With the development of smart grid, meter reading based on machine vision is gradually replacing manual meter reading. In substations or some industrial scenes, the environment of the meter is relatively harsh, and it is often affected by different environmental factors, such as light and pollution. The traditional meter reading shows significant performance degradation in noisy environments, we propose an end-to-end algorithm for automatic identification and reading of single-pointer meters based on deep learning. This algorithm achieves the automatic identification and reading of single-pointer meters without requiring prior knowledge. The YOLOv5 network is employed for effective detection of the meter. Then the perspective transformation method is used to correct the tilted dial plate. In order to solve the problem of low accuracy of pointer and tick mark extraction in low quality images, we enhance the traditional U2NET network with a channel and spatial attention module. The CRAFT text detection algorithm and CRNN text recognition algorithm are utilized to determine the scale value of the dial plate. The experimental results show that the reading method proposed in this paper has higher applicability than the existing methods, and the average reading error is less than 5% in the industrial field.
26 December 2019 in International Journal of Control, Automation, and Systems
An omnidirectional vision system consisting of a fisheye camera and structured light projector has its own distinctive advantages towards other vision systems. Vision system provides a wide range of information by one single capture. Moreover, the laser feature is easy to detect and extract from the given image for further analysis. In order to obtain reliable results, the vision system must be calibrated. There have been improvements towards calibration techniques for vision systems with laser illumination. However, these techniques are not suitable for an omnidirectional vision system in a particular configuration. Motivated by the above-mentioned problem, this paper presents a novel omnidirectional vision system with laser illumination in a flexible configuration and a method for its extrinsic calibration. Flexible configuration means that both camera and laser plane are no longer parallel to the floor. The calibrated vision system is evaluated experimentally by using simulated and real data to demonstrate that the proposed calibration method is more robust and provides a higher measurement accuracy.
Omnidirectional three-dimensional (3D) measurement based on the binocular omnidirectional vision system attracts increasing attention in computer vision. A 360° horizontal field of view (FOV) and 180° vertical FOV are obtained by the binocular omnidirectional vision system with only a pair of images. However, a nonlinear relationship exists between points on image planes and in 3D space for the binocular omnidirectional vision system, and traditional calibration methods are no longer applicable. This study proposes a hybrid calibration approach, which effectively fuses the unified camera model and the backpropagation neural network with a virtual 3D target, in which the neural network improved with genetic algorithm compensates all kinds of distortions and errors. First, the unified camera model is used to calibrate the intrinsic parameters, extrinsic parameters, and a transformation matrix. Second, the pixel coordinates on image planes are converted to 3D coordinates by 3D reconstruction through binocular vision. The errors between actual and calculated 3D coordinates are obtained. In addition, the neural network improved with genetic algorithm is established and trained with pixel coordinates on image planes and errors. Experimental results indicate that the average measurement errors in the X and Y directions of the central area are respectively reduced to 0.1057 and 0.1548 mm, and those of the edge area are reduced to 0.1992 and 0.2748 mm. Compared with mathematical and neural network methods, the hybrid calibration method provides more stable, robust, and accurate results. It is proven suitable for complex calibration scenes with large FOV, high distortion, and strong nonlinearity.
Mobile omnidirectional structured light vision is increasingly used in scene perception and robot navigation. A wide range of information is obtained by means of the vision system by only one image and laser image features are detected and extracted easily and quickly. In this paper a novel calibration method for mobile omnidirectional camera based on structured light is presented. Firstly, a set of parallel laser planes is emitted on the walls of corridor as auxiliary targets by structured light and intersects with wall orthogonally. Secondly, the constraint relationship is analyzed between the vanishing points in fisheye images and intrinsic parameters of imaging model. Finally, effects of the laser stripes’ interval and the angle between the wall which contains laser stripes and ground on calibration results are evaluated. Compared to Scaramuzza method, the calibration method shows its superiority in terms of both feasibility and efficiency. The method with the characteristic of self-calibration since the planar target is replaced by actively projected laser stripes. The result illustrates that our method has the advantages of simple and feasible operation, but result is effective and accurate. The calibration parameters are independent of the laser stripes’ interval and the angle between the wall and ground. Therefore, the method of the mobile omnidirectional structured light vision presented in this paper can be applied to many areas.