Uniaxial MEMS-based 3D reconstruction using pixel refinement

Abstract

A uniaxial micro-electro-mechanical systems (MEMS) micro-vibration mirror can be used to construct a new type of fringe projection profilometry (FPP) system. In FPP system calibration, some pixels may be calibrated worse than other pixels due to various error sources, which will affect the final reconstruction accuracy. In addition, there are some difficulties in calibrating the MEMS-based system because a projector using the uniaxial vibration mirror does not have focusing optics and can only project unidirectional fringes. In this paper, we developed an FPP system using a uniaxial MEMS micro-vibration mirror. To solve the calibration problems, we propose a calibration model suitable for the MEMS-based system and a pixel refinement method. These pixels with relatively large calibration errors are called outlier-pixels, which will significantly increase the error of the following 3D mapping. Therefore, the pixel refinement method classifies all pixels based on a frequency distribution histogram of calibration errors during calibration and prevents outlier-pixels from participating in the following 3D mapping. The experimental results show that the proposed method can improve the accuracy of 3D reconstruction, and the feasibility of the self-developed system is verified.

Colorful 3D Reconstruction and an Extended Depth of Field for a Monocular Biological Microscope Using an Electrically Tunable Lens

“…Unfortunately, the switching of the objective or movement translation stage requires manual or mechanical operations, which makes the microscope complicated, bulky, complex, heavy, and expensive [ 18 ]. Moreover, manual or mechanical operation inevitably causes sample vibration that will affect the 3D reconstruction performance [ 19 ].…”

MHPCG:Multi-modal Hyperspectral Point Cloud Generation Based on Single RGB Image

“…In concern of above facts, effort devoted to realize flexible multi-modal data generation has been a hot-hoc trend. Technique of 3D reconstruction from image [12][13][14][15][16][17] presents another solution to get color texture as well as the 3D structure information without using multi-sensors. In recent years, there has available studies explore 3D reconstruction of hyperspectal images [2,18] for above problem solving.…”

“…It enables hyperspectral image generation from a RGB image, thus presents a new flexible way to get hyperspectral image at low-cost. Along with the widely emerging of deep learning approaches, there have emerged amazing achievements in both 3D reconstruction [12,13,16,17,22] and spectral reconstruction [20,21,23,24] for the sake of data-driven and computational imaging capability of deep learning. Therefore, there is a potential solution to achieve convenient acquisition of the multi-mode hyperspectral point cloud data from simple RGB imaging.…”

Research on visual SLAM system based on near-infrared perception

“…On one hand, it maintains the cost-effectiveness and lightweight nature of visual odometry, making it suitable for deployment on various platforms such as small unmanned aerial vehicles and compact wearable devices [20][21][22][23][24]. On the other hand, in contrast to active ranging sensors like LiDAR [25][26][27][28][29], visual odometry exhibits strong environmental adaptability and provides richer information. It performs well in environments with repetitive and uniform geometric structures [30], a crucial feature for scenarios such as underground tunnels [31][32][33].…”