CN-122023171-A - Method, system, equipment and medium for generating simulation training data based on camera noise

Abstract

The application discloses a method, a system, equipment and a medium for generating simulation training data based on camera noise. The method for generating the simulation training data based on the camera noise comprises the steps of predicting a camera image without noise through a prediction model to obtain prediction result data, determining a value range of camera noise parameters according to the confidence of the prediction model, and applying noise to the camera image without noise according to the value range of the camera noise parameters to generate the simulation training data. By adopting the application, the value range of the noise parameters of the camera is determined through the confidence coefficient of the prediction model, and the noise parameters are selected in the range to noise the camera image, so that the simulation training data which meets the model requirement and is close to the real noise is generated.

Inventors

• MIAO CHANGLONG
• MA ZHIYAO

Assignees

• 浙江越影科技有限公司

Dates

Publication Date

20260512

Application Date

20260116

Claims (10)

1. 1. A method for generating simulated training data based on camera noise, the method comprising: Predicting a camera image without noise through a prediction model to obtain prediction result data, wherein the prediction result data comprises the confidence coefficient of the prediction model; Determining a value range of a camera noise parameter according to the confidence coefficient of the prediction model; and applying noise to the camera image without noise according to the value range of the camera noise parameter to generate simulation training data.
2. 2. The method for generating simulated training data based on camera noise according to claim 1, further comprising, prior to said determining a range of values for camera noise parameters based on a confidence level of said predictive model: Determining one or more noise sources from the camera image noise, wherein each noise source is controlled by a camera noise parameter; and establishing a mapping model of the model confidence coefficient and the value range of the camera noise parameter.
3. 3. The method for generating simulation training data based on camera noise according to claim 2, wherein the modeling a mapping model of the confidence coefficient of the model and the range of values of the camera noise parameter comprises: And if the model confidence coefficient is larger than a first confidence coefficient threshold, adopting a first noise parameter value range, if the model confidence coefficient is positioned between the first confidence coefficient threshold and a second confidence coefficient threshold, adopting a second noise parameter value range, and if the model confidence coefficient is smaller than the second confidence coefficient threshold, adopting a third noise parameter value range, wherein the first confidence coefficient threshold is larger than the second confidence coefficient threshold, the first noise parameter value range is larger than the second noise parameter value range, and the second noise parameter value range is larger than the third noise parameter value range.
4. 4. The method for generating simulated training data based on camera noise according to claim 2, wherein determining the range of values of the camera noise parameters according to the confidence level of the prediction model comprises: and determining the value range of the camera noise parameter according to the confidence coefficient of the prediction model based on the mapping model of the model confidence coefficient and the value range of the camera noise parameter.
5. 5. The method for generating simulated training data based on camera noise as claimed in claim 1, wherein said applying noise to said noise-free camera image based on said range of values of said camera noise parameter, generating simulated training data comprises: randomly selecting a plurality of groups of camera noise parameter values according to the value range of the camera noise parameters; generating a noise image according to the camera noise parameter value; And synthesizing the noise image and the camera image without noise to obtain a camera image with noise as simulation training data.
6. 6. The method for generating camera noise-based simulated training data of claim 5, further comprising: Another set of simulated training data is generated from the noisy camera image.
7. 7. The method for generating camera noise-based simulated training data of claim 1, further comprising: Analyzing links affecting the prediction stability of the prediction model; And generating targeted camera image data according to the links influencing the prediction stability of the prediction model.
8. 8. A system for generating simulated training data based on camera noise, comprising: the prediction module is used for predicting the camera image without noise through a prediction model to obtain prediction result data, wherein the prediction result data comprises the confidence level of the prediction model; The parameter determining module is used for determining the value range of the noise parameter of the camera according to the confidence coefficient of the prediction model; and the noise adding module is used for applying noise to the camera image without noise according to the value range of the camera noise parameter to generate simulation training data.
9. 9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of generating camera noise based simulation training data according to any of claims 1-7 when the program is executed by the processor.
10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method of generating camera noise based simulation training data according to any of claims 1-7.

Description

Method, system, equipment and medium for generating simulation training data based on camera noise Technical Field The present application relates to the field of image processing technologies, and in particular, to a method, a system, an apparatus, and a medium for generating simulation training data based on camera noise. Background In the application fields of automatic driving, intelligent body building and the like, a large amount of simulation image data is generally required to train a model so as to improve the performance of the model. However, in the imaging process of the camera, there are various noise effects, and statistical models such as gaussian noise, pretzel noise and the like adopted in some image simulation methods often cannot reflect the physical imaging process of a real camera, so that a trained model does not perform well on a real data set, in addition, random or uniform noise injection strategies are adopted in some technologies, and due to the lack of a dynamic adaptation mechanism with the model state, edge cases are difficult to actively generate, the data generation efficiency is low, and meanwhile, the influence of real factors such as temperature, sensitivity, exposure time and the like on the noise is not generally considered in the existing methods, so that simulation training data close to the real physical imaging is difficult to generate. Disclosure of Invention Based on the above, it is necessary to provide a method, a system, a device and a medium for generating simulation training data based on camera noise, which determine a value range of camera noise parameters by predicting confidence of a model, and select noise parameters in the range to noise a camera image, so as to generate simulation training data which meets the model requirement and is close to real noise. In a first aspect, a method for generating simulated training data based on camera noise is provided, including: Predicting a camera image without noise through a prediction model to obtain prediction result data, wherein the prediction result data comprises the confidence coefficient of the prediction model; Determining a value range of a camera noise parameter according to the confidence coefficient of the prediction model; and applying noise to the camera image without noise according to the value range of the camera noise parameter to generate simulation training data. Further, before determining the range of values of the camera noise parameters according to the confidence coefficient of the prediction model, the method further comprises: Determining one or more noise sources from the camera image noise, wherein each noise source is controlled by a camera noise parameter; and establishing a mapping model of the model confidence coefficient and the value range of the camera noise parameter. Further, the establishing a mapping model of the model confidence coefficient and the value range of the camera noise parameter includes: And if the model confidence coefficient is larger than a first confidence coefficient threshold, adopting a first noise parameter value range, if the model confidence coefficient is positioned between the first confidence coefficient threshold and a second confidence coefficient threshold, adopting a second noise parameter value range, and if the model confidence coefficient is smaller than the second confidence coefficient threshold, adopting a third noise parameter value range, wherein the first confidence coefficient threshold is larger than the second confidence coefficient threshold, the first noise parameter value range is larger than the second noise parameter value range, and the second noise parameter value range is larger than the third noise parameter value range. Further, the determining the range of the camera noise parameter according to the confidence coefficient of the prediction model includes: and determining the value range of the camera noise parameter according to the confidence coefficient of the prediction model based on the mapping model of the model confidence coefficient and the value range of the camera noise parameter. Further, the applying noise to the camera image without noise according to the value range of the camera noise parameter, and generating simulation training data, includes: randomly selecting a plurality of groups of camera noise parameter values according to the value range of the camera noise parameters; generating a noise image according to the camera noise parameter value; And synthesizing the noise image and the camera image without noise to obtain a camera image with noise as simulation training data. Further, the method further comprises the following steps: Another set of simulated training data is generated from the noisy camera image. Further, the method further comprises the following steps: Analyzing links affecting the prediction stability of the prediction model; And generating targeted camera image data according to the