Bridge catastrophe identification method and system based on image mode

Abstract

The invention relates to a bridge catastrophe identification method based on an image mode. The method comprises the following steps: acquiring a bridge target image and a bridge floor information image in real time; performing target point identification on the bridge target image, and calculating deviation of each target point along the bridge direction at different moments; recognizing the bridge floor information image through a prediction model obtained through pre-training to judge whether the appearance contour state of the bridge is abnormal or not; and judging whether to give an alarm or not according to the deviation and the bridge appearance contour state. According to the bridge catastrophe recognition method based on the image mode, the bridge target image and the bridge floor information image are collected in real time, whether major catastrophe such as bridge collapse occurs or not is accurately judged by judging the deviation amount and the bridge appearance contour state, alarm pushing is carried out in real time, and after the major catastrophe such as the bridge collapse accident occurs, the bridge target image and the bridge floor information image are recognized. And a secondary accident is caused.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge structure safety monitoring, and in particular to a bridge disaster identification method and system based on image patterns. Background Art

[0002] During the long-term operation of bridge structures, adverse factors such as extreme environments, excessive vehicle loads, and sudden, accidental forces inevitably lead to degradation of the safety and durability of bridge structures. Failure to promptly detect damage to bridge structures and implement appropriate maintenance and repairs can severely impact the normal operation of bridges and increase maintenance and repair costs. In severe cases, the service life of bridges can be shortened, leading to sudden damage and collapse, resulting in significant loss of life and property.

[0003] Small and medium-span bridges are the most numerous and widely distributed type of existing bridges in my country. Consequently, they are most susceptible to collapse accidents caused by extreme environmental conditions or sudden, accidental events. If a bridge collapse is not promptly reported, it can hinder vehicle response, leading to secondary accidents such as vehicle falls and collisions. Summary of the Invention

[0004] The present application aims to solve the technical problem in the related art that when a small or medium span bridge collapse accident occurs, an alarm cannot be given in time, which may lead to secondary accidents.

[0005] The present invention provides a method for identifying bridge disasters based on image patterns, comprising the following steps:

[0006] Real-time acquisition of bridge target images and bridge deck information images;

[0007] Performing target point recognition on the bridge target image, and calculating the deviation of each target point along the bridge direction at different times;

[0008] The bridge deck information image is recognized by a pre-trained prediction model to determine whether there is any abnormality in the appearance and contour of the bridge;

[0009] Determine whether to send an alarm based on the deviation and the appearance and contour status of the bridge.

[0010] In one embodiment, the bridge disaster identification method based on image patterns further includes:

[0011] Recognizing the bridge deck information image based on a background difference method to determine whether there is foreign matter intrusion on the bridge deck;

[0012] Determine whether to send an alarm based on the status of foreign objects on the bridge surface.

[0013] In one embodiment, the performing target point recognition on the bridge target image and calculating the deviation of each target point along the bridge direction at different times includes:

[0014] Converting the bridge target image into a binary image based on a Kmeans algorithm;

[0015] Performing denoising on the binarized image based on a Gaussian filtering algorithm;

[0016] Perform contour detection on the denoised image to identify the pixel center coordinates of each target point;

[0017] The pixel center coordinates of each target point are fitted with a least squares straight line to calculate the deviation of the target point along the bridge direction at different times.

[0018] In one embodiment, the pre-training process of the prediction model includes:

[0019] Obtain a positive sample set with normal bridge appearance and contour status and a negative sample set with abnormal bridge appearance and contour status;

[0020] Converting the positive sample set and the negative sample set into grayscale images;

[0021] semantically annotating the grayscale images, and dividing the annotated grayscale images into a training set and a validation set;

[0022] A prediction model is constructed and trained based on the training set and the validation set.

[0023] In one embodiment, when obtaining a negative sample set of abnormal bridge appearance contour states, negative sample sets of different bridge types and different collapse conditions are obtained through simulation.

[0024] In one embodiment, the identifying the bridge deck information image based on a background difference method to determine whether there is foreign object intrusion on the bridge deck includes:

[0025] Establishing a mixed Gaussian model according to the grayscale value of each pixel in the bridge deck information image;

[0026] Calculating the matching degree between each pixel point in a bridge deck information image and each Gaussian distribution in the mixed Gaussian model;

[0027] If it matches, it is considered as background;

[0028] If there is no match, it is considered as foreground.

[0029] In one embodiment, determining whether to push an alarm based on the deviation amount and the bridge appearance and contour status includes:

[0030] When the identified target points are less than the target number and the bridge appearance contour is normal, a first alarm message is pushed: part or all of the targets are lost;

[0031] When the number of identified target points is less than the target number and the bridge appearance contour state is abnormal, a second alarm message is pushed: target lost and contour recognition abnormal;

[0032] When the number of identified target points is normal but the deviation is greater than the set threshold, and the bridge appearance is normal, a third alarm message is pushed: the target is complete but the target deviation is too large;

[0033] If the number of identified target points is normal but the deviation is less than the set threshold, and the bridge appearance contour is abnormal, the data will not be pushed;

[0034] When the number of identified target points is normal but the deviation is greater than the set threshold, and the bridge appearance contour state is abnormal, a fourth alarm message is pushed: the target is complete but the target deviation is too large, and the contour recognition is abnormal.

[0035] In one embodiment, when identifying the deviation or the bridge appearance profile state, if the abnormality exceeds a set number of times continuously, it is identified as a state abnormality.

[0036] The present application also provides an image-based bridge disaster identification system, which uses any of the above-mentioned image-based bridge disaster identification methods. The image-based bridge disaster identification system includes:

[0037] The image acquisition module is configured to: acquire bridge target images and bridge deck information images in real time;

[0038] The first edge computing module is configured to: identify target points on the bridge target image and calculate the deviation of each target point along the bridge direction at different times;

[0039] The second edge computing module is configured to: identify the bridge deck information image using a pre-trained prediction model to determine whether there is any abnormality in the appearance and contour of the bridge;

[0040] The alarm push module is configured to determine whether to push an alarm based on the deviation amount and the appearance and contour status of the bridge.

[0041] In one embodiment, the second edge computing module is further configured to: identify the bridge deck information image based on a background difference method to determine whether there is foreign object intrusion on the bridge deck.

[0042] The beneficial effects of the technical solutions provided in the embodiments of the present application include:

[0043] The present application provides a bridge disaster identification method based on image patterns, which collects bridge target images and bridge deck information images in real time. On the one hand, by identifying target points in the bridge target images, the deviation of each target point along the bridge direction at different times is calculated; on the other hand, a prediction model is pre-trained based on deep learning, and the appearance contour state of the bridge is directly learned to determine whether the bridge has collapsed; by judging the deviation and the appearance contour state of the bridge, it is accurately determined whether a major disaster such as bridge collapse has occurred, and an alarm is pushed in real time to prevent secondary accidents after major disasters such as bridge collapse accidents occur. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0045] Figure 1 Flowchart of a bridge disaster identification method based on image patterns in one embodiment of the present invention.

[0046] Figure 2 Schematic diagram of the installation position of the image acquisition module in one embodiment of the present invention.

[0047] Figure 3 Flowchart of a Gaussian filtering algorithm in one embodiment of the present invention.

[0048] In the picture: 1. Camera; 2. Target. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the present invention, 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 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 this application.

[0050] like Figure 1 As shown, Figure 1 Flowchart of a bridge disaster identification method based on image patterns in one embodiment of the present invention.

[0051] This embodiment provides a bridge disaster identification method based on image patterns, comprising the following steps:

[0052] Step S1, real-time acquisition of bridge target images and bridge deck information images;

[0053] Step S2: identifying target points on the bridge target image and calculating the deviation of each target point along the bridge direction at different times;

[0054] Step S3: Recognize the bridge deck information image using the pre-trained prediction model to determine whether there is any abnormality in the bridge appearance contour state;

[0055] Step S4: Determine whether to send an alarm based on the deviation amount and the appearance and contour status of the bridge.

[0056] This embodiment provides a bridge disaster recognition method based on image patterns, which collects bridge target images and bridge deck information images in real time. On the one hand, by identifying target points in the bridge target images, the deviation of each target point along the bridge direction at different times is calculated; on the other hand, a prediction model is pre-trained based on deep learning, and the bridge appearance contour state is directly learned to determine whether the bridge has collapsed; by judging the deviation and the bridge appearance contour state, it is accurately determined whether a major disaster such as bridge collapse has occurred, and an alarm is pushed in real time to prevent secondary accidents after major disasters such as bridge collapse.

[0057] Each step is described and explained in detail below.

[0058] like Figure 2 As shown, Figure 2 Schematic diagram of the installation position of the image acquisition module in one embodiment of the present invention.

[0059] Taking a certain bridge as an example, highly reflective passive targets 2, such as 5 targets, are symmetrically installed in the middle of the bridge span and at the piers. In order to fully cover the entire bridge field of view, image acquisition modules 1 are arranged at both ends of the bridge. Each image acquisition module 1 contains N security cameras. Taking the image acquisition module 1 on the left as an example: among the N security cameras, one security camera is used to collect target information on one side of the bridge, and the remaining N-1 cameras are used to cover the bridge deck field of view. The value of N depends on the length of the bridge and the coverage field of view of the camera to ensure that the camera field of view at both ends can achieve full bridge deck coverage. For the convenience of description, in this embodiment, N is 2, and the camera that collects target information in real time is the first security camera A in the image acquisition module 1, which serves as the target video frame acquisition unit; the camera that collects bridge deck information in real time is the second security camera B, which serves as the bridge deck appearance information acquisition unit.

[0060] First Security Camera A captured the bridge target using infrared fill light in night mode, which is energy-efficient and concealed, facilitating image capture in dark environments where standard lighting is inadequate. The camera was also set to a fixed exposure setting, with the exposure adjusted to a low level to ensure image clarity. An 850nm filter was applied to the lens of First Security Camera A, allowing only light in the 850nm band to pass through, minimizing the impact of natural light. The camera sensor is also highly sensitive to light in the 850nm band, resulting in higher image quality. This ensures excellent image quality at all times and under varying lighting conditions, filtering out any interference.

[0061] The second security camera, Camera B, collects real-time information about the bridge surface. It's a dual-use camera that automatically switches between day and night modes depending on light intensity. In day mode, it uses visible light to capture color images; in night mode, it uses infrared light to capture clear black and white images. This ensures clear, all-weather coverage of the bridge's exterior.

[0062] In one embodiment, step S2, identifying target points on the bridge target image and calculating the deviation of each target point along the bridge direction at different times, includes:

[0063] Step S21: Convert the bridge target image into a binary image based on the Kmeans algorithm.

[0064] The Kmeans algorithm uses optical flow or feature matching technology to extract key feature points in the bridge target image. By setting thresholds, it filters out unnecessary changes (such as wind and rain). It has high reliability and is not affected by abnormal points.

[0065] Specifically, a. Take any two integer points on the grayscale histogram of the bridge target image b. Move the remaining points on the histogram to the left The distance is classified into two categories A1 and A2, one is the target in the bridge target image, and the other is the background in the bridge target image, so as to filter out the target and facilitate the subsequent positioning of the target. c. Calculate the center points a1 and a2 of the two types of data A1 and A2, and the dividing point is m1 = (a1 + a2) / 2. d. Repeat steps b and c to calculate the dividing point m i , until the change is less than the threshold eps = 1. e. The grayscale of the bridge target image is greater than m i The part is the foreground, which is smaller than m i The part is taken as the background and a binary image is obtained.

[0066] Step S22: De-noising the binary image based on a Gaussian filtering algorithm.

[0067] The image contour detection process is very sensitive to noise. The Gaussian filter algorithm is used to reduce noise in the foreground image of the binary image and smooth the image. The filter operator is generated according to the Gaussian formula. The matrix composed of each pixel in the foreground image of the binary image and its neighboring pixels is convolved with the filter operator to remove noise. The Gaussian filter operator is:

[0068]

[0069] like Figure 3 As shown, Figure 3 Flowchart of a Gaussian filtering algorithm in one embodiment of the present invention.

[0070] The specific process is as follows: Assuming that the coordinates of the center point (i, j) are (0, 0), when the convolution kernel is set to 3, only the 8 points closest to it need to be calculated, such as Figure 3 The first matrix in is shown. Substitute the coordinates in this matrix into the Gaussian filter operator formula, take σ = 1.5, and the weight matrix with a radius of 1 is as follows Figure 3 The second matrix is shown in . Because we need to use this weight matrix for weighted averaging, we need to normalize the original matrix. The specific operation is to find the sum of the second matrix to be 0.4783, and then divide the nine values of the matrix by 0.4783 to obtain the final convolution kernel (weight matrix).

[0071] Step S23: Perform contour detection on the denoised image to identify the pixel center coordinates of each target point.

[0072] Perform contour canny edge detection on the denoised image to obtain the two-dimensional pixel center point of each target point. The specific process is as follows:

[0073] a. Use the Sobel edge detection operator to obtain the pixel coordinates of the target edge points;

[0074] b. Perform least squares ellipse fitting on the edge pixels to obtain the center pixel coordinates of the target point;

[0075] The ellipse fitting equation is: Ax 2 +Bxy+Cy 2 +Dx+Ey+1=0;

[0076] Where x and y are the pixel coordinates of the edge point. According to the principle of least squares, the fitted objective function is:

[0077] In order to minimize F, it is necessary to make the partial derivatives of F equal to 0, and then the parameters of the ellipse [A, B, C, D, E] can be obtained.

[0078] Thus we can find the center of the ellipse:

[0079]

[0080] Step S24: performing least squares straight line fitting on the pixel center coordinates of each target point to calculate the deviation of the target point along the bridge direction at different times.

[0081] Specifically, the center coordinates of the five target points are fitted with a least squares straight line l: ax+by+c=0.

[0082] Calculate the distance of all points from the straight line and sum them up to get the deviation D t :

[0083]

[0084] Record the deviation D0 at the initial moment and calculate the deviation D at different moments t The ratio r of the deviation relative to the initial moment:

[0085]

[0086] Through the above scheme, the deviation is calculated based on the real-time acquisition of bridge target images. When the bridge structure is normal, multiple target points are roughly distributed on both sides of a straight line. If the distance between the detected points and the straight line is too large, or some points disappear, it is considered that the structure is abnormal, which helps to determine whether the bridge has suffered a major disaster such as collapse.

[0087] In one embodiment, the pre-training process of the prediction model in step S3 includes:

[0088] a. Obtain a set of positive samples showing normal bridge outlines and a set of negative samples showing abnormal bridge outlines. The positive sample set is relatively easy to collect. For example, by using a web crawler to search and filter, we collected 1,214 photos of normal bridges (as the positive sample set) and 148 photos of collapsed bridges (as the negative sample set showing abnormal bridge outlines). Because the number of negative samples is relatively small compared to the positive samples, we used image enhancement techniques to perform operations such as rotation, flipping, and scaling on the negative samples, expanding the negative sample set to 500 to achieve a consistent training ratio.

[0089] In one embodiment, when obtaining a negative sample set of abnormal bridge appearance and contour, a simulation is performed to obtain a negative sample set of different bridge types and different collapse conditions. The simulation simulates different failure modes of bridge structures to obtain training samples and improve the accuracy of subsequent predictions.

[0090] Improve the accuracy of algorithm recognition.

[0091] b. Convert the positive sample set and negative sample set into grayscale images.

[0092] Converting the positive and negative sample sets into grayscale images has many advantages. First, for identifying major disasters such as collapses, the model only needs to learn changes in appearance contours, eliminate color interference, and reduce noise in the data. Fewer features means a simpler model, reducing the tendency of overfitting. Second, grayscale images pay more attention to shape, edges, and textures, which are more effective for collapse identification tasks. Third, grayscale images have only one channel, while color images have three channels. The computational cost and memory requirements for training grayscale images are lower. Fourth, the second security camera has been using night mode and outputs black and white images, which can be directly predicted using a model trained with grayscale images.

[0093] c. Perform semantic annotation on the grayscale images, and divide the annotated grayscale images into a training set and a validation set.

[0094] Specifically, the open-source labeling tool labelImg was used to annotate the grayscale images of the positive and negative sample sets. The annotation information is (x, y, w, h, c), representing the horizontal coordinate, vertical coordinate, width, height, and category of the bridge in the image. Since there are only two categories, c is labeled as 0 for positive samples and 1 for negative samples. The dataset is then divided into training and validation sets in a ratio of 7:3. The training set samples are used for training, while the validation set images are used to optimize the trained model.

[0095] d. Build a prediction model and train the prediction model based on the training set and validation set.

[0096] Specifically, we used the YOLO 11 network architecture to train the samples, balancing efficiency and accuracy. During training, the image input size was set to 640*640. After model training, we exported the model to the onnx format for subsequent prediction and recognition.

[0097] During deep learning, black and white image samples are used for learning, eliminating interference from color and other factors unrelated to appearance and contour. This improves accuracy while reducing model training costs.

[0098] In one embodiment, when the deviation amount or the bridge appearance profile status is identified, if the abnormality exceeds a set number of times continuously, it is identified as a status abnormality.

[0099] When identifying deviation or bridge contour status, anomalies must occur for a set number of times (e.g., 10 times) before the status is considered abnormal. However, confirmation of normality only requires a single occurrence. That is, if the first nine times are considered abnormal but the tenth time is normal, the status is considered normal, thus reducing the false alarm rate.

[0100] In one embodiment, step S4, determining whether to issue an alarm based on the deviation amount and the bridge appearance and contour status, includes:

[0101] (1) When the number of identified target points is less than the number of targets and the bridge's appearance and contour are normal, the first alarm message is pushed: some or all targets are lost;

[0102] Specifically, when the number of identified target points is less than 5, and is less than 5 for 10 consecutive times, and the appearance and contour of the bridge are normal, the bridge status is considered suspicious. Based on the number of lost targets, the first alarm information is pushed: some or all targets are lost.

[0103] (2) When the number of identified target points is less than the number of targets and the bridge appearance contour state is abnormal, a second alarm message is pushed: target lost and contour recognition abnormal;

[0104] Specifically, when the number of identified target points is less than 5 and is less than 5 for 10 consecutive times, and the bridge appearance contour status is identified as abnormal, and is judged as abnormal for 10 consecutive times, the bridge status is considered to be very suspicious, and the second alarm information is pushed: the target is lost and the contour recognition is abnormal.

[0105] (3) When the number of identified target points is normal but the deviation is greater than the set threshold, and the bridge appearance contour is normal, the third alarm message is pushed: the target is complete but the target deviation is too large;

[0106] Specifically, when the identified target point = 5, according to the pre-set deviation threshold of 0.5, if the deviation is greater than the set threshold (and exceeds it for 10 consecutive times), and the bridge appearance contour is normal, the third alarm information is pushed: the target is complete but the target deviation is too large.

[0107] (4) If the number of identified target points is normal but the deviation is less than the set threshold, and the bridge appearance contour status is abnormal, the data will not be pushed;

[0108] Specifically, when the identified target point = 5, the deviation is less than the set threshold (and is less than this value for 10 consecutive times), and the bridge appearance contour status is identified as abnormal, and is judged as abnormal for 10 consecutive times, then the bridge appearance contour status is considered to be incorrectly identified and no push is performed.

[0109] (5) When the number of identified target points is normal but the deviation is greater than the set threshold, and the bridge appearance contour state is abnormal, the fourth alarm message is pushed: the target is complete but the target deviation is too large, and the contour recognition is abnormal.

[0110] Specifically, when the identified target point = 5, the deviation is greater than the set threshold (and is greater than this threshold for 10 consecutive times), and the bridge appearance contour status is identified as a bridge abnormality, and is judged as abnormal for 10 consecutive identifications, the bridge status is considered to be suspicious, and the fourth alarm information is pushed: the target is complete but the target deviation is too large, and the contour recognition is abnormal.

[0111] Through the above solution, different push information is provided according to different recognition results to ensure the reliability of the alarm information.

[0112] Furthermore, when alarm information is pushed, on-site alarm prompts can be provided. For example, alarm prompt modules can be deployed on bridges to provide multi-dimensional alarm prompts to drivers and passengers. For example, alarm information can be displayed on information boards to warn drivers and passengers of danger ahead and prompt them to stop immediately. For example, alarm information can be broadcast through loud speakers to alert drivers and passengers of danger ahead and prompt them to stop immediately. For example, strobe lights can flash in a conspicuous red at night to warn of danger ahead. Remote alarm prompts can also be provided. For example, through the 5G network, the danger alarm information can be transmitted to the bridge monitoring center to alert the monitoring center of the danger. The monitoring center will then report the danger incident step by step, and the bridge management unit will carry out emergency disaster relief.

[0113] In one embodiment, the bridge disaster identification method based on image patterns further includes:

[0114] Step S5: Identify the bridge deck information image based on the background difference method to determine whether there is foreign object intrusion on the bridge deck; and determine whether to push an alarm based on the state of the foreign object on the bridge deck.

[0115] In one embodiment, step S5, identifying the bridge deck information image based on a background difference method to determine whether there is foreign object intrusion on the bridge deck, includes:

[0116] Step S51: establishing a mixed Gaussian model according to the grayscale value of each pixel in the bridge deck information image;

[0117] Step S52: Calculate the matching degree between each pixel point in a bridge deck information image and each Gaussian distribution in the mixed Gaussian model;

[0118] If it matches, it is considered as background;

[0119] If there is no match, it is considered as foreground.

[0120] Through the above scheme, based on the background difference method, according to the pixel size and residence time of the object, it is possible to detect whether there is foreign object intrusion on the bridge deck, and judge whether there is abnormal parking on the bridge deck, or foreign object intrusion caused by mudslides, landslides, etc. Such foreign objects are set as vehicles cannot be crushed and can only be detoured or stopped to avoid secondary accidents caused by foreign object intrusion.

[0121] Background subtraction is a method for detecting moving foreground objects, offering high accuracy and speed. It fully extracts moving foreground objects by using the difference between the current frame image and a reference background image. Using this method requires continuously updating the background model using online estimation methods to mitigate the effects of lighting changes, chaotic background motion, and other factors. The background subtraction method, based on a mixed Gaussian distribution model, is not only highly adaptable to complex scenes but also allows for adaptive updating of background image parameters to accommodate environmental changes. This method can reliably handle interference from lighting changes, chaotic background motion, and prolonged scene changes.

[0122] The specific process is:

[0123] a. Establish a mixed Gaussian model for the grayscale value of each pixel in the bridge deck information image. The number of mixed Gaussian models is 3. Each Gaussian distribution needs to be represented by three parameters, namely weight W k , mean μ k , standard deviation σ k The value of K ranges from 1 to 3, representing the parameters of the three Gaussian distributions. k Initialize with 1 / 3, μ k Take the first gray value of the pixel, σ k Take 5. The first pixel is regarded as the first Gaussian distribution.

[0124] Read a bridge deck information image at a time (since it is a grayscale image, only one channel is considered). For each pixel, calculate the absolute value of the difference between the mean of each Gaussian at that pixel, and then calculate whether this difference is less than D times the variance of the corresponding Gaussian, where D is defined as 2.5. Adjust the parameters of the matching Gaussian model as follows: (where α is an artificially defined learning rate of 0.001, used to update the weights, and ρ = α / W k , used to update the Gaussian model): such as:

[0125]

[0126] The weights of all Gaussian models are updated as follows:

[0127] W k，t =(1-α)W k，t-1 +αM k，t ;

[0128] If the current pixel matches the corresponding Gaussian model, M k，t =1, otherwise M k，i =0.

[0129] For example, the grayscale value of the pixel in the tth frame is 105, |105-μ 1，t |<2.5σ 1，t, that is, the pixel of the t-th frame belongs to the first Gaussian distribution, so the pixel is considered to be the background, and the mean and standard deviation of the first distribution are updated.

[0130] After the parameters are updated, the weight of the matched Gaussian will become larger, and the weight of the unmatched Gaussian will become smaller. Therefore, after the update is completed, the sum of the weight values is usually not equal to 1, and the weights need to be normalized.

[0131] Calculate the rank value of all Gaussians for each pixel, rank = W k / σ t , meaning that models with larger weights and smaller variances are ranked higher. If all Gaussian models for a pixel are not matched, it means that the model for that pixel is not well established. The Gaussian model with the smallest rank is deleted, and a new Gaussian model matching the current pixel value is established. The pixel grayscale value is set to the mean of the Gaussian model, with a variance of 5, and the weight remains unchanged.

[0132] The first three models are selected as the background. When a new video frame (i.e., the bridge deck information image) comes in, the pixel points are detected to see if they belong to the three Gaussian distributions. If not, they are considered to be foreground, otherwise they are considered to be background.

[0133] When judging whether to push an alarm based on the status of foreign objects on the bridge surface, if a foreign object is identified at the same location 10 times in a row, a circular fitting is performed on the grayscale imaging of the foreign object. If the fitting radius exceeds 1 / 4 of the lane pixel width, the fifth alarm information is pushed: foreign object intrusion alarm.

[0134] The present application also provides an image-based bridge disaster identification system, which uses the above-mentioned image-based bridge disaster identification method. The image-based bridge disaster identification system includes:

[0135] The image acquisition module is configured to: acquire bridge target images and bridge deck information images in real time;

[0136] The first edge computing module is configured to: identify target points on the bridge target image and calculate the deviation of each target point along the bridge direction at different times;

[0137] The second edge computing module is configured to: identify the bridge deck information image using a pre-trained prediction model to determine whether there is any abnormality in the bridge appearance contour;

[0138] The alarm push module is configured to determine whether to push an alarm based on the deviation amount and the appearance and contour status of the bridge.

[0139] In one embodiment, the second edge computing module is further configured to: identify the bridge deck information image based on the background difference method to determine whether there is foreign object intrusion on the bridge deck.

[0140] The calculation frequency of the first edge computing module is 2 times / second; the calculation frequency of the second edge computing module is 1 time / second.

[0141] The image-based bridge disaster identification system provided in the embodiment of the present application has edge computing capabilities, and has integrated perception, transmission, and application. It can effectively meet the needs of early warning and monitoring of structural damage of bridges under disasters such as mountain torrents, mud and rock flows, and landslides. It can provide real-time early warning, has controllable costs, and has low power consumption.

[0142] The functions of image acquisition, the first edge computing module, and the second edge computing module have been explained in the above method and will not be repeated here.

[0143] In the description of this application, it should be noted that the terms "upper" and "lower" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on this application. Unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be internal communication between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to the specific circumstances.

[0144] It should be noted that, in this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises", "includes" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0145] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0146] The above are merely specific embodiments of the present application to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but rather is intended to conform to the widest scope consistent with the principles and novel features of the present application.

Claims

1. A bridge disaster identification method based on image pattern, characterized in that: The following steps are involved: Real-time acquisition of bridge target images and bridge deck information images; Performing target point recognition on the bridge target image, and calculating the deviation of each target point along the bridge direction at different times; The bridge deck information image is recognized by a pre-trained prediction model to determine whether there is any abnormality in the appearance and contour of the bridge; Determine whether to send an alarm based on the deviation and the appearance and contour status of the bridge.

2. The bridge disaster identification method based on image pattern according to claim 1, characterized in that: The bridge disaster identification method based on image patterns also includes: Recognizing the bridge deck information image based on a background difference method to determine whether there is foreign matter intrusion on the bridge deck; Determine whether to send an alarm based on the status of foreign objects on the bridge surface.

3. The bridge disaster identification method based on image pattern according to claim 1, characterized in that: The step of identifying target points on the bridge target image and calculating the deviation of each target point along the bridge direction at different times includes: Converting the bridge target image into a binary image based on a Kmeans algorithm; Performing denoising on the binarized image based on a Gaussian filtering algorithm; Perform contour detection on the denoised image to identify the pixel center coordinates of each target point; The pixel center coordinates of each target point are fitted with a least squares straight line to calculate the deviation of the target point along the bridge direction at different times.

4. The bridge disaster identification method based on image pattern according to claim 1, characterized in that: The pre-training process of the prediction model includes: Obtain a positive sample set with normal bridge appearance and contour status and a negative sample set with abnormal bridge appearance and contour status; Converting the positive sample set and the negative sample set into grayscale images; semantically annotating the grayscale images, and dividing the annotated grayscale images into a training set and a validation set; A prediction model is constructed and trained based on the training set and the validation set.

5. The bridge disaster identification method based on image pattern according to claim 4, characterized in that: When obtaining a negative sample set with abnormal bridge appearance contour status, a negative sample set with different bridge types and different collapse conditions is obtained through simulation.

6. The bridge disaster identification method based on image pattern according to claim 2, characterized in that: The identifying of the bridge deck information image based on the background difference method to determine whether there is foreign matter intrusion on the bridge deck includes: Establishing a mixed Gaussian model according to the grayscale value of each pixel in the bridge deck information image; Calculating the matching degree between each pixel point in a bridge deck information image and each Gaussian distribution in the mixed Gaussian model; If it matches, it is considered as background; If there is no match, it is considered as foreground.

7. The bridge disaster identification method based on image pattern according to claim 1, characterized in that: The determining whether to push an alarm based on the deviation amount and the bridge appearance and contour state includes: When the identified target points are less than the target number and the bridge appearance contour is normal, a first alarm message is pushed: part or all of the targets are lost; When the number of identified target points is less than the target number and the bridge appearance contour state is abnormal, a second alarm message is pushed: target lost and contour recognition abnormal; When the number of identified target points is normal but the deviation is greater than the set threshold, and the bridge appearance is normal, a third alarm message is pushed: the target is complete but the target deviation is too large; If the number of identified target points is normal but the deviation is less than the set threshold, and the bridge appearance contour is abnormal, the data will not be pushed; When the number of identified target points is normal but the deviation is greater than the set threshold, and the bridge appearance contour state is abnormal, a fourth alarm message is pushed: the target is complete but the target deviation is too large, and the contour recognition is abnormal.

8. The bridge disaster identification method based on image pattern according to claim 7, characterized in that: When identifying the deviation or the appearance and outline of the bridge, if the abnormality exceeds the set number of times continuously, it will be identified as an abnormality.

9. A bridge disaster identification system based on image patterns, using the bridge disaster identification method based on image patterns according to any one of claims 1 to 8, characterized in that: The bridge disaster identification system based on image pattern includes: The image acquisition module is configured to: acquire bridge target images and bridge deck information images in real time; The first edge computing module is configured to: identify target points on the bridge target image and calculate the deviation of each target point along the bridge direction at different times; The second edge computing module is configured to: identify the bridge deck information image using a pre-trained prediction model to determine whether there is any abnormality in the appearance and contour of the bridge; The alarm push module is configured to determine whether to push an alarm based on the deviation amount and the appearance and contour status of the bridge.

10. The bridge disaster identification system based on image pattern according to claim 9, characterized in that: The second edge computing module is further configured to: identify the bridge deck information image based on a background difference method to determine whether there is foreign object intrusion on the bridge deck.

Images