AI video deep learning-based agglomerate fog intelligent identification technology

Abstract

The invention discloses an intelligent agglomerate fog recognition technology based on AI video deep learning, and relates to the technical field of intelligent agglomerate fog recognition. According to the method, the video stream image in the configuration file is acquired, different images at multiple positions in the video stream are acquired, then parameter correction is carried out on the images, and the information of the multiple images can be combined by fitting the information of the multiple images, so that the information of the multiple images can be combined together; therefore, image information sending errors can be avoided, the problem of an agglomerate fog visibility recognition algorithm can be solved, the recognition accuracy and efficiency are improved, and the recognition accuracy and efficiency are improved by obtaining the video stream image in the configuration file, obtaining multiple different images in the video stream and then carrying out gray processing and Gaussian blur on the obtained images. The input image is preprocessed, the image brightness, contrast, color balance and other operations are adjusted, and the recognition accuracy of the model is improved while the image information is reserved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent fog recognition technology, and specifically to an intelligent fog recognition technology based on AI video deep learning. Background Art

[0002] Fog is a common weather phenomenon with strong regionality, low visibility, fast changes and great difficulty in prediction. It is extremely harmful to highway driving safety. Fog is different from large-scale advection fog and radiation fog. Its atmospheric visibility value varies greatly, and the visibility monitored in the fog area rises and falls repeatedly. Due to the unique climatic conditions in the cross-river basin, the air humidity is relatively high. When it is in the season with a large temperature difference between day and night, water evaporates into the air during the day. When the temperature drops at night, the water vapor in the air is very easy to liquefy to form fog. Fog formed in different ways has different impacts on traffic operations. The difference makes it very difficult to identify and predict fog. The formation conditions of fog include three major factors: clear night, light breeze and sufficient water vapor. Fog is different from the heavy fog we usually see. It is block-shaped and has a relatively small control range, just like a pile of cotton wool falling on the ground. The visibility outside the fog is good, but the visibility inside the fog is hazy and extremely low. Fog also has the characteristics of small range, long accumulation time, easy drift and strong suddenness. Highway is an industry related to the development of the lifeline of the national economy. According to statistics, heavy fog is the main cause of frequent serious traffic accidents on highways.

[0003] However, the current visibility algorithms of fog cluster intelligent recognition technology mostly rely on human intervention, have low recognition degree and low efficiency, cannot be recognized based on camera video, can only be processed by manual input image, and are not convenient for uploading the recognized data and cannot be displayed intuitively. In response to the above problems, the inventors propose a fog cluster intelligent recognition technology based on AI video deep learning to solve the above problems. Summary of the invention

[0004] In order to solve the problem that the current intelligent fog recognition technology has low recognition degree and low efficiency, the purpose of the present invention is to provide an intelligent fog recognition technology based on AI video deep learning.

[0005] To solve the above technical problems, the present invention adopts the following technical solutions: an intelligent fog recognition technology based on AI video deep learning, firstly, reading the file materials in the storage card configuration through a reading device, and then obtaining an image from the video stream of the read configuration file, then deleting the image data one month ago from the obtained image, and graying the deleted image and converting it into a grayscale image, and then performing Gaussian blur processing on the grayscale image to blur the image, and then performing adaptive threshold processing on the Gaussian blurred image, processing the light and dark difference of the image through adaptive threshold processing, and then calculating the visibility of the processed image, and then calculating the result of the image visibility, and performing parameter correction on the result, and then storing the processed image data through a storage device, and finally sending the stored image data to a cloud server for storage through the cloud server.

[0006] Preferably, the method for acquiring images is to directly pull out the video in the configuration file to perform image acquisition, and multiple image information in one video is acquired. The method for deleting image data older than one month is to determine the retention and deletion of the image by determining the storage date of the image, and regularly delete the image data older than one month, and delete duplicate information, eliminate image errors caused by camera spot problems, and perform grayscale processing on the image data after deletion to make the image a grayscale image. Gaussian blur is a processing effect widely used in image processing. By Gaussian blurring the image, image noise can be reduced and the level of detail can be reduced to ensure that the acquired image is clearer.

[0007] Preferably, the adaptive threshold processing technology uses a variable threshold to complete the threshold processing of the image. When performing the threshold processing, the adaptive threshold processing obtains the threshold by calculating the weighted average of the adjacent areas around each pixel, and uses the threshold to process the current pixel. The threshold t0 is set. When the t value is less than t0, t=t0 is set. Taking t0=0.1 as the standard, the processed three-channel images are synthesized to restore the clear image J(x) after defogging. The image visibility observation model is obtained by curve fitting using the brightness mean of the digital image near the object in the transmittance distribution map and the visibility value of the forward scattering instrument at the corresponding time. The relationship between the brightness mean of the digital image transmittance distribution map and the atmospheric horizontal visibility is obtained. The image visibility observation model is expressed by the following polynomial: V(t)=a1T(t)4+a2T(t)3+a3T(t)2+a4T(t)+a5, where t is the observation time, T(t) is the brightness mean of the image transmittance distribution map, V(t) is the atmospheric visibility value, a1-a5 are the polynomial coefficients, and the atmospheric visibility is the distance when the parallel light attenuates to 0.05 of the original luminous flux through the atmosphere. The visibility value V can be derived, and the visibility value V is obtained by the atmospheric extinction coefficient β obtained in the process of obtaining a clear image.

[0008] Preferably, different error values ​​of fog group image information are recorded in different images. By fitting these error values, correction parameters are calculated, and parameter correction is performed on the calculation results. The obtained correction parameters are applied to the measurement data, and higher-precision measurement results are obtained by subtraction. The images after parameter correction are stored in a database and transmitted to the cloud server through the database.

[0009] Compared with the prior art, the present invention has the following beneficial effects:

[0010] 1. By acquiring the video stream images in the configuration file, multiple different images in the video stream are acquired, and then the image parameters are corrected. By fitting the information of multiple images, the information of multiple images can be combined, and then the information of multiple images can be combined together to calculate the comprehensive parameters, thereby avoiding the error in image information transmission, and then solving the problem of fog visibility recognition algorithm, and improving recognition accuracy and efficiency;

[0011] 2. By acquiring the video stream images in the configuration file, multiple different images in the video stream are acquired, and then grayscale processing and Gaussian blur are performed on the acquired images, and the input images are pre-processed to adjust the image brightness, contrast, color balance and other operations, thereby improving the recognition accuracy of the model while retaining the image information, thereby improving the recognition accuracy and processing efficiency, making it more intelligent and efficient;

[0012] 3. Combining intelligent control technology and sensor technology, the function of automatically collecting camera video streams can be realized. By using mature Internet of Things technology, the data can be uploaded to the cloud for processing and analysis. Combining intelligent control technology and sensor technology, the function of automatically collecting camera video streams can be realized. By using mature Internet of Things technology, the data can be uploaded to the cloud for processing and analysis, which facilitates the upload of data and the intuitive display of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 It is a schematic diagram of the fog group intelligent identification technology system of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Example: Figure 1 As shown, the present invention provides an intelligent identification technology for group fog based on AI video deep learning. First, the file material in the storage card configuration is read by a reading device, and then an image is obtained from the video stream of the read configuration file, and then the image data one month ago is deleted from the obtained image, and the deleted image is grayed and converted into a grayscale image, and then Gaussian blur processing is performed on the grayscale image to blur the image, and then the image after Gaussian blur processing is performed on the image, and the brightness difference of the image is processed by adaptive threshold processing, and then the visibility of the processed image is calculated, and then the result of the image visibility is calculated, and the result is parameter corrected, and then the processed image data is stored through a storage device, and finally the stored image data is sent to a cloud server for storage through the cloud server.

[0017] By adopting the above technical solution, the file material in the storage card can be read by reading the configuration file, so as to facilitate the acquisition of the image in the file material, so that multiple image information can be obtained in the video stream to fit the multiple image information and avoid errors in the image information. Deleting image data from one month ago can avoid continuous occupation of hard disk space, and thus free up some space on the hard disk, so that the hard disk can be used for a longer time. Image grayscale processing can be used as a preprocessing step for image processing to prepare for subsequent upper-level operations such as image segmentation, image recognition and image analysis. By processing the image with an adaptive threshold, the brightness difference of the image can be better processed. Calculating the image visibility can calculate the brightness mean of the image transmittance distribution map, and parameter correction of the calculation results can avoid errors in image information.

[0018] The method for acquiring images is to directly pull the video in the configuration file to collect the image, collect multiple image information in one video, and delete the image data older than one month by determining the image storage date to determine the retention and deletion of the image. The image data older than one month is regularly deleted, and the duplicate information is deleted to eliminate the image errors caused by the camera spot problem. At the same time, the image data after deletion is gray-processed to make the image a gray-scale image. Gaussian blur is a widely used processing effect in image processing. By Gaussian blurring the image, the image noise can be reduced and the detail level can be reduced to ensure that the acquired image is clearer.

[0019] By adopting the above technical scheme, a Gaussian mixture model GMM classifier is trained offline through an image training library of a foggy or non-fog image classifier. The Gaussian mixture model GMM classifier extracts one-dimensional grayscale histogram image features of foggy images and non-fog images according to the grayscale value frequency characteristics of foggy images, and classifies foggy images and non-fog images according to the histogram of the images. The processed road images are classified as foggy or non-fog using the Gaussian mixture model GMM classifier for foggy or non-fog images and the support vector machine SVM classifier for foggy or non-fog images, respectively. The foggy images classified by the Gaussian mixture model GMM classifier for foggy or non-fog images and the foggy images classified by the support vector machine SVM classifier for foggy or non-fog images are taken as a union to obtain a real-time foggy road image.

[0020] The adaptive threshold processing technology uses a variable threshold to complete the threshold processing of the image. When performing threshold processing, the adaptive threshold processing obtains the threshold by calculating the weighted average of the adjacent areas around each pixel, and uses the threshold to process the current pixel. The threshold t0 is set. When the t value is less than t0, t=t0 is set. Taking t0=0.1 as the standard, the processed three-channel images are synthesized to restore the clear image J(x) after defogging. The image visibility observation model is obtained by curve fitting the brightness mean of the digital image near the object in the transmittance distribution map and the visibility value of the forward scattering instrument at the corresponding time. The relationship between the brightness mean of the digital image transmittance distribution diagram and the atmospheric horizontal visibility is expressed by the following polynomial: V(t)=a1T(t)4+a2T(t)3+a3T(t)2+a4T(t)+a5, where t is the observation time, T(t) is the brightness mean of the image transmittance distribution diagram, V(t) is the atmospheric visibility value, a1-a5 are the polynomial coefficients, and the atmospheric visibility is the distance when the parallel light attenuates to 0.05 of the original luminous flux through the atmosphere. The visibility value V can be derived, and the visibility value V can be obtained by the atmospheric extinction coefficient β obtained in the process of obtaining a clear image.

[0021] By adopting the above technical solution, the adaptive threshold processing technology uses a variable threshold to complete the threshold processing of the image. When performing threshold processing, the adaptive threshold processing obtains the threshold by calculating the weighted average of the adjacent area around each pixel, and uses the threshold to process the current pixel. Since noise is inevitably present in the actual imaging process, the differential image also contains noise, but the noise value is usually relatively small. Therefore, a proper threshold can be used to better detect the moving object in the differential image.

[0022] Different fog image information error values ​​are recorded in different images. By fitting these error values, the correction parameters are calculated, and the calculation results are corrected. The obtained correction parameters are applied to the measurement data, and higher-precision measurement results are obtained by subtraction. The images after parameter correction are stored in the database and transmitted to the cloud server through the database.

[0023] By adopting the above technical solution, by correcting the parameters of the image calculation results, the information of multiple images can be fitted, and the information of multiple images can be combined, and then the information of multiple images can be combined together to calculate the comprehensive parameters, thereby avoiding the problem of image information transmission errors, and avoiding image loss by storing the image, and then transmitting the stored data to the cloud server.

[0024] Working principle: When working, first read the file material in the storage card configuration through the reading device, then obtain multiple images from the video stream of the configuration file, then determine the storage date of the obtained images, and regularly delete image data older than one month to avoid continuous occupation of hard disk space, thereby freeing up some space on the hard disk, so that the hard disk can be used for a longer time, and grayscale processing is performed on the deleted images, converted into grayscale images, and then Gaussian blur processing is performed on the grayscale images respectively, and the Gaussian mixture model GMM classification is trained offline through the image training library with and without fog image classifier The Gaussian mixture model GMM classifier extracts the one-dimensional grayscale histogram image features of the foggy image and the fog-free image according to the grayscale value frequency characteristics of the foggy image, and classifies the foggy image and the fog-free image according to the image histogram. The Gaussian mixture model GMM classifier for foggy and foggy images and the support vector machine SVM classifier for foggy and foggy images are used to classify the processed road image as foggy or foggy respectively, and the foggy image classified by the Gaussian mixture model GMM classifier for foggy and foggy images and the foggy image classified by the support vector machine SVM classifier for foggy and foggy images are taken as the union to obtain the real-time foggy road image. Image blurring and image grayscale processing can be used as preprocessing steps for image processing to prepare for the subsequent upper-level operations such as image segmentation, image recognition and image analysis. Then, the image after Gaussian blur processing is subjected to adaptive threshold processing. When performing threshold processing, the adaptive threshold processing obtains the threshold by calculating the weighted average of the adjacent areas around each pixel, and uses the threshold to process the current pixel. Since noise is inevitable in the actual imaging process, the differential image also contains noise, but the value of the noise is usually small. Therefore, an appropriate threshold can be used to better detect the moving object in the differential image. The brightness difference of the image is processed by adaptive threshold processing, and then the visibility of the processed image is calculated. Then, the result of the image visibility is calculated, and the result is parameter corrected. By fitting the information of multiple images, the information of multiple images can be combined, and then the information of multiple images can be combined together to calculate the comprehensive parameters, so as to avoid the problem of image information transmission error, and then the parameter correction of the result is realized. Then the processed image data is stored through the storage device, and finally the stored image data is sent to the cloud server.

[0025] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A cloud fog intelligent recognition technology based on AI video deep learning, characterized by: First, the file material in the storage card configuration is read by a reading device, and then an image is obtained from the video stream of the read configuration file, and then the image data one month ago is deleted from the obtained image, and the deleted image is grayed and converted into a grayscale image, and then Gaussian blur processing is performed on the grayscale image to blur the image, and then the image after Gaussian blur processing is performed on the image, and the brightness difference of the image is processed by adaptive threshold processing, and then the visibility of the processed image is calculated, and then the result of the image visibility is calculated, and the result is parameter corrected, and then the processed image data is stored through a storage device, and finally the stored image data is sent to a cloud server for storage.

2. The method of intelligent fog recognition based on AI video deep learning according to claim 1, characterized in that: The method for acquiring images is to directly pull the video in the configuration file to collect the image, and collect multiple image information in one video.

3. The method of intelligent fog recognition based on AI video deep learning according to claim 1, characterized in that: The method for deleting image data older than one month is to determine the retention and deletion of the image by determining the storage date of the image, regularly delete the image data older than one month, delete duplicate information, eliminate image errors caused by camera spot problems, and at the same time perform grayscale processing on the image data after deletion to make the image a grayscale image.

4. The method of intelligent fog recognition based on AI video deep learning according to claim 1, characterized in that: Gaussian blur is a widely used processing effect in image processing. By Gaussian blurring the image, image noise can be reduced and the level of detail can be lowered to ensure that the acquired image is clearer.

5. The method of intelligent fog recognition based on AI video deep learning according to claim 1, characterized in that: Adaptive threshold processing technology uses a variable threshold to complete the threshold processing of the image. When performing threshold processing, adaptive threshold processing obtains the threshold by calculating the weighted average of the adjacent area around each pixel, and uses the threshold to process the current pixel.

6. The method of intelligent fog recognition based on AI video deep learning according to claim 5, characterized in that: The threshold t0 is set. When the t value is less than t0, t=t0 is set. Taking t0=0.1 as the standard, the processed three-channel images are synthesized to restore the clear image J(x) after defogging.

7. The method of intelligent fog identification based on AI video deep learning according to claim 6, characterized in that: The image visibility observation model is the relationship between the brightness mean of the digital image transmittance distribution map and the atmospheric horizontal visibility obtained by curve fitting the digital image brightness mean near the ground object in the transmittance distribution map and the visibility value of the forward scattering instrument at the corresponding time. The image visibility observation model is expressed by the following polynomial: V(t)=a1T(t)4+a2T(t)3+a3T(t)2+a4T(t)+a5.

8. The method of intelligent fog identification based on AI video deep learning according to claim 1, characterized in that: Where t is the observation time, T(t) is the brightness mean of the image transmittance distribution diagram, V(t) is the atmospheric visibility value, a1-a5 are polynomial coefficients, and atmospheric visibility is the distance at which the parallel light attenuates through the atmosphere to 0.05 of the original luminous flux. The visibility value V can be derived, and the visibility value V can be obtained by the atmospheric extinction coefficient β obtained in the process of obtaining a clear image.

9. The method of intelligent fog identification based on AI video deep learning according to claim 8, characterized in that: Different fog image information error values ​​are recorded in different images. By fitting these error values, correction parameters are calculated and the calculation results are corrected.

10. The method of intelligent fog recognition based on AI video deep learning according to claim 1, characterized in that: The obtained correction parameters are applied to the measurement data, and a higher-precision measurement result is obtained by subtraction. The image after correction parameters is stored in the database and transmitted to the cloud server through the database.

Images