A calibration method of computer vision system based on dual attention mechanism
Introduction
With the development of computer vision, the calibration method of computer vision system has been paid more and more attention. Nowadays, system calibration has a very wide range of applications. As a hot field in computer vision, it is widely used in the field of industrial, agricultural, measurement, medical, aerospace etc. Therefore, research on computer vision system calibration method has a broad prospect.
One of the basic tasks of computer vision is to calculate the geometric information of objects in 3d space from the image information obtained by camera. Then, the object is identified according to the geometric information. The relationship between the three-dimensional geometric position of a point on the surface of a space object and its corresponding point in the image is determined by the camera imaging geometric model, and these geometric model parameters are camera parameters.
This paper introduces the dual-attention mechanism which a common algorithm in natural language processing into the image visual processing and calibrates the image system with the combination of Zhang Zhengyou algorithm. Attention mechanism has always been a well-known mechanism in the field of natural language processing. It can effectively extract the deep meaning of the text context through the attention mechanism, and the feature area of the image can be extracted by introducing the attention mechanism into the field of image processing. On the basis of image acquisition, various feature information in the visual image is processed, analyzed and calculated. Combined with the Zhang Zhengyou algorithm, it can realize the measurement of the three-dimensional geometric size, morphology and position of the measured object, and the algorithm has good versatility, strong applicability and high accuracy.
Section snippets
Related works
As one of the most important steps in visual measurement, many scholars at home and abroad have studied it deeply and put forward a variety of calibration algorithms. At present, the most widely used method is Zhang Zhengyou calibration method [1], which realizes the coordinate coordinate transformation between world coordinates and image coordinates. Compared with the classical two-step calibration method, the complexity of calibration quasi-algorithm is reduced, but the accuracy of the
Construction of dual attention model
The dual attention mechanism is a kind of resource allocation model. At a certain moment, people's attention is always focused on a characteristic part of the picture, while ignoring other parts [5]. The dual attention mechanism makes use of this principle. The structure diagram of the model used in this paper is shown in Fig. 1.
In the image module, the camera is first used to obtain images, which are input into ResNet for image processing, and the trained model is used to obtain the
Extract image feature area module
Image feature extraction has always been an extremely important branch of computer vision, and ResNet provides a learning model that is very suitable for image processing. The trained ResNet can learn the features in the image, and complete the accurate extraction of features. As an important branch of the neural network field, the advantage of ResNet is that can get the eigenvector of the next layer by the convolution kernel excitation which shares the weight of the previous layer [8]. This
Zhang Zhengyou's system calibration method
Zhang Zhengyou's system calibration method is widely used in computer vision. It can build a camera imagine geometry model according to the camera parameters. Then the 3D scene will reconstruct from the obtained image and calibrated.
In three-dimensional world coordinates, a certain point in the image is X = [X, Y, Z, 1]T, and in a two-dimensional camera plane pixel, a certain point is m = [u, v, 1]T, therefore, The relation existing between the corresponding image planes of feature planes is as
Conclusion
In this paper, introduce the dual attention mechanism into the image visual processing. Then use the multi-level feature extraction method to extract the feature regions of the three-dimensional image. On the basis of the neural network, the algorithm model focuses on some valuable values. The image area can effectively describe the picture theme, and bring the area with large weight ratio into the calibration algorithm of Zhang Zhengyou’ system calibration method. Finally, the space
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The study was supported by “Science and Technology Project of Sichuan Provincial Department of Education, China - Design and Implementation of MOOC-oriented Large Data Real-time Stream Recommendation System Based on Storm (No. 18ZB0099).
References (17)
- et al.
A calibration method for stereovision system based on solid circle target
Measurement
(2019) - et al.
Single-image calibration method for multiple virtual binocular vision system
Measurement
(2019) - et al.
Accurate calibration approach for non-overlapping multi-camera system
Opt. Laser Technol.
(2019) - et al.
Sequence searching with CNN features for robust and fast visual place recognition
Comput. Graph.
(2018) - et al.
Research on logistics supply chain of Iron and steel enterprises based on block chain technology
Futur. Gener. Comput. Syst.
(2019) - et al.
Simulation study on characteristics of information extraction in multiple-image radiography
Nucl. Sci. Tech.
(2018) - et al.
Camera calibration algorithm based on Harris-Zhang Zhengyou plane calibration method
J. Xi’an Univ. Architect. Technol. (NATURAL SCIENCE EDITION)
(2014) - et al.
Integrating an attention mechanism and convolution collaborative filtering for document context-aware rating prediction
IEEE Access
(2019)
Cited by (13)
Accelerating Attention Mechanism on FPGAs based on Efficient Reconfigurable Systolic Array
2023, ACM Transactions on Embedded Computing SystemsMonitoring of biodiversity of water communities
2023, E3S Web of Conferences