3D reconstruction of dense model based on the sparse frames using RGBD camera

With the popularity of consumer-grade depth cameras on mobile devices, 3D reconstruction based on the RGBD camera has once again become a hot topic in the field of 3D vision. The current application of RGBD cameras is mainly in the field of large-scale 3D reconstruction and VSLAM that rely on high-performance graphics cards to achieve real-time reconstruction and currently reach a relatively mature stage. The reconstruction of specific scene models ( such as objects, human bodies, human faces ) has a lot of application prospects in the fields of intelligent volume measurement, face recognition, motion capture, etc, attracting the attention of researchers. In this paper, a lightweight 3D reconstruction algorithm framework is designed for scene objects. The selecting frame module is introduced. The dense 3D reconstruction based on sparse frames is implemented on the Raspberry Pi 4B to improve the model reconstruction efficiency and we design the point cloud pose-processing module to improve the quality of the model reconstruction.