Machine vision-based detection methods have been widely applied in the detection of aircraft skin damage. During drone inspection processes, a key step is to spatially locate high-resolution detailed images of aircraft skin from multiple angles onto a three-dimensional point cloud model of the aircraft. This relies on the rigid registration of image center position coordinate point cloud with the aircraft 3D point cloud. To address the issues of low accuracy and poor robustness encountered by existing registration algorithms when dealing with heterogeneous point clouds with significant differences in density and low overlap, this paper presents a novel cross-source point cloud registration network. The network integrates multi-scale information from the point cloud and employs an attention mechanism to identify representative overlapping points. First, the network achieves initial correspondences using the multi-scale geometric features and positional information of the point cloud. Then, an overlapping feature guidance module predicts the overlapping score of the point cloud. By utilizing information interaction through the attention mechanism, the network combines point overlapping scores with fused features to filter out representative overlapping points, achieving precise correspondences in the point cloud. The network employs weighted singular value decomposition (SVD) to estimate two sets of transformation matrices, yielding the relative pose parameters of the point cloud. Experiments were conducted in an unsupervised manner. The experimental results on the ModelNet40 dataset and the aero object dataset aircraft measurement data showed that, compared to other existing traditional and learning-based methods, this approach demonstrated excellent performance in terms of registration accuracy and robustness.

Monocular visual positioning systems are valued for their low cost and straightforward calibration. However, the lack of real-scale information and the complexity of initialization processes limit their application in scenarios requiring accurate absolute positioning. Existing solutions present trade-offs: markerless approaches depend on environmental priors (e.g., fixed camera height constraints) for scale recovery, while marker-based methods typically necessitate that target patterns remain within the camera’s field of view throughout the process. To address these challenges, we propose a 3D target-assisted initialization method that enables scale recovery with just two target images. This modular approach can be seamlessly integrated into monocular simultaneous localization and mapping (SLAM) frameworks. We validated our proposed initialization method through integration with ORB-SLAM3 and semi-direct visual odometry (SVO). Experimental results demonstrated that our method provides real-scale information without compromising real-time performance, making it suitable for applications such as indoor navigation and industrial robot localization, where accurate absolute positioning is essential.

Grating fringe projection 3D measurement techniques are extensively applied in various fields. However, in high dynamic range scenarios with significant surface reflectivity variations, uneven greyscale distribution may lead to phase errors and poor reconstruction results. To address this problem, an adaptive fringe projection method is introduced. The method involves projecting two sets of dark and light fringes onto the object, enabling the full-field projection intensity map to be generated adaptively based on greyscale analysis. First, dark fringes are projected onto the object to extend exposure time as long as possible without causing overexposure in the image. Subsequently, bright fringes are projected under the same exposure settings to detect overexposed pixels, and the greyscale distribution of these overexposed points from the previous dark fringe projection is analyzed to calculate the corresponding projection intensities. Finally, absolute phase information from orthogonal fringes is used for coordinate matching, enabling the generation of adaptive projection fringe patterns. Experiments on various high dynamic range objects show that compared to conventional fringe projection binocular reconstruction method, the proposed algorithm achieves complete reconstruction of high dynamic range surfaces and shows robust performance against phase calculation errors caused by overexposure and low modulation.

Aiming at the problem that the traditional visual pose measurement depends on the known structure information or artificial mark of the target in the scene, a relative pose measurement method is proposed and then the corresponding system is designed based on feature matching. The relative pose measurement does not need to know the prior information of the target in the scene, and the system has high measurement accuracy. Firstly, the binocular camera is used to collect the sequence images of the targets in the scene, the computer uses the accelerated KAZE (AKAZE) algorithm to extract feature points of the image, the improved k-nearest neighbors (KNN) and random sample consensus (RANSAC) algorithms are used to perform feature points matching on adjacent images and eliminate mismatched points. Afterwards, the three-dimensional coordinates of the feature points are obtained by triangulation measurement and bundle adjustment optimization, and the three-dimensional feature point library is established by using three-dimensional coordinates and two-dimensional image feature vectors. During pose measurement, the monocular camera is used to collect the image of the scene target, and the AKAZE algorithm is used to extract feature points of the image to be measured. The obtained feature points are matched with the three-dimensional feature point library, and then EPnP + Gauss-Newton method is used to solve the relative pose. In the experiment, a high-precision turntable is used to rotate camera, and the camera takes multiple images for measurement. The results show that the maximum measurement error of the designed pose measurement system is less than 0.2° in the range of -20° to 20°, which can meet the application requirements.

The traditional tongue diagnosis process has the problem of poor objectivity. Applying computer vision technology to tongue diagnosis can effectively promote its objectivity. Binocular stereo vision combined with structured light fringe projection technology is a common method for 3D measurement. However, in the measurement scenario of tongue diagnosis, due to the presence of saliva and fluids on the tongue surface, there are high-reflectance areas with significant random distribution in the fringe images, leading to errors in phase calculation and point cloud loss. A trinocular measurement system was proposed based on fringe projection, where a trinocular system and three binocular subsystems were composed of three cameras. Dual-epipolar constraint based on phase and order constraints was introduced to enhance the accuracy of trinocular stereo matching. Supplementary matching points were utilized to optimize the trinocular matching point sets, reconstructing point clouds in high-reflectance areas. The results indicated that, compared to traditional binocular systems, this system achieved improved matching and reconstruction accuracy. Particularly in real tongue surface measurements, it could generate point clouds with clear textures and complete features. It could effectively measure the highly reflective area of the tongue surface and facilitate objective tongue diagnosis.

In visual measurement, high-precision camera calibration often employs circular targets. To address issues in mainstream methods, such as the eccentricity error of the circle from using the circle’s center for calibration, overfitting or local minimum from full-parameter optimization, and calibration errors due to neglecting the center of distortion, a stepwise camera calibration method incorporating compensation for eccentricity error was proposed to enhance monocular camera calibration precision. Initially, the multi-image distortion correction method calculated the common center of distortion and coefficients, improving precision, stability, and efficiency compared to single-image distortion correction methods. Subsequently, the projection point of the circle’s center was compared with the center of the contour’s projection to iteratively correct the eccentricity error, leading to more precise and stable calibration. Finally, nonlinear optimization refined the calibration parameters to minimize reprojection error and boosts precision. These processes achieved stepwise camera calibration, which enhanced robustness. In addition, the module comparison experiment showed that both the eccentricity error compensation and the camera parameter optimization could improve the calibration precision, but the latter had a greater impact. The combined use of the two methods further improved the precision and stability. Simulations and experiments confirmed that the proposed method achieved high precision, stability, and robustness, suitable for high-precision visual measurements.