Multi-camera systems are widely applied in many fields, but the camera calibration is particularly important and difficult. In the application of a multi-camera system, it is very common for multiple cameras to be distributed on both sides of the measured object with overlapping field of view. In this paper, we present a novel calibration method for a multi-camera system based on flat refractive geometry. All cameras in the system can acquire calibration images of a transparent glass calibration board (TGCB) at the same time. The application of TGCB leads to a refractive phenomenon that can generate calibration error. The theory of flat refractive geometry is employed to eliminate the error. The proposed method combines the camera projection model with flat refractive geometry to determine the intrinsic and extrinsic camera parameters. The bundle adjustment method is employed to minimize the reprojection error and obtain optimized calibration results. The simulation is performed with zero-mean Gaussian noise of the standard deviation changes from 0 to 0.4 pixels, and the results show that the error of rotation angle is less than $5.6\mathrm{e}-3\mathrm{deg}$, and the error of translation is less than $4.6\mathrm{e}-3\mathrm{mm}$. The four-camera calibration results of real data show that the mean value and standard deviation of the reprojection error of our method are $4.3411\mathrm{e}-05$ and 0.4553 pixels, respectively. Both the simulative and real experiments show that the proposed method is accurate and feasible.

Multi-camera systems are widely applied in the three dimensional (3D) computer vision, especially when multiple cameras are distributed on both sides of the measured object. The calibration methods of multi-camera systems are critical to the accuracy of vision measurement and the key is to find an appropriate calibration target. In this paper, a high-precision camera calibration method for multi-camera systems based on transparent glass checkerboards and ray tracing is described, and is used to calibrate multiple cameras distributed on both sides of the glass checkerboard. Firstly, the intrinsic parameters of each camera are obtained by Zhang’s calibration method. Then, multiple cameras capture several images from the front and back of the glass checkerboard with different orientations, and all images contain distinct grid corners. As the cameras on one side are not affected by the refraction of glass checkerboard, extrinsic parameters can be directly calculated. However, the cameras on the other side are influenced by the refraction of glass checkerboard, and the direct use of projection model will produce a calibration error. A multi-camera calibration method using refractive projection model and ray tracing is developed to eliminate this error. Furthermore, both synthetic and real data are employed to validate the proposed approach. The experimental results of refractive calibration show that the error of the 3D reconstruction is smaller than 0.2 mm, the relative errors of both rotation and translation are less than 0.014%, and the mean and standard deviation of reprojection error of the four-camera system are 0.00007 and 0.4543 pixels, respectively. The proposed method is flexible, highly accurate, and simple to carry out.

Continuous point cloud stitching provides accurate and rapid environmental information for mapping, which plays an essential role in autonomous vehicles. Many methods related to autonomous vehicles are either inaccurate or time-consuming when they reconstruct large scenes. In this work, a feature matching score-based stitching method to reconstruct environments for autonomous vehicles is proposed. The purpose of the method is to address the problems of stitching quality and time consumption by using the feature match- completeness scores. The feature matching score is obtained by calculating the matching rate within an adjacent image pairs that produces a frame of point cloud. The score maintains the correct points, so fewer points require to be stitched. Then, the corresponding criterion is given to evaluate the performance of stitching candidate point clouds. Finally, the points are fit into planes to form a complete and comprehensive scene. The proposed stitching algorithm was verified in several experiments to demonstrate the efficiency and the accuracy of the overall system.

The hollow turbine blade plays an important role in the propulsion of the aeroengine. However, due to its complex hollow structure and nickel-based superalloys material property, only industrial computed tomography (ICT) could realize its nondestructive detection with sufficient intuitiveness. The ICT detection precision mainly depends on the segmentation accuracy of target ICT images. However, because the hollow turbine blade is made of special superalloys and contains many small unique structures such as film cooling holes, exhaust edges, etc., the ICT image quality of the hollow turbine blades is often deficient, with artifacts, low contrast, and inhomogeneity scattered around the blade contour, making it hard for traditional mathematical model-based methods to acquire satisfying segmentation precision. Therefore, this paper presents a deep learning-based approach, i.e., the enhanced U-net with multiscale inputs, dense blocks, focal loss function, and residual path in the skip connection to realize the high-precision segmentation of the hollow turbine blade. The experimental results show that our proposed enhanced U-net can achieve better segmentation accuracy for practical turbine blades than conventional U-net and traditional mathematical model-based methods.

The forward traffic environment cannot be fully perceived by a single sensor or multiple homogeneous sensors in the driving condition. Therefore, it is necessary to fuse the heterogeneous sensors to realize the mutual cooperation and compensation. Aiming at the advantages and disadvantages of different vehicle sensors, millimeter wave radar and monocular camera are selected as vehicle sensors to perceive the forward information. Firstly, the coordinate relationship between the sensors is calculated and sensor coordinates are mapped into the same vehicle coordinate system. Secondly, target information from the sensors are processed in parallel. Thirdly, in the case of time and space synchronization of two kinds of signals, the signals are processed by the matching algorithm of Global Nearest Neighbor (GNN) and the two matched targets are combined into one by the weighted average method. Finally, the unmatched targets and matched targets are tracked to determine the final state by Extended Kalman Filter(EKF) algorithm. Results in real car environment reveal that the fusion method can make up the deficiency of single sensor and improve the recognition rate of target.

Multi-camera calibration plays an important role in many field. In the paper, we present a novel multi-camera calibration method based on flat refractive geometry. All cameras can acquire calibration images of transparent glass calibration board (TGCB) at the same time. The application of TGCB leads to refractive phenomenon which can generate calibration error. The theory of flat refractive geometry is employed to eliminate the error. The new method can solve the refractive phenomenon of TGCB. Moreover, the bundle adjustment method is used to minimize the reprojection error and obtain optimized calibration results. Finally, the four-cameras calibration results of real data show that the mean value and standard deviation of the reprojection error of our method are 4.3411e-05 and 0.4553 pixel, respectively. The experimental results show that the proposed method is accurate and reliable.