For important camera calibration in the field of computer vision, a new target form, namely, a grid spherical target (GST) that is different from the spherical target, is proposed. The GST has advantages of spherical and checkerboard targets because of the grid on the sphere. And the latitude and longitude circles and the intersection points between latitude and longitude circles on the GST are used to calibrate the camera. Firstly, the Image of Absolute Conic should be obtained using the elliptic curves of latitude and longitude circles on the GST in the images. After obtaining the initial intrinsic and extrinsic parameters of the camera using the Image of Absolute Conic, optimum solutions of the intrinsic and extrinsic parameters are solved through nonlinear optimization by using the latitude circles and the intersection points of the latitude and longitude lines. Finally, the effectiveness of the GST-based method is proven in simulation and physical experiments.

Nonindustrial low-cost cameras have the advantages of cheap and simple structure, but have the disadvantages of low resolution and large image noise. When the existing camera calibration methods are used to calibrate nonindustrial low-cost cameras, high-accuracy calibration cannot be obtained. A high-accuracy calibration method using a high-accuracy planar target is introduced in this study to solve this problem. First, the initial values and the uncertainties of all image feature points are determined by the multiscale image analysis method. Then, the image disturbance factor is added to each target image feature point. In addition, the image projection error is established as the minimum objective function according to the homography matrix between the target plane and the image plane. Thus, the optimal coordinates of all image feature points are obtained by the nonlinear optimization method. Finally, the calibration of the intrinsic and extrinsic parameters of the camera will be achieved by using Zhang’s method according to the image feature points obtained from the previous step. Simulative and real experiments have been conducted to evaluate the performance of the proposed method, and results show that the calibration accuracy of the proposed method is at least three times that of Zhang’s method.

Automatic fault inspection, aiming at either a single target or an object of a position predictive in the image, is a common paradigm in railway maintenance. However, the fault detection of multiple targets with unknown initial locations is rarely conducted. In this paper, an automatic fault detection system for multiple targets based on time-scale normalization is proposed with the use of high-quality linear camera. For location of the multiple targets in the image sequences, it is important to make sure that the target and standard images are exactly aligned. To deal with this problem, the image distortion, caused by the velocity fluctuation of the moving train or the discordant movement between the linear scan camera and the train, must be corrected according to the size of the standard image. The scale-invariant feature transformation points based on Sobel gradient are first extracted and matched accurately with the geometric constraint. Then, the matched images are divided into subblocks in horizontal direction. The feature points in each subblock are quantized to a typical feature point. Finally, the time-scale normalization is used to achieve the accurate alignment for the target images. In the procedure of fault inspection, an image subtraction approach is presented for locating the positions of the potential fault regions. According to the priori knowledge, the type of the possible fault regions can be finally confirmed, and the customized high-level image understanding knowledge is used to identify the status of multiple targets. The practical application in detecting the abnormalities of China Railway High-Speed parts demonstrates that the proposed method exhibits an excellent performance in monitoring multiple targets with unknown positions in the high-speed railway.

In the product surface defect detection system, the detection accuracy can be greatly improved if the surface grayscale image and depth image can be adopted simultaneously. Existing 3-D cameras have the problem in precise registration of 2-D images and 3-D data, which results in inflexibility and inevitable error in practical vision tasks. Hence, we propose a novel stereo vision measurement system which is able to simultaneously acquire registered scan grayscale images and depth images without special data registration algorithm as well as the calibration method. The proposed measurement system mainly consists of a line laser system, a line scan camera, and a frame camera. The line scan camera captures clear images of objects illuminated by the line laser. In addition, the line scan camera and the frame camera form a stereo vision sensor. In this case, the line laser stripe provides location feature of the matching point for frame camera. The epipolar constraint implements the matching process between the line scan camera and the frame camera. Finally, the stereo vision measurement model is utilized to realize the 3-D coordinate reconstruction, and the z-axis coordinate is used to generate depth image. The proposed system can simultaneously obtain the image and depth information corresponding to each pixel by means of a push sweep. In the experiment, the measurement system can achieve the accuracy of 0.13 mm within a measurement range of 500 mm x 300 mmx 200 mm.

Images collected by linear scan cameras are stretched and compressed due to the speed regulation of trains. This condition changes the shape of objects and considerably increases the false and missed alarm rates. Moreover, small size and dense distribution of the key components in railway trains increase the difficulty in fault diagnosis. Therefore, a two-module troubleshooting and positioning methodology is proposed in this study for state diagnosis of key small components of high-speed running trains. Firstly, the image with deformation is reshaped to be exactly the same as the standard one via omnidirectional scale correlation normalization (OSCN), which performs well even in low-texture, high-light situations. Secondly, we propose three feature-enhanced models to expand the receptive field of deep feature maps. This novel detector, namely, refine-inception net (RIN), not only reduces the rate of missed and false detection, but also minimizes the influence of target size and occlusion. Augmentation is performed to increase robustness because the detector is data driven. Experimental results show that the optimal strategy combining OSCN and RIN can troubleshoot high-speed trains of different models with an accuracy higher than 99%. Our method can be extended to foreign object recognition on the train roof while maintaining railway safety.

A novel calibration method for the line-structured light vision sensor that only requires the image of the light stripe on the target using a movable parallel cylinder target is proposed in this paper. The corresponding equations between two ellipses obtained from the intersection of the light stripe and the target and their projected images are established according to the perspective projection transformation, and the light plane equation is solved based on the constraint conditions that the minor axis of the ellipse is equal to the diameter of the cylinder. In the physical experiment, the field of view of the line-structured light vision sensor is about 500 mm × 400 mm, and the measurement distance is about 700 mm. A calibration accuracy of 0.07 mm is achieved using the proposed method, which is comparable to that when planar targets are used.