A real-time three-dimensional reconstruction method and system based on Kinect V2

Abstract

The invention discloses a real-time three-dimensional reconstruction method and system based on Kinect V2. The system comprises a constant-speed stepping motor electric rotating platform capable of timing, a Kinect V2 camera fixed to a tripod and a computer. According to the invention, based on a classical KinectFusion three-dimensional reconstruction algorithm, the filtering and denoising of thedepth image are improved, so that the reconstruction quality is improved; An objective function in an ICP algorithm is optimized, the robustness of a system is enhanced, RGB data and a point cloud model are added into a fusion algorithm of a model for splicing fusion, the model is endowed with colors to improve the visual visual visual effect, and finally a more accurate and specific three-dimensional point cloud model is obtained through a projection mapping algorithm. According to the system, the GPU is used for parallel acceleration processing of data, so that the reconstruction speed reaches about 25 frames per second, an object can be accurately reconstructed in real time, and the system can be widely applied to low-cost rapid three-dimensional reconstruction. Finally, the computer processing process is converted into the ARM microprocessor, a wifi transmission module is added, and a simple three-dimensional reconstruction device is obtained through integration.

Description

technical field [0001] The invention belongs to the technical field of computer vision, in particular to a real-time three-dimensional reconstruction method and system based on Kinect V2. Background technique [0002] Fast, low-cost, accurate and convenient acquisition of 3D models of objects has always been a research hotspot in the field of computer vision and graphics. Although image-based passive 3D reconstruction can obtain the 3D model of the object conveniently and at low cost, its feature extraction and matching depend on the lighting environment and the texture characteristics of the object surface, and the reconstructed point cloud is mostly sparse. The full name of Kinect v2 is Kinect for Windows v2sensor, which is Microsoft's second-generation Kinect peripheral, which uses a completely different TOF (time-of-flight) measurement technology from the first generation. Both Kinect v2 and the first-generation Kinect v1 are peripherals for the depth sensor hardware of...

AI Technical Summary

Problems solved by technology

Although image-based passive 3D reconstruction can obtain the 3D model of the object conveniently and at low cost, its feature extraction and matching depend on the lighting environment and the texture characteristics of the object surface, and most of the reconstructed point clouds are relatively sparse.

Images