Traffic sign real-time monitoring system and method based on artificial intelligence

Abstract

The present invention discloses a real-time monitoring system and method for traffic signs based on artificial intelligence, which relates to the technical field of real-time monitoring of traffic signs, including a detection and acquisition module, a data processing and AI analysis module, an alarm and feedback module, and a remote management module; the intelligent traffic sign monitoring system uses AI technology to achieve real-time recognition and monitoring, but in complex environments, recognition is inaccurate, abnormalities are not handled in a timely manner, and there is a lack of intelligent analysis, which affects traffic safety and smart city construction. Now, the collected data is processed and analyzed through the data processing and AI analysis module of the present invention; problems such as recognition in complex environments are solved, centralized management and analysis of monitoring data are realized, problems can be quickly discovered, and data support is provided for long-term traffic sign optimization and layout adjustment; the ability of real-time monitoring and dynamic adjustment improves the management efficiency of traffic signs and road safety, and provides a strong technical guarantee for smart city construction.

Description

Technical Field

[0001] The present invention relates to the technical field of real-time monitoring of traffic signs, and in particular to a real-time monitoring system and method for traffic signs based on artificial intelligence. Background Art

[0002] With the development of intelligent transportation systems, AI-based real-time traffic sign monitoring systems have become an essential component of traffic management. These systems leverage computer vision technology to capture images of traffic signs on the road using cameras and, combined with deep learning models, perform real-time recognition and analysis. This provides drivers with accurate road information and helps traffic management departments monitor the status of traffic signs, such as damage, obstruction, or aging. This technology offers significant advantages in improving road safety and traffic flow, and is widely used in autonomous vehicles and intelligent monitoring systems.

[0003] The current system still faces some problems in practical application: increased traffic safety risks. In complex environments (such as rain, fog, and at night), existing technologies cannot accurately recognize traffic signs, which may cause drivers to misunderstand road information and increase the risk of traffic accidents. Traffic management efficiency is low. Due to the inability to fully analyze the status and functional rationality of signs, abnormal signs (such as blurred displays or functional failures) are not discovered or handled in a timely manner, affecting the efficiency of traffic management departments and hindering the construction of smart cities. The existing system lacks real-time data support and intelligent analysis, making it difficult to provide a reliable basis for the long-term optimization of traffic signs and urban planning. Summary of the Invention

[0004] The purpose of the present invention is to solve the above-mentioned problems and to propose a real-time monitoring system and method for traffic signs based on artificial intelligence.

[0005] The object of the present invention can be achieved by the following technical solutions: a real-time monitoring system for traffic signs based on artificial intelligence, including a data processing and AI analysis module;

[0006] Perform data standardization on the monitoring data to obtain a standardized image;

[0007] Perform clarity analysis on monitoring data; calculate the image quality clarity of monitoring data after standard processing, and obtain G based on the calculation of image gradient. x is the horizontal gradient, G y is the vertical gradient and gradient amplitude G, and the mean or variance of the gradient amplitude is used as the clarity evaluation index; mean method: through the formula: Output clarity evaluation index - C1; variance method: through the formula: Output clarity evaluation index C2, where M×N is the image size and μ is the mean of the gradient amplitude. Then use the Laplace operator to evaluate clarity. The Laplace operator calculation formula is used to calculate the absolute value mean or variance of the Laplace result. The absolute value mean method is: Output clarity evaluation three C3; variance method: Output clarity evaluation index C4, where L(i, j) is the pixel value of the Laplace result; perform weighted calculation on the calculated clarity evaluation index 1, clarity evaluation index 2, clarity evaluation index 3, and clarity evaluation index 4 to obtain a total clarity evaluation index C0; compare the total clarity evaluation index C0 with a preset clarity index threshold; if the total clarity evaluation index C0 is less than the preset clarity index threshold, perform image processing, and recalculate the total clarity evaluation index; if the recalculated total clarity evaluation index is still less than the preset clarity index threshold, generate a position acquisition signal;

[0008] If the total clarity evaluation index is greater than the preset clarity index threshold, the sign in the image is at a standard position and angle, and irrelevant background is removed to obtain the final cropped image; the final cropped image is subjected to standard detection to generate an alarm instruction.

[0009] As a preferred embodiment of the present invention, it also includes a detection and acquisition module, an alarm and feedback module and a remote management module;

[0010] The detection and acquisition module is used to capture traffic sign images in real time and mark them as monitoring data;

[0011] The alarm and feedback module is used to receive alarm instructions and perform corresponding operations;

[0012] The remote management module is used to receive the position acquisition signal. When the position acquisition signal is received, it obtains the marked position, marks the monitoring data of the traffic sign at this position as abnormal, and then controls other vehicles equipped with high-definition cameras to collect data and perform clarity analysis again.

[0013] As a preferred embodiment of the present invention, the specific process of performing data standardization on the monitoring data to obtain a standardized image is as follows:

[0014] The first step is to denoise the collected image data, convert the grayscale value of the color image into a grayscale image, and smooth the image; then calculate the gradient amplitude and direction of each pixel; use the Sobel operator to perform convolution to output the horizontal gradient and G y is the vertical gradient; by the formula: Output the gradient amplitude G and direction θ; set high and low thresholds Tg and Td. If the gradient amplitude G is greater than the high threshold Tg, it is determined to be a strong edge. If the gradient amplitude G is greater than the low threshold Td but less than the low threshold Tg, the weak edge is retained and the strong edge is connected. If the gradient amplitude G is less than the low threshold Td, the weak noise point is suppressed.

[0015] Step 2: Use Gaussian filtering to remove high-frequency noise and retain low-frequency information. The calculation is based on the Gaussian kernel convolution formula: define the Gaussian kernel matrix, then apply the Gaussian kernel matrix K to the input image I to generate the smoothed image I'(x,y). Then move the Gaussian kernel and repeat the operation to process all pixels of the image.

[0016] Step 3: Use median filtering to replace the pixel value with the median of the pixel values ​​in its neighborhood to smooth the image: determine the neighborhood window of the pixel. For each pixel I(x,y) in the image, select the neighborhood window around it and sort all the pixel values ​​in the window from small to large. Select the middle value from the sorted pixel values ​​as the new pixel value. For 9 pixel values, the middle value is the 5th value. Replace the original pixel value with the center pixel value, move the window to each pixel position in the image, repeat the above process, and finally generate the filtered image.

[0017] As a preferred embodiment of the present invention, the specific process of obtaining the final cropped image is as follows:

[0018] Use the Canny algorithm to extract the edges of the image: E(x,y) is the edge image. The edge points are mapped from the Cartesian coordinate system to the polar coordinate system: ρ = x cosε + y sinε, where ρ is the distance from the line to the image origin and ε is the tilt angle of the line. The tilt angle ε of the line is extracted based on the Hough transform result, and the rotation angle α of the sign relative to the image center is calculated: α = 90° - |ε|.

[0019] Perform an affine transformation on the image to adjust the tilted sign to a horizontal or vertical state. Generate the rotation affine matrix W through the transformation moment, which is: tx and ty are the offsets of the image center; use affine transformation to rotate the image: J' = J·W, where J' is the corrected image; if the sign is circular, apply minimum circumscribed circle detection to adjust the center and radius of the rotated circle; if the sign is triangular, use vertex detection to ensure that the vertex positions match the boundaries; combine with the object detection model to lock the sign area: input the data into the detection model for processing, and output the predicted box: (x min ,y min ,x max ,y max ); extract the cropping area, by the formula: Jc=J[y min :ymax , x min : x max is obtained; then use the edge detection algorithm to re-extract the edges of the signboard, calculate the convex hull CH of the edge points, and output the convex hull CH of the edge points by the formula: CH = ConvexHull(P), P = {(x, y)|E(x, y) = 1}; generate a rectangular crop B according to the boundary of the convex hull CH, and find the minimum bounding rectangle of the convex hull: (x min , y min , x max , y max ) = BoundingBox(CH), (x min , y min) and (x max , y max ) are the upper left and lower right coordinates of the rectangular frame; crop the original image with the optimized crop frame (x min , y min , x max , y max ): Jr = J[y min : y max , x min : x max [[ID=3⑦]]] to output the finally cropped image Jr.

[0020] As a preferred embodiment of the present invention, the specific process of performing standard detection on the finally cropped image to generate an alarm instruction is as follows:

[0021] Calculate the matching degree P for the integrity of the cropped image. If P < Tp, where Tp is a preset template matching threshold, trigger an alarm and generate an alarm instruction 1; calculate the overall evaluation clarity Cz for the clarity of the image content of the cropped image, and compare the overall evaluation clarity Cz with a preset threshold Tc. If Cz is less than the threshold Tc, issue an alarm instruction 2; then calculate the functional relevance fitness Ag of the sign by combining the actual environment and the sign content, and set a threshold Tf according to traffic management requirements. If Ag ≥ 0.8, the fitness is reasonable; if Ag < 0.8: the fitness is insufficient and there is a problem, and an alarm instruction 3 is generated.

[0022] As a preferred embodiment of the present invention, the specific process of calculating the matching degree P for the integrity of the cropped image is as follows:

[0023] Use template matching technology to match the cropped image Jr with the standard template Tr of the signboard; through the formula: Output the image matching degree P.

[0024] As a preferred embodiment of the present invention, the specific process of calculating the image content clarity of the cropped image to obtain the overall evaluation clarity Cz is as follows:

[0025] Use the Laplace operator to calculate the high-frequency component of the image, using the formula: Output a new matrix H(x,y), Represents the second-order derivative of the grayscale value in the x direction, Represents the second-order derivative of the grayscale value in the y direction; the overall clarity of the entire image is obtained, and the absolute value mean of the matrix H(x,y) is calculated: by the formula: Output the overall evaluation clarity Cz, where X and Y are the total number of detection areas on the X and Y axes respectively.

[0026] As a preferred embodiment of the present invention, the specific process of calculating the functional relevance and functional adaptability Ag of the marker is as follows:

[0027] The correlation weight Fx is output through the formula: Fx = Fd × β1 + Ff × β2 + Fp × β3, where Fd is the matching degree between the sign content and the road design speed, Ff is the matching degree between the sign content and the traffic volume, and Fp is the matching degree between the sign content and the accident rate; β1, β2, and β3 are the weights of each factor; the weights of all expected functions of the sign category are then accumulated. For speed limit signs, the following functional weights are included: controlling speed exceeding behavior; improving driving safety; and maintaining traffic flow stability; Where K is the total number of functions; substitute Fx and Ft into the formula: Output the functional suitability Ag of the sign.

[0028] A traffic sign real-time monitoring system based on artificial intelligence, the method comprising:

[0029] Step 1: Install a high-definition camera at a traffic intersection or on a mobile carrier to capture images of traffic signs in real time;

[0030] Step 2: The collected images are subjected to image denoising, image correction and extraction, image matching and evaluation, and functional fitness calculation;

[0031] Step 3: Classify and execute the alarm instructions;

[0032] Step 4: Receive the abnormal location signal and mark the traffic sign data at this location as abnormal; dispatch other devices equipped with cameras to collect data at the abnormal location and re-analyze and confirm the clarity.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. The present invention has multiple advantages. It integrates high-definition cameras and AI analysis modules to achieve efficient monitoring and accurate analysis of traffic signs. Through steps such as image denoising, clarity assessment, edge enhancement, and functional adaptability analysis, the system can comprehensively evaluate the shape, display status, and functional rationality of the signs. In particular, through the combination of multiple algorithms, such as edge detection, Gaussian filtering, and Laplace operators, the system can effectively remove noise, enhance image details, and make the detection and recognition of signs more accurate. For abnormal situations, the system can quickly generate multi-dimensional alarm instructions including shape abnormalities, display blur, functional incompatibility, etc., providing all-round intelligent support for traffic management. In addition, the use of vehicles as mobile platforms for monitoring equipment greatly improves the monitoring range and efficiency.

[0035] 2. At the same time, the present invention further demonstrates the advanced nature of intelligent transportation. When an abnormal location is marked, the system can immediately synchronize the relevant information to the management center, guiding other vehicles equipped with high-definition cameras to conduct re-inspections, thereby ensuring the accuracy and reliability of the data. Through the centralized management and analysis of monitoring data, not only can problems be quickly discovered, but data support can also be provided for long-term traffic sign optimization and layout adjustment. This real-time monitoring and dynamic adjustment capability significantly improves the management efficiency of traffic signs and road safety, provides a strong technical guarantee for the construction of smart cities, and effectively reduces the cost of manual inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0037] Figure 1 It is a principle block diagram of the present invention;

[0038] Figure 2 A diagram showing the steps of the method of the present invention;

[0039] Figure 3 This is a schematic diagram of the detection scenario of the present invention Figure 1 ;

[0040] Figure 4 This is a schematic diagram of the detection scenario of the present invention Figure 2 . DETAILED DESCRIPTION

[0041] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] It should be understood that the terms “include” and “comprising” used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0043] It should also be understood that the terminology used in this disclosure is for the purpose of describing specific embodiments only and is not intended to limit the disclosure. As used in this disclosure and the claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should be further understood that the term "and / or" as used in this disclosure and the claims refers to any and all possible combinations of one or more of the associated listed items, including and including these combinations.

[0044] See also Figure 1 、 3 As shown in Figure 4, a real-time monitoring system for traffic signs based on artificial intelligence includes a detection and acquisition module, a data processing and AI analysis module, an alarm and feedback module, and a remote management module.

[0045] The measurement and acquisition module captures traffic sign images in real time by installing high-definition cameras at traffic intersections or on vehicles; for example, by cooperating with taxi or bus companies, taxis or buses are equipped with corresponding detection and acquisition equipment to collect monitoring data; the monitoring data includes still image data and dynamic image data.

[0046] The data processing and AI analysis module is used to analyze monitoring data, specifically:

[0047] Standardize the collected monitoring data;

[0048] Step 1: De-noise the collected image data, use edge detection algorithm to separate noise and obtain the main structure of the sign for processing, calculate the grayscale value of the color image and convert it into a grayscale image, then use Gaussian filtering to denoise and smooth the image; then calculate the gradient amplitude and direction of each pixel; use Sobel operator for convolution: through the formula: Among them G x is the horizontal gradient, G y is the vertical gradient, I is the input image matrix, that is, the grayscale value of each pixel in the image, and the value of each pixel I(x,y) represents the grayscale value of the image at position (x,y) (if it is a grayscale image) or the color component value (if it is a single channel of a color image); through the formula: Output the gradient magnitude G and direction θ; suppress non-edge pixels along the gradient direction and retain significant edge points; then set the high and low thresholds Tg and Td. If the gradient magnitude G is greater than Tg, it is determined to be a strong edge. If the gradient magnitude G is greater than Td but less than Tg, the weak edge is retained and the strong edge is connected; if the gradient magnitude G is less than Td, the weak noise points are suppressed.

[0049] Step 2: Use Gaussian filtering to remove high-frequency noise while retaining low-frequency information; calculate using the Gaussian kernel convolution formula: define the Gaussian kernel matrix K (taking a 3×3 kernel as an example): Then apply the Gaussian kernel matrix K to the input image I to generate the smoothed image I'(x,y), specifically: through the formula: Output the smoothed image I'(x,y), then move the Gaussian kernel, and repeat the operation to process all pixels of the image. It should be noted that in actual applications, the size and standard deviation σ of the Gaussian kernel can be adjusted according to the intensity of image noise. When there is a lot of noise, a larger Gaussian kernel (such as 5×5 or 7×7) is used. The standard deviation σ adjusts the width of the Gaussian distribution. A larger σ will result in a stronger blurring effect.

[0050] Step 3: Use median filtering to smooth the image by replacing the pixel value with the median of the pixel values ​​in its neighborhood. Specifically:

[0051] Determine the neighborhood window of the pixel. For each pixel I(x,y) in the image, select a neighborhood window around it (commonly 3×3, but also 5× or 7×7), and then sort all the pixel values ​​in the window in ascending order; for example: neighborhood pixel value = {10,20,30,40,50,60,70,80,90}, after sorting: sorting result = {10,20,30,40,50,60,70,80,90}, select the middle value from the sorted pixel values ​​as the new pixel value, for 9 pixel values ​​(odd number), the middle value is the 5th value: I'1(x,y) = 50, and then replace the original pixel value with the center pixel value, for example: the original pixel value I(x,y) = 30, I'1(x,y) = 50; move the window to each pixel position in the image, repeat the above process, and finally generate the filtered image.

[0052] The collected monitoring data is analyzed for clarity, specifically:

[0053] The image quality clarity is calculated for the monitoring data level after standard processing, and G is obtained by calculating the image gradient. x is the horizontal gradient, G y is the vertical gradient and G, and on this basis, the mean or variance of the gradient amplitude is used as the clarity evaluation index: Mean method: through the formula: Output clarity evaluation index - C1; variance method: through the formula: Output clarity evaluation index 2 C2, where M×N is the image size and μ is the mean value of the gradient amplitude;

[0054] Then use the Laplace operator to evaluate the clarity (based on the second-order derivative of the image, the Laplace operator can highlight the edge information in the image), and use the Laplace operator calculation formula: For the input image; the Laplacian operator is a high-pass filter that focuses on edges and rapidly changing areas; the absolute mean or variance of the Laplacian results is statistically calculated: the absolute mean method: Output clarity evaluation three C3; variance method: Output clarity evaluation four C4, where L(i,j) is the pixel value of the Laplace result;

[0055] The clarity evaluation index 1 C1, clarity evaluation index 2 C2, clarity evaluation index 3 C3 and clarity evaluation index 4 C4 are weighted and calculated using the formula: Output the total clarity evaluation index C0, where a1, a2, a3 and a4 are the preset weights corresponding to the clarity evaluation index; compare the total clarity evaluation index C0 with the preset clarity index threshold. If the total clarity evaluation index C0 is less than the preset clarity index threshold, perform linear brightness transformation on the image to increase the grayscale value to a more reasonable range; contrast enhancement: use histogram equalization or contrast-limited adaptive histogram equalization (CLAHE); edge enhancement: use a high-pass filter or a sharpening filter through convolution operation to highlight image details, noise reduction processing: if noise affects clarity, use Gaussian filtering or median filtering for noise reduction, and after image enhancement methods, recalculate the total clarity evaluation index. If the recalculated total clarity evaluation index is still less than the preset clarity index threshold, obtain the current position information, mark it, and generate a position acquisition signal.

[0056] If the total clarity evaluation index is greater than the preset clarity index threshold, the sign in the image is at the standard position and angle, and irrelevant background is removed, specifically:

[0057] Extract the edges of the image by using the Canny algorithm: Here, E(x, y) is the edge image. The edge points are then mapped from the Cartesian coordinate system to the polar coordinate system: ρ = x cosε + y sinε, where ρ is the distance from the line to the image origin and ε is the tilt angle of the line. The ε of the main line is extracted based on the Hough transform results, and the rotation angle α of the sign relative to the image center is calculated: α = 90° - |ε|.

[0058] Perform an affine transformation on the image to adjust the tilted sign to a horizontal or vertical state. Generate the rotation affine matrix W through the transformation moment, which is: Where tx and ty are the offsets of the image center (keeping the center unchanged). The image is then rotated using an affine transformation: J' = J·W, where J' is the corrected image. If the sign is circular, a minimum circumscribed circle test is applied to adjust the center and radius of the rotated circle. If the sign is triangular, vertex detection (such as Harris corner detection) is used to ensure that the vertex positions match the boundary.

[0059] Combined with the target detection model (such as YOLO or SSD), lock the sign area: input the data into the detection model for processing, and output the predicted box: (x min ,y min ,x max ,y max ), represents the minimum bounding box of the sign; extract the cropping area, through the formula: Jc=J[y min :y max ,x min :x max ]; then use the edge detection algorithm (Canny) to re-extract the edge of the sign, calculate the convex hull CH of the edge points, and output the convex hull CH of the edge points through the formula: CH = ConvexH ull(P), P = {(x, y) | E(x, y) = 1}; generate a rectangular crop B based on the boundary of the convex hull CH, that is, find the minimum circumscribed rectangle of the convex hull: (x min ,y min ,x max ,y max )=BoundingBox(CH),(x min ,y min) and (x max ,y max) The coordinates of the upper left corner and lower right corner of the rectangular frame; the optimized cropping frame (x min ,y min ,x max ,y max ), crop the original image: Jr=J[y min :y max ,x min :x max ] Output the final cropped image Jr.

[0060] The final cropped image is then subjected to standard testing, specifically:

[0061] The integrity of the cropped image is calculated to obtain the matching degree P. The template matching technology is used to match the cropped image Jr with the standard template Tr of the signboard; according to the formula: Output the cosine similarity P (matching degree P) of the output image, ranging from [0, 1]. The closer M is to 1, the more similar the cropped image is to the template. If P < Tp (preset template matching threshold), it means the cropped image does not meet the standard, trigger an alarm, and generate alarm instruction one.

[0062] Calculate the clarity of the content of the cropped image. Specifically: Use the Laplacian operator to calculate the high-frequency components of the image through the formula: Output a new matrix H(x, y), whose value represents the intensity of the gray-scale change of each pixel point in the image (in an area with剧烈变化 (such as edges), the Laplacian value will be large; if the image is blurred and the gray-scale change is gentle, the Laplacian value is small). Represents the second-order derivative of the gray-scale value in the x direction. Represents the second-order derivative of the gray-scale value in the y direction. Then, obtain the overall clarity of the entire image and calculate the absolute value mean of the matrix H(x, y) through the formula: Output the overall evaluation clarity Cz. X and Y are the total numbers of the X-axis and Y-axis of the detection area respectively. The higher Cz is, the clearer the image is and the richer the edge information is. Compare the overall evaluation clarity Cz with the preset threshold Tc. If Cz is less than the threshold Tc, it means the image is blurred and the edges and details are insufficient, and alarm instruction two is issued; on the contrary, if Cz is higher than the threshold Tc, it means the image clarity meets the requirements.

[0063] Perform weighted calculation on the cosine similarity P of the image matching and the overall evaluation clarity Cz: Through the formula: TXZ = P×P' + Cz×Cz' to output the image value TXZ, where P' and Cz' are the respective preset weight factors.

[0064] Then, by combining the actual environment and the sign content, calculate the functional relevance functional adaptability of the sign. For example, for a speed limit sign: Through the formula: Fx = Fd×β1 + Ff×β2 + Fp×β3 to output the relevance weight Fx. Fd is the matching degree value between the sign content and the road design speed, Ff is the matching degree value between the sign content and the traffic flow, Fp is the matching degree value between the sign content and the accident rate (obtained directly from the database); β1, β2, and β3 are the weights of each factor (set by traffic management experts); then accumulate the weights of all expected functions of the sign category. For example, for a speed limit sign, it includes the following function weights: Control speed over-limit behavior; Improve driving safety; Keep traffic flow smoothness; Where K is the total number of functions; Substitute Fx and Ft into the formula: Output the functional adaptability Ag of the signboard. Then, set the threshold Tf according to traffic management requirements. If Ag ≥ 0.8, the adaptability is reasonable; if Ag < 0.8: The adaptability is insufficient and there are problems, and alarm instruction three is generated.

[0065] Then, the functional adaptability Ag of the signboard and the image value TXZ are weighted to obtain the final image value ZTZ, ZTZ = TXZ × B1 + Ag × B2, where B1 and B2 are preset weights; the final image value ZTZ is compared with the corresponding preset threshold value. If the final image value ZTZ is less than the corresponding preset threshold value, an alarm instruction four is generated;

[0066] The alarm and feedback module is used to receive alarm instructions, which include alarm instruction one, alarm instruction two, alarm instruction three and alarm instruction four. When alarm instruction one is received, the current position information is obtained, and the text corresponding to the abnormal shape of the traffic sign is generated for display; when alarm instruction two is received, the current position information is obtained, and the text corresponding to the abnormal display of the traffic sign is generated for display; when alarm instruction three is received, the current position information is obtained, and the text corresponding to the abnormal function of the traffic sign is generated for display; when alarm instruction four is received, the current position information is obtained, and the traffic sign shape function, display, and abnormal text are generated for display.

[0067] The remote management module is used to receive the location acquisition signal. When the location acquisition signal is received, it obtains the marked location, marks the monitoring data of the traffic sign at this location as abnormal, and then controls other vehicles equipped with high-definition cameras to collect data and perform clarity analysis again.

[0068] See also Figure 2 As shown, another aspect of the present invention provides a method for real-time monitoring of traffic signs based on artificial intelligence, the specific steps of which are:

[0069] Step 1: Install a high-definition camera at a traffic intersection or on a mobile vehicle (such as a taxi or bus) to capture images of traffic signs in real time;

[0070] Step 2: The collected images are subjected to image denoising, image correction and extraction, image matching and evaluation, and functional fitness calculation;

[0071] Step 3: Classify and execute the alarm instructions;

[0072] Step 4: Receive the abnormal location signal and mark the traffic sign data at this location as abnormal; dispatch other devices equipped with cameras to collect data at the abnormal location and re-analyze and confirm the clarity.

[0073] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A real-time traffic sign monitoring system based on artificial intelligence, including data processing and AI analysis modules; characterized by: The data processing and AI analysis module is used to analyze the monitoring data, specifically: Perform data standardization on the monitoring data to obtain a standardized image; Perform clarity analysis on monitoring data; calculate the image quality clarity of monitoring data after standard processing, and obtain G based on the calculation of image gradient. x is the horizontal gradient, G y is the vertical gradient and gradient amplitude G, and the mean or variance of the gradient amplitude is used as the clarity evaluation index; mean method: through the formula: Output clarity evaluation index - C1; variance method: through the formula: Output clarity evaluation index C2, where M×N is the image size and μ is the mean of the gradient amplitude. Then use the Laplace operator to evaluate clarity. The Laplace operator calculation formula is used to calculate the absolute value mean or variance of the Laplace result. The absolute value mean method is: Output clarity evaluation three C3; variance method: Output clarity evaluation index C4, where L(i, j) is the pixel value of the Laplace result; perform weighted calculation on the calculated clarity evaluation index 1, clarity evaluation index 2, clarity evaluation index 3, and clarity evaluation index 4 to obtain a total clarity evaluation index C0; compare the total clarity evaluation index C0 with a preset clarity index threshold; if the total clarity evaluation index C0 is less than the preset clarity index threshold, perform image processing, and recalculate the total clarity evaluation index; if the recalculated total clarity evaluation index is still less than the preset clarity index threshold, generate a position acquisition signal; If the total clarity evaluation index is greater than the preset clarity index threshold, for the signboard in the standard position and angle in the image, remove the irrelevant background to obtain the finally cropped image; calculate the integrity of the cropped image to obtain the matching degree P. If P < Tp (Tp is the preset template matching threshold), trigger an alarm and generate alarm instruction 1; calculate the overall evaluation clarity Cz of the cropped image. Compare the overall evaluation clarity Cz with the preset threshold Tc. If Cz is less than the threshold Tc, issue alarm instruction 2; then, by combining the actual environment and the sign content, calculate the functional relevance and functional adaptability Ag of the sign: Through the formula: Output the relevance weight Fx. Fd is the matching degree value between the sign content and the road design speed, Ff is the matching degree value between the sign content and the traffic flow, and Fp is the matching degree value between the sign content and the accident rate; β1, β2, and β3 are the weights of each factor; then accumulate the weights of all expected functions of the sign category. For speed limit signs, the following function weights are included: controlling speed over-limit behavior; improving driving safety; maintaining smooth traffic flow; , where K is the total number of functions; substitute Fx and Ft into the formula: Output the functional adaptability Ag of the signboard. Then, set the threshold Tf according to traffic management requirements. If Ag ≥ 0.8, the adaptability is reasonable; if Ag < 0.8: the adaptability is insufficient, there are problems, and generate alarm instruction 3.

2. The traffic sign real-time monitoring system based on artificial intelligence according to claim 1 is characterized in that: It also includes detection and acquisition module, alarm and feedback module and remote management module; The detection and acquisition module is used to capture traffic sign images in real time and mark them as monitoring data; The alarm and feedback module is used to receive alarm instructions and perform corresponding operations; The remote management module is used to receive the position acquisition signal. When the position acquisition signal is received, it obtains the marked position, marks the monitoring data of the traffic sign at this position as abnormal, and then controls other vehicles equipped with high-definition cameras to collect data and perform clarity analysis again.

3. The traffic sign real-time monitoring system based on artificial intelligence according to claim 1 is characterized in that: The specific process of performing data standardization on the monitoring data to obtain the standardized image is as follows: The first step is to denoise the collected image data, convert the grayscale value of the color image into a grayscale image, and smooth the image; then calculate the gradient amplitude and direction of each pixel; use the Sobel operator to perform convolution to output the horizontal gradient and G y is the vertical gradient; by the formula: Output the gradient amplitude G and direction θ; set high and low thresholds Tg and Td. If the gradient amplitude G is greater than the high threshold Tg, it is determined to be a strong edge. If the gradient amplitude G is greater than the low threshold Td but less than the low threshold Tg, the weak edge is retained and the strong edge is connected. If the gradient amplitude G is less than the low threshold Td, the weak noise point is suppressed. Step 2: Use Gaussian filtering to remove high-frequency noise and retain low-frequency information; calculate using the Gaussian kernel convolution formula: define the Gaussian kernel matrix, and then apply the Gaussian kernel matrix K to the input image I to generate a smoothed image , then move the Gaussian kernel and repeat the operation to process all pixels of the image; Step 3: Use median filtering to replace the pixel value with the median of the pixel values ​​in its neighborhood to smooth the image: determine the neighborhood window of the pixel. For each pixel I(x,y) in the image, select the neighborhood window around it and sort all the pixel values ​​in the window from small to large. Select the middle value from the sorted pixel values ​​as the new pixel value. For 9 pixel values, the middle value is the 5th value. Replace the original pixel value with the center pixel value, move the window to each pixel position in the image, repeat the above process, and finally generate the filtered image.

4. The traffic sign real-time monitoring system based on artificial intelligence according to claim 1 is characterized in that: The specific process of obtaining the final cropped image is as follows: Use the Canny algorithm to extract the edges of the image: , E(x,y) is the edge image, which maps the edge points from the Cartesian coordinate system to the polar coordinate system: , ρ is the distance from the line to the image origin, and ε is the inclination angle of the line. The inclination angle ε of the line is extracted based on the Hough transform result, and the rotation angle α of the sign relative to the image center is calculated: ; Perform an affine transformation on the image to adjust the tilted sign to a horizontal or vertical state. Generate the rotation affine matrix W through the transformation moment, which is: , tx and ty are the offsets of the image center; use affine transformation to rotate the image: , is the corrected image; if the sign is circular, apply minimum circumscribed circle detection to adjust the center and radius of the rotated circle; if the sign is triangular, use vertex detection to ensure that the vertex positions match the boundary; Combined with the target detection model, lock the sign area: input the data into the detection model for processing, and output the predicted box: (x min ,y min ,x max ,y max ); extract the cropping area by the formula: Get; then use the edge detection algorithm to re-extract the edge of the signboard and calculate the convex hull CH of the edge points, according to the formula: Output the convex hull CH of the edge point; generate a rectangular clipping B based on the boundary of the convex hull CH and find the minimum circumscribed rectangle of the convex hull: (x min ,y min ,x max ,y max )=BoundingBox(CH),(x min ,y min） and (x max ,y max ) are the coordinates of the upper left corner and lower right corner of the rectangular frame; the optimized cropping frame (x min ,y min ,x max ,y max ), crop the original image: Output the final cropped image Jr.

5. The traffic sign real-time monitoring system based on artificial intelligence according to claim 1 is characterized in that: The specific process of performing integrity calculation on the cropped image to obtain the matching degree P is as follows: Use template matching technology to match the cropped image Jr with the standard template Tr of the sign; by formula: Output image matching degree P.

6. The traffic sign real-time monitoring system based on artificial intelligence according to claim 5 is characterized in that: The specific process of calculating the image content clarity of the cropped image to obtain the overall evaluation clarity Cz is as follows: Use the Laplace operator to calculate the high-frequency component of the image, using the formula: Output a new matrix H(x,y), which represents the second-order derivative of the gray value in the x direction. Represents the second-order derivative of the grayscale value in the y direction; the overall clarity of the entire image will be obtained, and the absolute value mean of the matrix H (x, y) is calculated: by the formula: Output the overall evaluation clarity Cz, where X and Y are the total number of detection areas on the X and Y axes respectively.

7. A real-time monitoring method for traffic signs based on artificial intelligence, characterized in that: Applied to implement any one of claims 1-6 of the artificial intelligence-based real-time traffic sign monitoring system, the method comprises: Step 1: Install a high-definition camera at a traffic intersection or on a mobile carrier to capture images of traffic signs in real time; Step 2: The collected images are subjected to image denoising, image correction and extraction, image matching and evaluation, and functional fitness calculation; Step 3: Classify and execute the alarm instructions; Step 4: Receive the abnormal location signal and mark the traffic sign data at this location as abnormal; dispatch other devices equipped with cameras to collect data at the abnormal location and re-analyze and confirm the clarity.

Images