruction Technology Based On Kinect Depth Sensor

With the development of computer vision technology and the continuous improvement of the hardware devices, emerged a large number of visual sensor devices such as Microsoft's Kinect, ASUS's Xtion PRO LIVE, etc. Such devices are widely used in daily life and scientific research, such as robot navigation, virtual reality, three-dimensional reconstruction, has great research significance and value in use. Among them, the significance of three-dimensional reconstruction is particularly significant. 3D reconstruction through depth data acquisition, preprocessing, point cloud registration and fusion, and generating surface, describe the real scene into a mathematical model which is consistent with the expression of computer logic. This model can be of such as the protection of cultural relics, game development, architectural design and clinical medicine research to the auxiliary role. The traditional methods of obtaining 3D object images are laser scanning, structure light scanning, etc., these methods are expensive, complex assembly and complicated operation. How to achieve the low consumption and cost, high performance and accuracy of 3D reconstruction, has great significance for the practical application and scientific research.In view of the opportunities and challenges, this paper uses Kinect depth sensor as the point cloud acquisition equipment, to analyze and improve the principle and algorithm of 3D reconstruction process. A real scene 3D reconstruction method based on Kinect depth image is proposed. The main work and research contents of this paper are as follows:First of all, introduces the commonly used method and characteristics of 3D reconstruction, summarizes the research status. Then, combined with the Kinect ranging principle and mathematical model of depth image of error analysis, for depth image acquisition Kinect existing edge mismatching, invalid pixels and noises, this paper present a depth image enhancement method based on improved anisotropic diffusion algorithm and can effectively repair the black holes composed of invalid pixels in the original depth image. It can remain the detailed information of the object edge in the depth image while suppressing the noise. Finally, according to the point cloud data registration in the process of iterative closest point algorithm to find the corresponding recently point slower, lower the efficiency of the registration, error precision of the larger problem is proposed in this paper based on box structure of the ICP algorithm, through the classification and sorting of the structure of the box to clear the nearest point search range, thereby significantly improving the efficiency of point cloud registration. After point cloud registration, the Delaunay triangulation is used to realize the surface generation of point cloud model, and the process of 3D reconstruction of the whole scene is achieved.Experimental results show that the proposed method has a significant improvement in the speed and accuracy of the reconstruction results, and it has good robustness while suitable for automated production and deeper research applications.