A cleaning method suitable for a camera of an automatic driving system of a vehicle

Abstract

The application discloses a cleaning method suitable for a camera of an automatic driving system of an automobile, and comprises the following steps: step one, acquiring images shot by the camera of the automatic driving system in real time; step two, identifying the images based on a rainbow refraction method to obtain water quantity corresponding to a pollution area; step three, determining a corresponding cleaning intensity level according to the water quantity; and step four, controlling a spraying speed and a spraying time of a cleaning device to clean the camera of the automatic driving system according to the cleaning intensity level. The cleaning method suitable for the camera of the automatic driving system of the automobile can accurately identify pollution conditions of the camera of the automatic driving system of the automobile, and can control a cleaning mode of a cleaning system according to the corresponding pollution conditions, so that the cleaning efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of automotive autonomous driving technology, and specifically relates to a cleaning method for cameras in automotive autonomous driving systems. Background Technology

[0002] With the rapid development of the automotive industry, cars are becoming increasingly intelligent and green, meeting the demands of the times and catering to the needs of the public. Various countries are also continuously issuing corresponding policies to encourage and support the development of the automotive industry. Companies are seizing this opportunity to vigorously develop technologies in areas such as intelligent connectivity, comfort, and safety.

[0003] Technologies such as intelligent navigation, driver assistance, and intelligent voice assistants are becoming increasingly widespread, and with system optimization, upgrades, and iterations, the user experience is constantly improving.

[0004] However, while intelligent technology brings convenience and comfort, the inherent safety risks also warrant attention. During autonomous driving, in severe weather conditions such as heavy rain, blizzards, and fog, the surfaces of the vehicle's external sensors can become contaminated with water. If this contamination is not addressed promptly, the intelligent driving system may be unable to accurately assess road conditions, leading to various dangerous driving safety issues. Therefore, making intelligent recognition systems more sophisticated, accurate, and reliable, and improving the driving safety and comfort of new energy vehicles, has become a top priority.

[0005] Current camera cleaning devices mostly rely on photoresistors, ultrasonic waves, or other methods combined with various algorithms for cleaning, or use complex cleaning devices. Intelligent recognition systems may experience reduced accuracy and stability under unfavorable lighting conditions, such as strong light or shadows. Changes in lighting conditions affect the sensors and cameras, limiting the performance of intelligent recognition systems. Furthermore, these methods also suffer from high costs and low cleaning efficiency. Summary of the Invention

[0006] The purpose of this invention is to provide a cleaning method for cameras in automotive autonomous driving systems, which can accurately identify the contamination status of automotive autonomous driving cameras and control the cleaning method of the cleaning system according to the corresponding contamination status, thereby improving cleaning efficiency.

[0007] The technical solution provided by this invention is as follows:

[0008] A cleaning method for cameras in automotive autonomous driving systems includes the following steps:

[0009] Step 1: Acquire images captured by the cameras of the autonomous driving system in real time;

[0010] Step 2: Identify the image to obtain the water volume corresponding to the polluted area;

[0011] Step 3: Determine the corresponding cleaning intensity level based on the water volume;

[0012] Step 4: Clean the autonomous driving camera by controlling the spray speed and spray time of the cleaning device according to the cleaning intensity level.

[0013] Preferably, in step two, the method for obtaining the water volume corresponding to the polluted area is as follows:

[0014] The full-field rainbow signal of the contaminated area in the image was recorded using the rainbow refraction method. The Active-set method was used for inversion iteration to obtain the average refractive index and particle size distribution of the droplets in the contaminated area. The average refractive index and particle size distribution were matched with known mapping relationships to obtain the thickness of the liquid in the contaminated area. The amount of water corresponding to the contaminated area was calculated based on the thickness and the area of ​​the contaminated area.

[0015] The mapping relationship is the mapping relationship between the average refractive index and particle size distribution of the droplets and the thickness of the liquid.

[0016] Preferably, in step three, the cleaning intensity level is divided into six levels, and the water volume is set as Q;

[0017] If the water volume Q in the polluted area is greater than or equal to 0.001 ml and less than 0.05 ml, it corresponds to the first cleaning intensity level;

[0018] If the water volume Q in the polluted area is greater than or equal to 0.05 ml and less than 0.1 ml, it corresponds to the second cleaning intensity level;

[0019] If the water volume Q in the polluted area is greater than or equal to 0.1 ml and less than 0.5 ml, it corresponds to the third cleaning intensity level;

[0020] If the water volume Q in the polluted area is greater than or equal to 0.5 ml and less than 1.5 ml, it corresponds to the fourth cleaning intensity level;

[0021] If the water volume Q in the polluted area is greater than or equal to 1.5 ml and less than 3 ml, it corresponds to the fifth cleaning intensity level;

[0022] If the water volume Q in the polluted area is 3 ml or more, it corresponds to the sixth level of cleaning intensity.

[0023] Preferably, step four further includes setting the spray pressure of the cleaning device to 100 Pa.

[0024] Preferably, when the cleaning intensity level is the first cleaning intensity level, the spray speed is set to 1 m / s and the spray time is 0.3 min;

[0025] When the cleaning intensity level is the second cleaning intensity level, the spray speed is set to 1m / s and the spray time is 0.5min;

[0026] When the cleaning intensity level is the third cleaning intensity level, the spray speed is set to 2m / s and the spray time is 0.3min;

[0027] When the cleaning intensity level is the fourth cleaning intensity level, the spray speed is set to 2m / s and the spray time is 0.5min;

[0028] When the cleaning intensity level is the fifth level, the spray speed is set to 3m / s and the spray time is 0.3min.

[0029] When the cleaning intensity level is the sixth level, the spray speed is set to 3m / s and the spray time is 0.5min.

[0030] Preferably, in step four, the spray distance is set to 2 cm.

[0031] Preferably, in step four, the spray angle is set to 50-70 degrees.

[0032] The beneficial effects of this invention are:

[0033] The cleaning method for cameras in automotive autonomous driving systems provided by this invention can accurately identify the contamination status of automotive autonomous driving cameras and control the cleaning method of the cleaning system according to the corresponding contamination status, thereby improving cleaning efficiency. Attached Figure Description

[0034] Figure 1 This is a flowchart of a cleaning method for cameras used in automotive autonomous driving systems, as described in this invention.

[0035] Figure 2 This is a flowchart of the rainbow refraction method described in this invention. Detailed Implementation

[0036] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0037] like Figure 1-2 As shown, this invention provides a cleaning method for cameras in automotive autonomous driving systems, which acquires images captured by the cameras in real time; performs photoelectric conversion on the patterns, converts the relevant information into a first signal, and transmits the signal to a recognition control system. The recognition control system includes an image recognition module and a cleaning device control module. The image recognition module incorporates a rainbow refraction method, and the cleaning device control module controls the spray parameters of the cleaning device. The specific implementation process is as follows.

[0038] S1: Acquire images captured by the camera of the autonomous driving system in real time and use these images as input for the rainbow refraction algorithm.

[0039] S2: Based on the rainbow refraction method, the image is identified to obtain the water volume corresponding to the polluted area.

[0040] Rainbow refraction technology obtains rainbow signals from images, and by using Airy integral theory and geometric optics theory to calculate the scattering angles at low-frequency Airy peaks and valleys, as well as the high-frequency ripple structure superimposed on them, based on the related geometric rainbow angular positions and light intensity. This allows us to obtain the angular distribution and frequency of the rainbow signal, and further obtain the refractive index and droplet size of the droplets.

[0041] Specifically, without considering evaporation, the system utilizes the robustness of the signal inversion algorithm to pre-set the droplet size distribution. Therefore, the rainbow signal of non-uniformly sized droplets in the contamination profile region obtained by the Rainbow Refraction Detection (GRR) technique is based on the aforementioned droplet size distribution, and parameters such as the droplet's average refractive index and particle size distribution are then inverted.

[0042] The GRR technique selected in this invention is applicable to the measurement of droplets with different sizes and even shapes. Compared with standard rainbow images, the high-frequency ripple structure superimposed on the main peak of the rainbow disappears. At the same time, due to the random orientation of non-spherical droplets, the superposition of their rainbow light intensities forms a uniform background, which does not affect the inversion of the rainbow signal.

[0043] Compared to currently known high-speed microscopy imaging techniques, the challenge in applying high-speed microscopy lies in its inability to simultaneously achieve a high measurement field of view and high resolution. For moving micrometer-sized droplets, it can only capture information within a few milliseconds of the measurement field of view. Most laser measurement techniques, such as particle holographic imaging, phase Doppler analyzers, and laser particle interferometry, currently also limit their particle size measurement accuracy to the micrometer level. Rainbow refraction detection technology, however, offers significantly higher accuracy.

[0044] Before inversion, different signal preprocessing is performed on the full-field rainbow signal, such as removing high-frequency structures and estimating the range of inversion parameters, to accurately accelerate the inversion convergence process.

[0045] The image processing mainly involves the inversion of rainbow signals and particle interference images, while the parameter inversion of droplet rainbow signals also requires precise angle calibration.

[0046] This invention employs a liquid column calibration method. It involves recording the rainbow signal of a liquid column with a known diameter and refractive index, then using light scattering theory to calculate the theoretical scattered light distribution of the liquid at that diameter and refractive index. Subsequently, by comparing the Airy peak distributions of the experimental and theoretical signals, the scattering angles corresponding to different pixels are obtained, and a calibration curve is obtained through linear fitting.

[0047] Compared to other scattering angle calibration methods, such as dual-wavelength calibration and mirror calibration, this method is easier to generate and control. Furthermore, its larger size produces richer details in the generated rainbow signal, thus improving the accuracy of angle calibration for scattering angle determination.

[0048] By using the scattering angle calibration method, the relationship between the pixels and the scattering angle of the recorded rainbow image can be obtained, and then phase rainbow image processing can be performed.

[0049] Specifically, firstly, rainbow signals from different pixel rows are extracted from the acquired image. The extracted rainbow intensity distribution is then processed using Fast Fourier Transform (FFT) and filtered to obtain the separated low-frequency Airy peaks of the rainbow. Subsequently, the separated Airy rainbow is subtracted from the total rainbow intensity to obtain the high-frequency ripple structure superimposed on it.

[0050] Specifically, the first goal of rainbow signal processing is to calculate the simulated rainbow signal based on the modified CAM theory, and at the same time, to use iterative search to find the optimal fit for the measured Airy rainbow intensity distribution (i.e., CAM theory analyzes the angular range of consistent approximation, limits the angular domain range close to and far from the rainbow signal, and divides the incident light into blocks according to the distribution of the rainbow angle for separate analysis and fitting), thereby inverting the droplet size and refractive index.

[0051] The following is a brief introduction to the formulas near the geometric rainbow angle involved in the inversion.

[0052]

[0053] Where α = πD / λ is a dimensionless size parameter, D is the droplet diameter, and λ is the wavelength of the incident light; The scattering angle is i; i is the imaginary unit. and These are the Airy function and its derivative, respectively, with ε being the scattering angle. With geometric rainbow angle The difference ( = +ε); parameters A, B, and C are respectively represented as:

[0054]

[0055]

[0056]

[0057] Where s and c are the incident angles of the incident light on the droplet surface, respectively. sine and cosine values

[0058]

[0059]

[0060]

[0061] Based on the above calculations, near the geometric rainbow, the intensity angle distribution of the first-order scattering (p=2) predicted by CAM theory and Debye series theory is in perfect agreement at all positions of the scattering angle.

[0062] The preprocessed rainbow signal is then read in and normalized to its maximum value. An inversion segment is then selected, requiring it to contain at least one complete main peak. The estimated refractive index and grain size are set as initial input values. Based on these initial values, the light intensity matrix is ​​calculated using CAM theory, and the current objective function value is calculated using the non-negative least squares method.

[0063] The algorithm checks if the objective function value is less than threshold 1. If so, it outputs the result; otherwise, it searches for and determines the next iteration point using an iterative method. Simultaneously, it checks if the change from the previous iteration point is less than threshold 2. If so, it stops iterating and outputs the result; otherwise, it repeats the above calculations. In this algorithm, determining whether to end the iteration consists of two parts: the threshold Δ for the objective function value and the threshold ε for the change in the parameter of the variable to be solved.

[0064] The threshold for the change in the objective function is set to Δ=10. -6 This is sufficient to ensure the accuracy of the inversion fitting. The threshold for the change of the parameter of the variable to be determined is set to ε = 2 × 10⁻⁶. -4 Because the accuracy of rainbow refraction technology in measuring refractive index is 2 × 10⁻⁶. -4 However, the measurement accuracy of particle size must be less than this threshold ε; setting this threshold avoids invalid iterative optimization calculations while ensuring measurement accuracy.

[0065] Specifically, this paper adopts the Active-set method, which is currently the best choice for iterative methods of rainbow signal inversion. Its inversion accuracy and timeliness are optimized compared to other iterative algorithms.

[0066] Furthermore, by using the algorithm of the identification control system, namely the Active-set method, parameters such as the average refractive index and particle size distribution of the droplets are calculated through inversion, and compared with preset values ​​to determine the amount of water currently adhering to the mirror surface.

[0067] Specifically, the control system uses full-field rainbow refraction technology to retrieve the average refractive index and particle size distribution. Based on the mapping relationship between the average refractive index and particle size distribution and the contaminated area, the thickness of the contaminated area is obtained. The identification control system identifies the area of ​​the contaminated area based on the image captured by the camera. Finally, based on the thickness and area of ​​the contaminated area, the water volume (or area) of the final contaminated outline region is calculated. In this embodiment, the area of ​​the contaminated area is obtained by processing the image using OpenCV-Python.

[0068] Furthermore, the identification and control system matches the current water volume with the cleaning intensity level in the preset scheme to obtain the specific parameters for controlling the cleaning device.

[0069] In this embodiment, the spray pressure, spray distance, and spray angle of the cleaning device are set to fixed values ​​that do not change with the level of contamination. Specifically, the spray pressure is set to 100 Pa, the spray distance is set to 2 cm, and the spray angle is set to 50-70 degrees.

[0070] As a preferred embodiment, the cleaning intensity level is divided into six levels, and the water volume is set as Q;

[0071] If the water volume Q in the polluted area is greater than or equal to 0.001 ml and less than 0.05 ml, it corresponds to the first cleaning intensity level;

[0072] If the water volume Q in the polluted area is greater than or equal to 0.05 ml and less than 0.1 ml, it corresponds to the second cleaning intensity level;

[0073] If the water volume Q in the polluted area is greater than or equal to 0.1 ml and less than 0.5 ml, it corresponds to the third cleaning intensity level;

[0074] If the water volume Q in the polluted area is greater than or equal to 0.5 ml and less than 1.5 ml, it corresponds to the fourth cleaning intensity level;

[0075] If the water volume Q in the polluted area is greater than or equal to 1.5 ml and less than 3 ml, it corresponds to the fifth cleaning intensity level;

[0076] If the water volume Q in the polluted area is 3 ml or more, it corresponds to the sixth level of cleaning intensity.

[0077] As a further preferred option, when the cleaning intensity level is the first cleaning intensity level, the spray speed is set to 1 m / s and the spray time is 0.3 min;

[0078] When the cleaning intensity level is the second cleaning intensity level, the spray speed is set to 1m / s and the spray time is 0.5min;

[0079] When the cleaning intensity level is the third cleaning intensity level, the spray speed is set to 2m / s and the spray time is 0.3min;

[0080] When the cleaning intensity level is the fourth cleaning intensity level, the spray speed is set to 2m / s and the spray time is 0.5min;

[0081] When the cleaning intensity level is the fifth level, the spray speed is set to 3m / s and the spray time is 0.3min.

[0082] When the cleaning intensity level is the sixth level, the spray speed is set to 3m / s and the spray time is 0.5min.

[0083] After cleaning is completed, images are captured by the autonomous driving camera and transmitted back to the recognition and control system for identification. If the water volume in the contaminated area is less than 0.001 ml, the cleaning process ends. Otherwise, the cleaning process continues according to the corresponding cleaning intensity level based on the identification result.

[0084] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A cleaning method for cameras in automotive autonomous driving systems, characterized in that, Includes the following steps: Step 1: Acquire images captured by the cameras of the autonomous driving system in real time; Step 2: Identify the image to obtain the water volume corresponding to the polluted area; Step 3: Determine the corresponding cleaning intensity level based on the water volume; Step 4: Clean the autonomous driving camera by controlling the spray speed and spray time of the cleaning device according to the cleaning intensity level; In step two, the method for obtaining the water volume corresponding to the polluted area is as follows: The full-field rainbow signal of the contaminated area in the image was recorded using the rainbow refraction method. The Active-set method was used for inversion iteration to obtain the average refractive index and particle size distribution of the droplets in the contaminated area. The average refractive index and particle size distribution were matched with known mapping relationships to obtain the thickness of the liquid in the contaminated area. The amount of water corresponding to the contaminated area was calculated based on the thickness and the area of ​​the contaminated area. Wherein, the mapping relationship is the mapping relationship between the average refractive index and particle size distribution of the droplets and the thickness of the liquid; In step three, the cleaning intensity level is divided into six levels, and the water volume is set as Q; If the water volume Q in the polluted area is greater than or equal to 0.001 ml and less than 0.05 ml, it corresponds to the first cleaning intensity level; If the water volume Q in the polluted area is greater than or equal to 0.05 ml and less than 0.1 ml, it corresponds to the second cleaning intensity level; If the water volume Q in the polluted area is greater than or equal to 0.1 ml and less than 0.5 ml, it corresponds to the third cleaning intensity level; If the water volume Q in the polluted area is greater than or equal to 0.5 ml and less than 1.5 ml, it corresponds to the fourth cleaning intensity level; If the water volume Q in the polluted area is greater than or equal to 1.5 ml and less than 3 ml, it corresponds to the fifth cleaning intensity level; If the water volume Q in the polluted area is 3 ml or more, it corresponds to the sixth level of cleaning intensity.

2. The cleaning method for cameras in automotive autonomous driving systems according to claim 1, characterized in that, Step four also includes setting the spray pressure of the cleaning device to 100 Pa.

3. The cleaning method for cameras in automotive autonomous driving systems according to claim 2, characterized in that, When the cleaning intensity level is the first cleaning intensity level, the spray speed is set to 1m / s and the spray time is 0.3min; When the cleaning intensity level is the second cleaning intensity level, the spray speed is set to 1m / s and the spray time is 0.5min; When the cleaning intensity level is the third cleaning intensity level, the spray speed is set to 2m / s and the spray time is 0.3min; When the cleaning intensity level is the fourth cleaning intensity level, the spray speed is set to 2m / s and the spray time is 0.5min; When the cleaning intensity level is the fifth level, the spray speed is set to 3m / s and the spray time is 0.3min. When the cleaning intensity level is the sixth level, the spray speed is set to 3m / s and the spray time is 0.5min.

4. The cleaning method for cameras in automotive autonomous driving systems according to claim 2 or 3, characterized in that, In step four, the spray distance is set to 2cm.

5. The cleaning method for cameras in automotive autonomous driving systems according to claim 4, characterized in that, In step four, the spray angle is set to 50-70 degrees.

Images