Intelligent inspection method for bridge piers

Abstract

The present invention belongs to the field of image processing technology, and specifically relates to an intelligent inspection method for bridge piers. The method comprises: performing Gaussian filtering on an inspection image using Gaussian filters of different scales, subtracting the original inspection image from the background images obtained at different scales to obtain intermediate images at different scales; calculating the degree of enhancement of the gradient features of the intermediate images of pixels at different scales compared to those in the original inspection image; calculating the degree of preservation of multiple LBP features of the intermediate images of pixels at different scales compared to those in the original inspection image; and calculating the weights of the grayscale values of the pixels in the intermediate images at different scales based on the degree of enhancement and the degree of preservation, performing weighted summation of the grayscale values of the pixels in the intermediate images at different scales to obtain a uniform light image. The present invention enables the ultimately obtained uniform light image to achieve a better balance in terms of detail and texture.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more particularly to a method for intelligent inspection of bridge piers. Background Art

[0002] With the rapid development of my country's transportation infrastructure, the number of bridges continues to increase, and the safe operation and maintenance of bridges has become a vital task; bridge piers are the key load-bearing structures of bridges, and their health status is directly related to the overall safety and service life of the bridge.

[0003] Traditional bridge pier inspections are mainly carried out manually. Workers need to use equipment such as boats, scaffolding or bridge inspection vehicles to approach the piers for close-range visual inspections and simple measurements. Inspection efficiency is low, especially for large bridges and bridge clusters distributed over a wide area. Inspection work is time-consuming and labor-intensive, and manual inspections carry certain safety risks. In particular, in severe weather conditions or complex water environments, the personal safety of workers cannot be fully guaranteed.

[0004] In order to overcome the shortcomings of traditional inspection methods and improve the efficiency, safety and accuracy of bridge pier inspections, drone technology has been introduced into the field of bridge inspection. Through the images and videos collected by drones, the surface condition of the piers can be clearly observed, and structural damage and defects can be discovered in a timely manner. At the same time, drone inspections can realize real-time data transmission and remote monitoring. Staff can analyze and evaluate inspection data in the control center, greatly improving the convenience and timeliness of inspection work.

[0005] However, due to the influence of environmental factors, lighting conditions, shooting angles and other factors, there is uneven lighting in the inspection videos of bridge piers collected by drones, which affects the subsequent accurate observation, analysis and evaluation of the surface condition of the bridge piers; the MASK uniform light algorithm uses a Gaussian filter to low-pass filter the inspection image, thereby obtaining and removing the background image with simulated brightness distribution from the inspection image, and obtaining an image with uniform lighting.

[0006] However, the Gaussian filter used in the MASK light-evening algorithm has a single scale and cannot balance the light-evening effect and detail information, which affects the inspection effect. Summary of the Invention

[0007] In order to solve the technical problem that the Gaussian filter used in the above-mentioned MASK light-evening algorithm has a single scale and cannot balance the light-evening effect and detail information, the present invention provides an intelligent inspection method for bridge piers, comprising: collecting inspection videos of bridge piers through a camera carried by a drone, and performing light-evening processing on each inspection image in the inspection video through the MASK light-evening algorithm, including: setting Gaussian filters of different scales; performing Gaussian filtering on the inspection image through Gaussian filters of different scales to obtain background images with simulated brightness distribution at different scales; performing subtraction between the original inspection image and the background image at different scales to obtain intermediate images at different scales. image; calculate the degree of enhancement of the gradient features of the pixel points in the intermediate images at different scales compared with the gradient features in the original inspection image; calculate the degree of preservation of the multiple LBP features of the pixel points in the intermediate images at different scales compared with the multiple LBP features in the original inspection image; calculate the weight of the grayscale value of the pixel points in the intermediate images at different scales according to the degree of enhancement and the degree of preservation, and perform weighted summation of the grayscale values of the pixel points in the intermediate images at different scales according to the weight of the grayscale value of the pixel points in the intermediate images at different scales as the grayscale value of the pixel points in the uniform light image, thereby obtaining the uniform light image.

[0008] The present invention constructs a brightness distribution model of the inspection image by setting Gaussian filters of different scales, providing richer background information for subsequent uniform light processing. For intermediate images at different scales, the degree of enhancement of the gradient features and the degree of preservation of multiple LBP features compared with the original inspection image are calculated. The weights of the grayscale values in the intermediate images at different scales are obtained by combining the degree of enhancement and the degree of preservation. Then, the information of the intermediate images at different scales is reasonably integrated through the weights, so that the uniform light image finally obtained achieves a better balance in details and texture, giving full play to the advantages of drones in collecting bridge pier inspection videos, realizing intelligent inspection of bridge piers, improving the level of bridge maintenance, and ensuring the safe operation of bridges.

[0009] Preferably, the setting of Gaussian filters of different scales includes: setting the range Each odd number within is considered as a standard deviation , used to construct Gaussian filters of different scales, and are the upper and lower limits of the range, respectively, and .

[0010] The present invention can more accurately construct the brightness distribution model of the inspection image by setting Gaussian filters of different scales, provide richer background information for subsequent uniform light processing, and thus avoid the problem of poor uniform light effect caused by inappropriate selection of filter scale.

[0011] Preferably, the step of calculating the enhancement degree of the gradient features of the pixel points in the intermediate images at different scales compared to the gradient features in the original inspection images includes: Where, , is the number of types of all scales, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, 、 The pixels are respectively The gradient magnitude and gradient direction in the intermediate image at different scales, 、 are the gradient amplitude and gradient direction of the pixel point in the original inspection image, Indicates taking the absolute value, Represents the Sigmoid function.

[0012] The present invention screens out scales that are better at enhancing image details, providing a basis for subsequently determining image grayscale value weights, so that the final uniform light image can better retain image detail information.

[0013] Preferably, the calculating of the preservation degree of the multiple LBP features of the pixel points in the intermediate image at different scales compared with the multiple LBP features in the original inspection image includes: Where, , is the number of types of all scales, For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, For the pixel point Multiple LBP features in the intermediate image at different scales, is the multiple LBP features of the pixel in the original inspection image, represents the Hamming distance.

[0014] The degree of preservation of the multiple LBP features of pixel points in the intermediate images at different scales calculated by the present invention compared with the multiple LBP features in the original inspection image is helpful to evaluate the performance of intermediate images of different scales in preserving image texture details, so that in subsequent processing, the relationship between detail enhancement and texture preservation can be balanced to avoid texture distortion caused by excessive enhancement or detail loss caused by excessive texture preservation.

[0015] Preferably, the step of calculating the weights of the grayscale values of the pixels in the intermediate images at different scales according to the enhancement degree and the preservation degree includes: Where, , is the number of types of all scales, For the pixel point The weight of the gray value in the intermediate image at different scales, For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, To keep the level to a minimum, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, Indicates taking the maximum value, Represents the normalization function.

[0016] By calculating the weights of the grayscale values of pixels in intermediate images at different scales, the present invention can rationally fuse the information of intermediate images at different scales, so that the final uniformly illuminated image achieves a better balance in terms of details and texture, avoiding the information loss that may result from simply averaging and fusing images of different scales.

[0017] Preferably, the method for obtaining the multiple LBP features of the pixel point includes: constructing a neighborhood with a size of 5×5 with the pixel point as the center pixel point, and the obtained neighborhood includes 1 center pixel point and 24 neighboring pixel points; calculating the absolute value of the difference between the grayscale value of each neighboring pixel point and the center pixel point as the grayscale difference between each neighboring pixel point and the center pixel point ; By putting The grayscale difference between the neighboring pixels and the central pixel Compare with the 4 sub-regions and obtain the The label values of the neighborhood positions corresponding to the neighborhood pixels are arranged in sequence and concatenated into a string as the multiple LBP features of the pixel points.

[0018] Preferably, the method for obtaining the four sub-areas includes: There are three threshold values set in 、 、 ,and ; The range is divided into three parts by 3 dividing values Divided into 4 sub-areas, namely , sub-area , sub-area , sub-area .

[0019] Preferably, the obtaining The label value of the neighborhood position corresponding to the neighborhood pixel point includes: if the grayscale difference In the sub-area Inside, then The mark value of the neighboring position corresponding to the neighboring pixel point is 00; if the grayscale difference In the sub-area Inside, then The label value of the neighboring position corresponding to the neighboring pixel point is 01; if the grayscale difference In the sub-area Inside, then The mark value of the neighboring position corresponding to the neighboring pixel point is 10; if the grayscale difference In the sub-area Inside, then The label value of the neighborhood position corresponding to the neighborhood pixel point is 11.

[0020] Preferably, the normalization function includes: , For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, To keep the level to a minimum, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, Indicates taking the maximum value, is the number of types at all scales.

[0021] Preferably, the method further comprises: performing stretching display processing on the uniform light image, wherein the stretching display processing method comprises 2% linear stretching and contrast parameter stretching.

[0022] The beneficial effects of the present invention are:

[0023] The present invention constructs a brightness distribution model of the inspection image by setting Gaussian filters of different scales, providing richer background information for subsequent uniform light processing. For intermediate images at different scales, the degree of enhancement of the gradient features and the degree of preservation of multiple LBP features compared with the original inspection image are calculated. The weights of the grayscale values in the intermediate images at different scales are obtained by combining the degree of enhancement and the degree of preservation. Then, the information of the intermediate images at different scales is reasonably integrated through the weights, so that the uniform light image finally obtained achieves a better balance in details and texture, giving full play to the advantages of drones in collecting bridge pier inspection videos, realizing intelligent inspection of bridge piers, improving the level of bridge maintenance, and ensuring the safe operation of bridges. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flow chart schematically illustrating a method for intelligent inspection of bridge piers in the present invention;

[0025] Figure 2 is a schematic diagram schematically illustrating an inspection image with uneven illumination;

[0026] Figure 3 The schematic diagram shows the MASK dodging algorithm Figure 2 Schematic diagram of the uniform light image obtained when performing uniform light processing;

[0027] Figure 4 The method of the present invention is schematically shown Figure 2 Schematic diagram of the uniform light image obtained when performing uniform light processing. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0029] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0030] The embodiment of the present invention discloses a bridge pier intelligent inspection method, referring to Figure 1 , including steps S1 to S3:

[0031] S1. Use the camera on the drone to collect inspection videos of bridge piers, including multiple inspection images.

[0032] Specifically, according to the bridge pier height, inspection range and accuracy requirements, a suitable model of industrial-grade drone is selected; using geographic information system (GIS) and satellite positioning technology (such as GPS or Beidou), combined with the precise coordinate data of the bridge piers, the drone's inspection flight path is pre-planned to ensure that the drone can perform circumferential shooting along multiple key points evenly distributed around the piers; then, the drone is controlled to follow the inspection flight path, and the onboard camera is used to collect inspection videos of the bridge piers. The obtained inspection videos are composed of multiple inspection images.

[0033] Furthermore, during the flight of the drone, the collected inspection video is transmitted back to the ground control station in real time through a wireless communication module (such as a 4G network or a 5G network). The ground control station can be deployed in a monitoring room near the bridge or in an operation and maintenance vehicle and equipped with professional image receiving and processing equipment to ensure stable video reception.

[0034] It should be noted that drones can quickly reach the vicinity of bridge piers and cover a large inspection area. Compared with traditional manual inspection methods, they greatly shorten the inspection time and improve work efficiency. At the same time, drones can replace humans to enter dangerous areas for close-up photography, reducing personnel safety risks.

[0035] S2. Gaussian filtering is performed on the inspection image through Gaussian filters of different scales to obtain background images with simulated brightness distribution at different scales. The original inspection image is subtracted from the background images at different scales to obtain intermediate images at different scales.

[0036] It should be noted that the inspection videos of bridge piers collected by drones can realize remote monitoring, clearly observe the surface conditions of bridge piers, and discover potential defects in time, which greatly improves the convenience and timeliness of inspection work; however, when collecting inspection videos of bridge piers by drones, they are affected by environmental factors, lighting conditions, shooting angles and other factors, resulting in different degrees of differences in brightness, hue and contrast within the inspection images, that is, there is uneven lighting, which affects the subsequent accurate observation, analysis and evaluation of the surface conditions of the bridge piers. Therefore, eliminating the uneven lighting phenomenon in the inspection videos of bridge piers and performing uniform light processing on the inspection videos of bridge piers have important theoretical and practical application value for obtaining high-quality inspection videos.

[0037] For example, a schematic diagram of an inspection image with uneven illumination is shown as follows: Figure 2 shown.

[0038] It should be further explained that the MASK light-evening algorithm is a method proposed to address uneven image illumination. It primarily views the image as the superposition of a background image with uneven brightness and an image with uniform illumination. From a frequency domain perspective, the brightness distribution of an image is primarily reflected in low-frequency information, while high-frequency information includes edges, details, and noise. By enhancing high-frequency information and suppressing low-frequency information, the image's detail contrast is enhanced, abnormal brightness variations are suppressed, and a uniform brightness distribution is achieved. Therefore, the MASK light-evening algorithm low-pass filters the original image using a Gaussian filter to obtain a background image with simulated brightness distribution. The original image is then subtracted from the background image to obtain an image with uniform illumination.

[0039] It should be noted that the Gaussian function in the Gaussian filter belongs to the normal distribution. Therefore, the larger the standard deviation of the Gaussian function, the larger the scale of the Gaussian filter, the more pixels involved in the calculation but the wider the distribution, the greater the degree of blur, and the insufficient simulation effect of the local brightness distribution, resulting in poor uniformity of the final uniformity image; and the smaller the standard deviation of the Gaussian function, the smaller the scale of the Gaussian filter, the fewer but more concentrated pixels involved in the calculation, and the obtained background image can better simulate the brightness distribution, but it also contains more detail information, resulting in the loss of local detail information in the final uniformity image.

[0040] Therefore, the inspection image is Gaussian filtered by Gaussian filters of different scales to obtain background images with simulated brightness distribution at different scales. The inspection image is subtracted from the background images at different scales to obtain intermediate images at different scales. The intermediate images at different scales are then combined to obtain the final uniform light image.

[0041] 1. Set Gaussian filters of different scales.

[0042] It should be noted that due to the limitations of a single-scale Gaussian filter: when the standard deviation is large, the local brightness simulation effect is not good; when the standard deviation is small, it is easy to cause the loss of local details; by setting Gaussian filters of different scales with different standard deviations, Gaussian filters of different scales can simulate the brightness distribution in different ranges respectively, from large-scale brightness changes to small-scale local brightness details, and the brightness distribution information of the inspection image can be obtained more comprehensively.

[0043] Specifically, the range Each odd number within is considered as a standard deviation , used to construct Gaussian filters of different scales, and are the upper and lower limits of the range, respectively, and ; As the standard deviation increases, the smoothing effect of the constructed Gaussian filter becomes more obvious.

[0044] Among them, the range middle and The specific value of can be set according to the actual application scenario and requirements, and ,in, The value range of is [7,11], The value range of is [15,19]. 、 Set to 9 and 15 respectively.

[0045] It should be noted that compared with Gaussian filtering of a single scale, setting Gaussian filters of different scales can more accurately construct the brightness distribution model of the inspection image, provide richer background information for subsequent uniform light processing, and thus avoid the problem of poor uniform light effect caused by improper selection of filter scale.

[0046] 2. Perform Gaussian filtering on the inspection image through Gaussian filters of different scales to obtain background images with simulated brightness distribution at different scales; perform subtraction between the original inspection image and the background images at different scales to obtain intermediate images at different scales.

[0047] It should be noted that the original inspection image contains background information with uneven brightness and image detail information. By subtracting the original inspection image from background images of different scales, the uneven brightness component in the original inspection image can be removed, and an intermediate image mainly containing image detail information can be obtained. The intermediate images at different scales correspond to the extraction of detail information at different scales.

[0048] S3. According to the weights of the grayscale values of the pixel points in the intermediate images at different scales, a weighted sum is performed on the grayscale values of the pixel points in the intermediate images at different scales, and the sum is used as the grayscale value of the pixel points in the uniform light image, thereby obtaining the uniform light image.

[0049] 1. Obtain intermediate images of pixels at different scales and detailed features in the original inspection image.

[0050] Specifically, the detailed features of the pixel point include the gradient feature and multiple LBP features of the pixel point; the gradient feature of the pixel point is obtained by calculating the Sobel operator, and the gradient feature includes the gradient direction and gradient amplitude; the multiple LBP features of the pixel point are obtained by calculating the gradient direction and gradient amplitude in the range It is obtained by calculating the multiple LBP operators constructed by setting multiple cutoff values.

[0051] Among them, the LBP operator is an operator used to describe the local features in the image. The multiple LBP operators constructed by the multiple cutoff values set in the algorithm are used to calculate the multiple LBP features of the pixel points. The specific steps include:

[0052] (1) Construct a neighborhood with a size of 5×5 and a pixel point as the center pixel point. The obtained neighborhood includes 1 center pixel point and 24 neighboring pixel points.

[0053] (2) Calculate the absolute value of the difference between the grayscale value of each neighborhood pixel and the central pixel as the grayscale difference between each neighborhood pixel and the central pixel .

[0054] (3) In scope There are three threshold values set in 、 、 ,and ; The range is divided into three parts by 3 dividing values Divided into 4 sub-areas, namely , sub-area , sub-area , sub-area .

[0055] Among them, the cutoff value 、 、 The specific value can be set according to the actual application scenario and requirements, and the demarcation value The value range is [4,7], the cutoff value The value range is [9,12], the cutoff value The value range is [18,25]. The present invention divides the 、 、 Set to 5, 10, and 20 respectively.

[0056] (4) By The grayscale difference between the neighboring pixels and the central pixel Compare with the 4 sub-regions and obtain the The label value of the neighborhood position corresponding to the neighborhood pixel point:

[0057] If the grayscale difference In the sub-area Inside, then The mark value of the neighborhood position corresponding to the neighborhood pixel point is 00;

[0058] If the grayscale difference In the sub-area Inside, then The label value of the neighborhood position corresponding to the neighborhood pixel point is 01;

[0059] If the grayscale difference In the sub-area Inside, then The label value of the neighborhood position corresponding to the neighborhood pixel point is 10;

[0060] If the grayscale difference In the sub-area Inside, then The label value of the neighborhood position corresponding to the neighborhood pixel point is 11.

[0061] (5) Arrange the label values of the neighborhood positions corresponding to all neighborhood pixels in sequence and concatenate them into a string as the multiple LBP features of the pixel points. Since the label value of the neighborhood position corresponding to each neighborhood pixel point is composed of two values, 0 and 1, and each label value includes two values, the multiple LBP features of the pixel point are essentially a binary sequence composed of two values, 0 and 1, and the length is equal to 24×2=48.

[0062] It should be noted that LBP feature is an effective method for describing local texture features of an image, and multiple LBP features can describe the texture details of an image from multiple angles.

[0063] 2. Calculate the weight of the grayscale value of the pixel in the intermediate image at different scales based on the change in the detail features of the pixel in the intermediate image at different scales compared to the detail features in the original inspection image.

[0064] Specifically, the detail features of the pixel points include the gradient features and multiple LBP features of the pixel points. Therefore, first, based on the gradient features of the pixel points in the intermediate images at different scales and the original inspection images, the degree of enhancement of the gradient features of the pixel points in the intermediate images at different scales compared with the gradient features in the original inspection images is calculated; secondly, based on the intermediate images at different scales and the multiple LBP features of the pixel points in the original inspection images, the degree of preservation of the multiple LBP features of the pixel points in the intermediate images at different scales compared with the multiple LBP features in the original inspection images is calculated; finally, based on the degree of enhancement of the gradient features of the pixel points in the intermediate images at different scales compared with the gradient features in the original inspection images and the degree of preservation of the multiple LBP features of the pixel points in the intermediate images at different scales compared with the multiple LBP features in the original inspection images, the weights of the grayscale values of the pixel points in the intermediate images at different scales are calculated.

[0065] The specific calculation process is:

[0066] (1) Based on the gradient features of the intermediate images at different scales and the original inspection images of the pixels, the degree of enhancement of the gradient features of the intermediate images at different scales compared with the gradient features of the original inspection images is calculated. The calculation formula is:

[0067] ;

[0068] Where, , is the number of types of all scales, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, 、 The pixels are respectively The gradient magnitude and gradient direction in the intermediate image at different scales, 、 are the gradient amplitude and gradient direction of the pixel point in the original inspection image, Indicates taking the absolute value, Represents the Sigmoid function.

[0069] Among them, the larger the gradient amplitude of the pixel point, the stronger the gradient feature of the pixel point. Therefore, by calculating the pixel point in the first Gradient magnitude in the intermediate image at different scales The gradient amplitude of the pixel in the original inspection image The difference , difference The larger the value, the higher the pixel The gradient features in the intermediate image at this scale are stronger than those in the original inspection image; on this basis, the difference in the gradient direction between the two is Correction is made when the difference between the two in the gradient direction The smaller it is, the greater the enhancement of the corrected gradient feature. Used for differences Perform normalization.

[0070] It should be noted that gradient features are important indicators for measuring edges and other detail information in an image. In the intermediate image, since some effects of uneven brightness have been removed, the gradient features may be enhanced. By comparing the gradient features of pixels in the intermediate image and the original image, the enhancement effect of the intermediate image on detail information at different scales can be quantified, which helps to evaluate the contribution of intermediate images of different scales in highlighting details.

[0071] (2) Based on the multiple LBP features of the intermediate images at different scales and the original inspection images, the degree of preservation of the multiple LBP features of the intermediate images at different scales compared with the multiple LBP features in the original inspection images is calculated. The calculation formula is:

[0072] ;

[0073] Where, , is the number of types of all scales, For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, For the pixel point Multiple LBP features in the intermediate image at different scales, is the multiple LBP features of the pixel in the original inspection image, represents the Hamming distance.

[0074] Among them, the pixel point Hamming distance between the multiple LBP features in the intermediate image at different scales and the multiple LBP features of the pixel in the original inspection image The smaller it is, the more pixels are in the The greater the similarity between the multiple LBP features in the intermediate image at this scale and the multiple LBP features of the pixel in the original inspection image, the more likely the pixel is to be in the first The better the local detail features in the intermediate image at this scale are preserved, the 48 in the denominator is used to calculate the Hamming distance. Perform normalization.

[0075] It should be noted that for objects with obvious texture features such as the surface of a bridge pier, texture information is an important factor in evaluating image quality. Therefore, by comparing the multiple LBP features of pixels in the intermediate image and the original image, the ability of the intermediate image to preserve image texture information at different scales can be determined.

[0076] (3) According to the degree of enhancement of the gradient features of the pixel points in the intermediate images at different scales compared with the gradient features in the original inspection image and the degree of preservation of the multiple LBP features of the pixel points in the intermediate images at different scales compared with the multiple LBP features in the original inspection image, the weight of the grayscale value of the pixel points in the intermediate images at different scales is calculated. The calculation formula is:

[0077] ;

[0078] Where, , is the number of types of all scales, For the pixel point The weight of the gray value in the intermediate image at different scales, For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, To keep the level to a minimum, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, Indicates taking the maximum value, represents the normalization function. In this embodiment, , is the number of types at all scales.

[0079] Among them, maintain the minimum The specific value of can be set according to the actual application scenario and requirements, and The value range of is [0.65, 0.85]. Set to 0.78.

[0080] Among them, when the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image Greater than the minimum level of maintenance When , it means the pixel point is The intermediate image at this scale maintains the same local detail features as in the original inspection image, so , and as the degree of maintenance Increase, the pixel point The greater the weight of the gray value in the intermediate image at this scale, the more pixels are The weight of the gray value in the intermediate image at this scale changes with the pixel point at the The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image increases with the increase of The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image Less than or equal to the minimum level of maintenance hour, , which means the pixel is at The local detail features in the intermediate image at this scale have changed significantly, so the pixel point The weights of the grayscale values in the intermediate images at these scales remain equal to 0.

[0081] It should be noted that intermediate images of different scales have their own advantages and disadvantages in detail enhancement and texture preservation. By comprehensively considering the degree of enhancement of gradient features and the degree of preservation of multiple LBP features, a weight can be assigned to the grayscale value of the intermediate image at different scales for each pixel. This weight reflects the contribution of the intermediate image to the pixel at that scale. Images with higher weights have better overall performance in detail enhancement and texture preservation.

[0082] 3. According to the weights of the grayscale values of the pixel points in the intermediate images at different scales, the grayscale values of the pixel points in the intermediate images at different scales are weighted summed as the grayscale value of the pixel points in the uniform light image, thereby obtaining the uniform light image.

[0083] For example, the MASK dodging algorithm is used to Figure 2 When performing the light uniformity processing, the schematic diagram of the light uniformity image obtained is as follows: Figure 3 As shown; by the method of the present invention Figure 2 When performing the light uniformity processing, the schematic diagram of the light uniformity image obtained is as follows: Figure 4 As shown; by comparison Figure 3 and Figure 4 It can be seen that Figure 3 There are still obvious grayscale jumps in the light area and shadow area, and the uniform light image appears gray and foggy as a whole. Figure 4 Not only is the transition between light and dark natural, but it also highlights details such as cracks in the uniformly lit image, improving image clarity and creating a stronger lighting consistency effect.

[0084] It should be noted that based on the weights calculated previously, the grayscale values of the intermediate images of different scales are weighted and summed up. This can combine the advantages of each scale and make full use of the information of the intermediate images of different scales in detail enhancement and texture preservation to generate the final uniform light image, in which the grayscale value of each pixel can better reflect the actual condition of the bridge pier surface, eliminate the influence of uneven lighting, and thus obtain a high-quality uniform light image, so that the bridge pier inspection video can more accurately display the real situation of the bridge pier surface, facilitate more effective observation, analysis and evaluation of the surface condition of the bridge pier, and improve the quality and efficiency of the inspection work.

[0085] In addition, since the previous processing process has undergone a subtraction operation, the contrast of the overall uniform light image, especially the darker areas, will be reduced. In order to improve the overall contrast of the image while maintaining the overall contrast consistency of the image, highlighting the image details, and maintaining image clarity, the uniform light image needs to be stretched and displayed to finally obtain a uniform light image.

[0086] Among them, commonly used stretching display methods include but are not limited to 2% linear stretching and contrast parameter stretching, and 2% linear stretching and contrast parameter stretching are both well-known technologies and will not be described in detail here.

Claims

1. A bridge pier intelligent inspection method, characterized in that: include: The camera on the drone collects inspection videos of bridge piers, and the MASK light-dodging algorithm is used to perform light-dodging on each inspection image in the inspection video, including: Set Gaussian filters of different scales; perform Gaussian filtering on the inspection image through Gaussian filters of different scales to obtain background images with simulated brightness distribution at different scales; perform subtraction between the original inspection image and the background images at different scales to obtain intermediate images at different scales; Calculate the degree of enhancement of the gradient features of the pixel points in the intermediate image at different scales compared to the gradient features in the original inspection image, including: , , is the number of types of all scales, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, 、 The pixels are respectively The gradient magnitude and gradient direction in the intermediate image at different scales, 、 are the gradient amplitude and gradient direction of the pixel point in the original inspection image, Indicates taking the absolute value, Represents the Sigmoid function; Calculate the degree of preservation of the multiple LBP features in the intermediate image at different pixel scales compared to the multiple LBP features in the original inspection image, including: , For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, For the pixel point Multiple LBP features in the intermediate image at different scales, is the multiple LBP features of the pixel in the original inspection image, represents the Hamming distance; According to the degree of enhancement and the degree of preservation, the weights of the grayscale values of the pixels in the intermediate images at different scales are calculated. According to the weights of the grayscale values of the pixels in the intermediate images at different scales, the weighted sum of the grayscale values of the pixels in the intermediate images at different scales is taken as the grayscale value of the pixels in the uniform light image, thereby obtaining the uniform light image.

2. The intelligent inspection method for bridge piers according to claim 1, characterized in that: The setting of Gaussian filters of different scales includes: The range Each odd number within is considered as a standard deviation , used to construct Gaussian filters of different scales, and are the upper and lower limits of the range, respectively, and .

3. The intelligent inspection method for bridge piers according to claim 1, characterized in that: Calculating the weights of the grayscale values of the pixels in the intermediate images at different scales according to the enhancement degree and the preservation degree includes: ； Where, , is the number of types of all scales, For the pixel point The weight of the grayscale value in the intermediate image at different scales, For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, To keep the level to a minimum, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, Indicates taking the maximum value, Represents the normalization function.

4. The intelligent inspection method for bridge piers according to claim 3 is characterized in that: The normalization function includes: , For the pixel point The degree of preservation of the multiple LBP features in the intermediate image at different scales compared to the multiple LBP features in the original inspection image, To keep the level to a minimum, For the pixel point The degree of enhancement of the gradient features in the intermediate image at this scale compared to the gradient features in the original inspection image, Indicates taking the maximum value, is the number of types at all scales.

5. The intelligent inspection method for bridge piers according to claim 1, characterized in that: The method for obtaining the multiple LBP features of the pixel point includes: Construct a 5×5 neighborhood with the pixel as the center pixel. The obtained neighborhood includes 1 center pixel and 24 neighboring pixels. Calculate the absolute value of the difference between the grayscale values of each neighborhood pixel and the central pixel as the grayscale difference between each neighborhood pixel and the central pixel ; By putting the The grayscale difference between the neighboring pixels and the central pixel Compare with the 4 sub-regions and obtain the The label value of the neighborhood position corresponding to the neighborhood pixel point; The label values of the neighborhood positions corresponding to all neighborhood pixels are arranged in sequence and concatenated into a string as the multiple LBP features of the pixel.

6. The intelligent inspection method for bridge piers according to claim 5, characterized in that: The method for obtaining the four sub-areas includes: In range There are three threshold values set in 、 、 ,and ; The range is divided into three parts by 3 dividing values. Divided into 4 sub-areas, namely , sub-area , sub-area , sub-area .

7. The intelligent inspection method for bridge piers according to claim 5, characterized in that: The obtained The label value of the neighborhood position corresponding to the neighborhood pixel point includes: If the grayscale difference In the sub-area Inside, then The mark value of the neighborhood position corresponding to the neighborhood pixel point is 00; If the grayscale difference In the sub-area Inside, then The label value of the neighborhood position corresponding to the neighborhood pixel point is 01; If the grayscale difference In the sub-area Inside, then The label value of the neighborhood position corresponding to the neighborhood pixel point is 10; If the grayscale difference In the sub-area Inside, then The label value of the neighborhood position corresponding to the neighborhood pixel point is 11.

8. The intelligent inspection method for bridge piers according to claim 1, characterized in that: The method further comprises: The uniform light image is stretched and displayed, and the stretching display processing methods include 2% linear stretching and contrast parameter stretching.

Images