Vehicle height limit early warning system and device based on intelligent video recognition technology

Abstract

The invention provides a vehicle height limit early warning system and device based on an intelligent video recognition technology, and relates to the technical field of intelligent traffic. The data acquisition module can automatically adjust camera parameters according to ambient light intensity to ensure video image quality; and the image enhancement module improves the contrast and definition of the video image through various algorithms. The vehicle height identification module is combined with a CNN model and a Transform model to accurately identify the type and height of a vehicle. The license plate recognition module efficiently recognizes a license plate number by using an OCR technology and deep learning, and judges whether a vehicle has a special passing permission or not. The detour guiding module generates an optimal detour path based on the real-time road condition, and transmits the path information to the driver through the V2X communication technology. In addition, the system also has a function of automatically recording violation behaviors, and can take corresponding measures for vehicles with violation for many times.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a vehicle height limit warning system and device based on intelligent video recognition technology. Background Art

[0002] With the development of economy, infrastructure construction such as roads, bridges, tunnels, overpasses, etc. is also developing rapidly. While accelerating infrastructure construction, it is also necessary to protect these infrastructure. At present, the means of protecting these infrastructure is usually to install height limit signs on the infrastructure to warn drivers, but this warning method cannot remind drivers in real time. Drivers need to maintain high attention to avoid collisions. Low safety and poor vigilance lead to height limit traffic accidents, causing huge losses to infrastructure and drivers' lives and property.

[0003] The document with the prior art publication number CN116447985A provides a vehicle height detection system for a height-limited scene, wherein the height measurement module is used to take a picture of a vehicle to be inspected, and the analysis module is used to mark the vehicle to be inspected according to the picture to be inspected and determine whether the vehicle to be inspected is super-high; the distance measurement module is used to take a super-high picture of a vehicle to be inspected according to the mark, and the analysis module is also used to predict the collision time and collision distance of the super-high vehicle to the height-limited scene according to the picture to be inspected and the super-high picture; the warning module is used to issue corresponding warning information according to the collision time and collision distance. The present invention improves the accuracy and practicality of height-limited detection and reduces the error of height-limited alarm information.

[0004] Although the existing technology has improved the accuracy and practicality of height limit detection to a certain extent, there are still some limitations. For example, the existing technology may not be able to effectively handle the problem of vehicle identification in complex environments, such as accurate identification at night, in rainy and snowy weather, or under strong light conditions. In addition, the real-time performance, intelligence level, and interactivity of the vehicle height limit warning system with the driver need to be further improved.

[0005] In view of this, we proposed a vehicle height limit warning system and device based on intelligent video recognition technology. Summary of the invention

[0006] The present invention provides a vehicle height limit warning system and device based on intelligent video recognition technology, aiming to improve road traffic safety and efficiency. The technical solution includes the following aspects.

[0007] According to one aspect of an embodiment of the present application, a vehicle height limit warning system based on intelligent video recognition technology is provided, and the system includes the following modules.

[0008] The data acquisition module uses light detection sensors to sense the ambient light intensity in real time, automatically adjusts the camera's capture parameters for different lighting environments, and obtains vehicle video data on connected roads.

[0009] The image enhancement module extracts video images from the collected vehicle video data, performs adaptive video image enhancement processing, and improves the video image quality.

[0010] The vehicle height recognition module is based on the CNN and Transformer architecture to identify the type, height, and roof attachments of vehicles approaching the height-restricted area.

[0011] The license plate recognition module uses OCR and deep learning technology to identify the vehicle's license plate number and compare it with the backend database to determine the vehicle's special access rights and whether it is an overheight vehicle.

[0012] The detour guidance module identifies over-height vehicles and uses navigation data or V2X communication technology to generate a detour route and send it directly to the driver or the on-board system to help bypass the height-restricted area.

[0013] The automatic violation recording module records the license plate information of overheight vehicles and compares it with the vehicle records in the database. For vehicles that have violated the rules multiple times, the system automatically generates penalty notices or traffic restriction instructions and handles them through relevant departments.

[0014] The multi-channel warning module accurately delivers warning information to the driver through V2X communication, LED display screens, and on-board terminals. It also provides detour suggestions for over-high vehicles through display screens and navigation systems.

[0015] As an optional solution of the present application, the data acquisition module includes a light detection sensor, an environmental perception sensor, a camera and a historical traffic flow collection device. The light detection sensor is used to monitor the ambient light intensity in real time. Accurately perceive the current lighting environment to ensure the quality of video image capture by the camera under different lighting conditions. The ambient light is measured by the light detection sensor to obtain the light intensity of the current scene, which is fed back to the data acquisition module for automatically adjusting the camera ISO, exposure and shutter speed to cope with different lighting environments. The camera is installed on both sides of the height limit area to capture the video data of vehicles entering the height limit area. The camera supports high-definition resolution, wide-angle field of view and night vision to ensure that every vehicle passing through the height limit area can be fully recorded. After adjusting the parameters through the light detection sensor, the camera can capture the video data of vehicles approaching the height limit area. The vehicle video data contains detailed information on the appearance and outline of the vehicle. The data acquisition module not only relies on physical equipment to collect data, but also obtains information from other data sources. External data, together with local real-time data, provides comprehensive information support for the height limit warning system.

[0016] As an optional solution of the present application, in the video, when the vehicle passes through the height-restricted area, the image enhancement module extracts the most representative video image by the optical flow method to ensure the recognition effect. The optical flow method calculates the video image by comparing the motion change between adjacent frames. The appropriate video image is enhanced, and the contrast of the video image is dynamically adjusted using the ACE algorithm under the condition of ambient light changes. In a low-light environment, Gamma correction is used to adjust the brightness and contrast of the video image. Gamma correction can linearly adjust the grayscale value of the video image to make the darker area clearer. The CLAHE algorithm is used to further enhance the contrast and details of the video image. By dividing the video image into multiple small blocks and performing histogram equalization processing respectively, the loss of details caused by overall over-enhancement is prevented, which is suitable for video image enhancement in complex lighting environments. In foggy weather or when the visual noise is large, a defogging algorithm, such as a defogging algorithm based on a dark channel prior, is used to enhance the clarity of the video image. By assuming that the dark channel in the non-sky area of ​​the video image has a lower brightness, the haze effect is effectively removed. SRGAN is used to improve the resolution of low-resolution video images. SRGAN generates high-resolution video images from low-resolution video images through a generative adversarial network architecture, enhancing the detail performance of vehicle video images. The enhanced video images are passed to the vehicle height recognition module and the license plate recognition module for subsequent vehicle type, license plate and height analysis. The enhancement effect is evaluated based on the PSNR peak signal-to-noise ratio and SSIM structural similarity. The quality of the enhanced video images is evaluated by PSNR and SSIM to ensure that the video images have high detail retention and structural similarity after enhancement.

[0017] As an optional solution of the present application, the vehicle height recognition module uses a deep learning model based on the convolutional neural network CNN and Transformer architecture to accurately identify the height and type of the vehicle, extract the structural features of the vehicle through the video image, and calculate its height. CNN is used to extract local features in video images. Through the stacking of convolutional layers and pooling layers, CNN can learn the edge, shape and spatial structure information of the vehicle. Transformer is mainly used to capture long-distance dependencies and global features in video images to ensure that even when the perspective changes and the vehicle angle is diverse, the model can still accurately identify the height and type of the vehicle. Specifically, the enhanced video image is first normalized and resized to ensure that the size of the video image input to CNN is consistent. The input video image is I(x,y). The video image is first normalized to the range of [0,1] and adjusted to a fixed size, I′(x,y)=[I(x,y)-min(I)] / [max(I)-min(I)]. The normalized video image I′(x,y) can accelerate the convergence of the model and ensure that the vehicle video images of different sizes are consistently input into CNN. Through multiple convolutional layers, the low-level features of the vehicle are extracted. After multiple convolutional layers are superimposed, CNN can gradually extract the local features of the vehicle. After the convolutional layer, the video image size is reduced through the pooling layer and important features are retained. Maximum pooling is used to retain important edge features in the video image while reducing the amount of calculation.

[0018] After being processed, the local features extracted by the convolutional layer enter the Transformer model. The Transformer processes the global contextual relationship in the video image through the self-attention mechanism to help identify the overall shape and height features of the vehicle. The multi-head self-attention mechanism divides the video image features into multiple attention heads, and processes different parts of the video image through different heads, thereby capturing the global information of the vehicle height. The input is the feature vector X, and the self-attention output of each head is, head i =Attention(Q i ,K i ,V i ), where Q i =W i Q X,K i =W i K X,V i =W i V X, W i Q ,W i K ,W i Vis the weight matrix of different heads. Through the multi-head attention mechanism, multiple video image regions are processed in parallel; adding position encoding helps understand the position information of pixels in the video image, and position encoding helps the Transformer model retain the spatial position information of the video image during feature processing.

[0019] The pixel height H of the vehicle is obtained by using the local features extracted by CNN and the global features processed by Transformer. pred The vehicle type is classified with the number of categories C and the output probability distribution P(y=c). The mean square error loss function is used for vehicle pixel height estimation, and the cross entropy loss is used for type classification. The total loss function L is the weighted sum of the two, L=α×MSE(H true ,H pred )+β×∑ c=1 C y c logP(y=c), where α and β are weight factors for height estimation and category classification, H true is the true height, H pred is the predicted height, y c is the indicator variable of the true category. A large-scale labeled dataset is used for model training, including video image data in different environments, angles, and vehicle types, to ensure that the model is robust to diverse scenarios. Based on the predicted vehicle pixel height, combined with reference objects around the vehicle, such as height limit poles, the actual height of the vehicle is inferred. The precise height, H, is obtained through the proportional relationship in the video image and the pre-set scale model. vehicle =H image S real / S pixel , where H vehicle is the actual height of the vehicle, H image is the pixel height of the vehicle in the video image, S pixel is the pixel value of the reference scale in the video image, S real is the actual height of the reference ruler.

[0020] As an optional solution of the present application, the license plate recognition module uses the vehicle video image captured by the camera to identify the vehicle's license plate number and compares it with the background database to determine whether the vehicle has special access rights. Specifically, the license plate video image is converted into a grayscale video image to reduce the amount of calculation and improve the contrast in subsequent processing. Adaptive histogram equalization CLAHE is used to enhance the contrast of the video image, especially in low-light environments, to ensure that the license plate area is clearly visible. Gaussian filtering is used to smooth the video image, remove noise interference, and make the edge of the license plate clearer. A convolutional neural network CNN is used to detect the specific location of the license plate. The network includes multiple convolutional layers and pooling layers to gradually extract video image features, and finally output the coordinates of the license plate area through a fully connected layer. The fully connected layer outputs the bounding box B of the license plate. plate , which is in the form of, B plate =(x1,y1,x2,y2), (x1,y1) and (x2,y2) are the coordinates of the upper left corner and lower right corner of the license plate area respectively. After the license plate area is determined, the characters in the license plate are segmented and recognized. The characters in the license plate video image are segmented separately by vertical projection method. The segmentation points between characters are found by calculating the sum of the gray values ​​of each column of pixels. P(x)=∑ y= y1 y2 I plate (x, y), where P(x) is the pixel value projection of the license plate video image on column x, I plate (x, y) is the video image of the license plate area. The OCR optical character recognition technology in the deep learning model is used to recognize the characters in the license plate based on the convolutional neural network. Each character is processed by CNN and the output is the probability distribution of the character category. The final output license plate is the character sequence S plate After recognizing the license plate number, the license plate number is compared with the vehicle record in the database to determine the special access rights of the vehicle. The vehicle database is queried to determine whether it is a special vehicle or a vehicle authorized to pass through the height restriction area. If the match is successful, the height restriction pole is automatically adjusted to allow the vehicle to pass. License plate area detection loss uses bounding box regression loss to optimize license plate area detection, L plate =∑ i=1 N ||B pred i -B true i || 2 , where B pred i is the predicted license plate bounding box in the i-th video image, B true i is the true bounding box. The character recognition loss uses the cross entropy loss to optimize the character recognition model, L char =-∑i=1 M ∑ c=1 C y c i logP(y=c|X i ), where y c i is the true category of the i-th character, P(y=c|X i ) is the character category probability predicted by the model. By processing the video image preprocessing, license plate detection, and character recognition steps in parallel on the edge computing device, license plate recognition can be completed in a short time.

[0021] As an optional solution of the present application, the detour guidance module obtains relevant information of the super-height vehicle from the license plate recognition module and the vehicle height recognition module, and generates the optimal detour path based on real-time road conditions, traffic flow, height restriction area location and other information. The detour path planning is implemented based on graph theory and dynamic programming. The goal is to generate an optimal path for the super-height vehicle to bypass the height restriction area. The road network graph is represented as a weighted directed graph G=(V,E), where V is a set of nodes representing intersections or key points of the road, and E is an edge set representing the roads connecting these nodes. Each edge e∈E has a weight w(e). Use the Dijkstra algorithm to calculate the shortest path from the current position of the vehicle to the destination. The current position of the vehicle is v0, and the entrance to the height restriction area is v L , the destination of the vehicle is v f , starting from the current position of the vehicle v0, traverse the nodes in the road network one by one, and gradually find the minimum path cost until the shortest path to bypass the height restriction area is found. Assign d(v0)=0 to the starting point v0 and d(v)=∞ to all other nodes; find the node u closest to v0 among the unprocessed nodes and mark it as processed; for each neighboring node v, update its distance d(v)=min(d(v),d(u)+w(u,v)); repeat the steps to find the node closest to the unprocessed node and update the distance of the neighboring nodes until the path to the destination v is found. f In the process of route planning, the route is adjusted dynamically in combination with real-time traffic flow data. If a section of the current route is congested suddenly and the travel time increases, the shortest route is recalculated and a new detour is generated. V2X communication technology is used to transmit the generated detour route to the vehicle's on-board terminal or the driver's smartphone. By communicating with the road infrastructure, the system can push detour suggestions to the vehicle within milliseconds.

[0022] According to one aspect of an embodiment of the present application, a vehicle height limit warning device based on intelligent video recognition technology is provided, and the device includes the following components.

[0023] Smart camera component, the smart camera has built-in light detection sensor and automatic adjustment module, which can sense the ambient light intensity in real time and automatically adjust the shooting parameters to adapt to different lighting environments. The camera is equipped with high-definition lens, wide-angle field of view and night vision function to ensure the capture of clear video images and videos around the clock. It is installed on both sides of the road and on the elevated structure in front of the height limit pole, covering the entire height limit area and access roads.

[0024] The data processing and storage components use high-performance GPU processors, support parallel computing, have built-in image processing units, and are equipped with large-capacity storage units to store historical traffic flow and video data collected in real time. They are centrally located in road management centers or cloud servers near height-restricted areas and connect to various terminal devices in the system for data processing and transmission.

[0025] The multi-channel warning component, including a large-screen LED display, an on-board communication module and a voice prompt system, can exchange data with the driver's on-board device or mobile phone to ensure the accurate delivery of warning information. The LED display is located within the visible range in front of the height-restricted area, and the on-board communication module is paired with the vehicle's on-board terminal.

[0026] The detour guidance component includes a high-precision GPS module, a navigation calculation unit and a communication module. It can generate the optimal detour route based on the real-time conditions of traffic flow, road conditions and height-restricted areas. It is located in the system's background server and performs calculations and information transmission. Some hardware is integrated into the driver's on-board terminal device.

[0027] The violation recording component is equipped with an automatic recording module, which combines with the license plate recognition system to update the information of illegal vehicles in real time, links with relevant departments through the network, sends penalty notices or restriction instructions, is integrated into the data processing and storage device, and is directly connected to the background system of the traffic management department.

[0028] The present application provides a vehicle height limit warning system and device based on intelligent video recognition technology, which has the following advantages.

[0029] Real-time monitoring of vehicle height and immediate warning when the vehicle is over-height greatly reduces the risk of accidents caused by over-height vehicles. Through advanced intelligent video recognition technology, the vehicle type, height and roof attachments are accurately identified to ensure the accuracy of the warning information. Automatically record the information of illegal vehicles and link with the traffic management department to realize the automatic recording of violations and the generation of penalty notices, improving management efficiency. Through a variety of warning methods, such as V2X communication, LED display screens and on-board terminals, ensure that the warning information can be conveyed to the driver in a timely and accurate manner, improving the user experience. Combined with real-time traffic flow data and vehicle travel time, the path planning is dynamically adjusted to provide the vehicle with the best detour route, reducing congestion and travel time. The smart camera component is equipped with night vision function and automatic adjustment module to ensure normal operation in various weather and lighting environments, realizing all-weather monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A structural diagram of a vehicle height limit warning system based on intelligent video recognition technology provided for one embodiment of the present application.

[0031] Figure 2 A flow chart of a vehicle height limit warning method based on intelligent video recognition technology is provided for one embodiment of the present application.

[0032] Figure 3 A structural diagram of a vehicle height limit warning device based on intelligent video recognition technology provided for one embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0034] Please refer to Figure 1 , provides a vehicle height limit warning system based on intelligent video recognition technology, including the following modules.

[0035] The data acquisition module uses light detection sensors to sense the ambient light intensity in real time, automatically adjusts the camera's capture parameters for different lighting environments, and obtains vehicle video data on connected roads.

[0036] The image enhancement module extracts video images from the collected vehicle video data, performs adaptive video image enhancement processing, and improves the video image quality.

[0037] The vehicle height recognition module identifies the type and height of vehicles approaching the height-restricted area based on the CNN and Transformer architecture.

[0038] The license plate recognition module uses OCR and deep learning technology to identify the vehicle's license plate number and compare it with the backend database to determine the vehicle's special access rights.

[0039] The detour guidance module identifies over-height vehicles and uses navigation data or V2X communication technology to generate a detour route and send it directly to the driver or the on-board system to help bypass the height-restricted area.

[0040] The automatic violation recording module records the license plate information of overheight vehicles and compares it with the vehicle records in the database. For vehicles with multiple violations, it automatically generates penalty notices or traffic restriction instructions and handles them through relevant departments.

[0041] The multi-channel warning module accurately delivers warning information to the driver through V2X communication, LED display screens, and on-board terminals. It also provides detour suggestions for over-high vehicles through display screens and navigation systems.

[0042] As an optional solution of this application, the data acquisition module includes a light detection sensor, an environmental perception sensor, a camera and a historical traffic flow collection device. The light detection sensor is used to monitor the ambient light intensity in real time. Accurately perceive the current lighting environment, such as daytime, nighttime, strong light, and shadow areas, to ensure the quality of video image capture by the camera under different lighting conditions. The ambient light is measured by the light detection sensor to obtain the light intensity L of the current scene, L=1 / A∑ i=1 A I(i), where A is the total number of pixels in the video image and I(i) is the brightness value of each pixel. Feedback to the data acquisition module is used to automatically adjust the camera ISO, exposure and shutter speed to cope with different lighting environments. For example, based on the light intensity, the camera exposure value E is dynamically adjusted, E=G / (L+ε), where G is the preset target exposure value, L is the current light intensity, and ε is a small constant to avoid the denominator being zero. The camera is installed on both sides of the height restriction area to capture the video data of vehicles entering the height restriction area. The camera supports high-definition resolution, wide-angle field of view and night vision to ensure that every vehicle passing through the height restriction area can be fully recorded. After adjusting the parameters through the light detection sensor, the camera component can capture the video data of vehicles approaching the height restriction area. The vehicle video data contains the appearance, outline, and detailed information of the roof load of the vehicle for subsequent vehicle height recognition and license plate recognition modules. The data acquisition module not only relies on physical equipment to collect data, but also obtains information from other data sources, such as weather forecast services, traffic management platforms, and fleet management systems. External data, together with local real-time data, provide comprehensive information support for the height restriction warning system.

[0043] As an optional solution of the present application, the image enhancement module extracts the most representative video image by optical flow method in the video when the vehicle passes through the height-restricted area to ensure the recognition effect. The optical flow method calculates the video image by comparing the motion change between adjacent frames. , where V(x,y,t) is the optical flow vector, is a partial derivative, which indicates the changes of the video image in space and time. Larger optical flow values ​​correspond to frames with more obvious object motion, which can be considered as video images. For appropriate video images, the ACE algorithm is used to dynamically adjust the contrast of the video image when the ambient light changes, such as strong light or low light conditions. I′(x,y)=α×(I(x,y)-μ(I))+μ(I), where I′(x,y) is the enhanced video image, I(x,y) is the original video image, α is the enhancement coefficient, and μ(I) is the average brightness of the video image. By adjusting the value of α, the contrast enhancement effect can be dynamically controlled. In low-light environments, Gamma correction is used to adjust the brightness and contrast of the video image. Gamma correction can linearly adjust the grayscale value of the video image to make the darker areas clearer. I′(x,y)=I(x,y) γ , where γ is the gamma correction coefficient. Usually γ∈(0,1) is used to enhance the details of low-brightness areas. According to the detection results of ambient light, the γ value is dynamically adjusted to ensure that the brightness of the video image is enhanced in a low-light environment. The CLAHE algorithm is used to further enhance the contrast and details of the video image. By dividing the video image into multiple small blocks and performing histogram equalization processing on each block, the loss of details caused by overall over-enhancement is prevented. It is suitable for video image enhancement in complex lighting environments. CLAHE (x,y)=[CDF clip (I(x,y))-min(CDF clip )] / [1-min(CDF clip )]×(L-1), where I CLAHE (x, y) is the video image after CLAHE enhancement, CDF clipis the cumulative distribution function after histogram equalization, and L is the grayscale level of the video image. In foggy weather or when the visual noise is large, a defogging algorithm, such as a defogging algorithm based on dark channel prior, is used to enhance the clarity of the video image. By assuming that the dark channel in the non-sky area of ​​the video image has a lower brightness, the effect of haze can be effectively removed. J(x)=(I(x)-t(x)) / A+t(x) where I(x) is a foggy video image, J(x) is a defogged video image, t(x) is a transmittance map, and A is atmospheric light. By estimating the transmittance map and atmospheric light of the video image, the real video image under the influence of haze is restored. Use SRGAN to improve the resolution of low-resolution video images. SRGAN generates high-resolution video images from low-resolution video images through a generative adversarial network architecture, thereby enhancing the detail performance of vehicle video images. The loss function is, L SRGAN =E IHR [logD(I HR )]+E ILR [log(1-D(G(I LR )))]Among them, I HR is a high-resolution video image, I LR is a low-resolution video image, G is the generator, and D is the discriminator. Through adversarial training, the generator can enhance the details of the video image, making the recognition model more accurate in judging the vehicle. The enhanced video image is passed to the vehicle height recognition module and the license plate recognition module for subsequent vehicle type, license plate, and height analysis. The enhancement effect is evaluated based on the PSNR peak signal-to-noise ratio and SSIM structural similarity, PSNR=10log 10 (MAX I 2 / MSE), SSIM(x,y)=(2μ x μ y +c1)(2σ xy +c2) / (μ x 2 +μ y 2 +c1)(σ x 2 +σ y 2 +c2), the quality of the enhanced video image is evaluated by PSNR and SSIM to ensure that the video image has high detail retention and structural similarity after enhancement.

[0044] As an optional solution of the present application, the vehicle height recognition module uses a deep learning model based on the convolutional neural network CNN and Transformer architecture to accurately identify the height and type of the vehicle, extract the structural features of the vehicle through the video image, and calculate its height. CNN is used to extract local features in video images. Through the stacking of convolutional layers and pooling layers, CNN can learn the edge, shape and spatial structure information of the vehicle. Transformer is mainly used to capture long-distance dependencies and global features in video images to ensure that even when the perspective changes and the vehicle angle is diverse, the model can still accurately identify the height and type of the vehicle. Specifically, the enhanced video image is first normalized and resized to ensure that the size of the video image input to CNN is consistent. The input video image is I(x,y). The video image is first normalized to the range of [0,1] and adjusted to a fixed size, such as 224x224 pixels, I′(x,y)=[I(x,y)-min(I)] / [max(I)-min(I)]. The normalized video image I′(x,y) can accelerate the convergence of the model and ensure the consistency of vehicle video images of different sizes when input into CNN. Through multiple convolutional layers, the low-level features of the vehicle are extracted. For the input video image I(x,y), the convolution kernel size is K×K, and the convolution output is O(x,y), O(x,y)=∑ i=-K / 2 K / 2 ∑ j=-K / 2 K / 2 W(i,j)I(x+i,y+j)+b, where W(i,j) is the weight of the convolution kernel and b is the bias term. After multiple convolution layers are superimposed, CNN can gradually extract the local features of the vehicle. After the convolution layer, the pooling layer is used to reduce the size of the video image and retain important features. pool (x,y)=max{O(x+i,y+j)},∀i,j∈pool_window, maximum pooling is used to retain important edge features in video images while reducing the amount of computation.

[0045] After being processed, the local features extracted by the convolutional layer enter the Transformer model. The Transformer processes the global contextual relationship in the video image through the self-attention mechanism to help identify the overall shape and height features of the vehicle. The multi-head self-attention mechanism divides the video image features into multiple attention heads, and processes different parts of the video image through different heads, thereby capturing the global information of the vehicle height. The input is the feature vector X, and the self-attention output of each head is, head i =Attention(Q i ,K i ,V i ), where Q i =W iQ X,K i =W i K X,V i =W i V X, W i Q ,W i K ,W i V is the weight matrix of different heads. Through the multi-head attention mechanism, multiple video image regions are processed in parallel; position encoding is added to help understand the position information of pixels in the video image, PE(pos,2i)=sin(pos / 10000 2i / dmodel ),PE(pos,2i+1)=cos(pos / 10000 2i / dmodel ), where pos is the position and dmodel is the dimension of the feature vector. Position encoding helps the Transformer model retain the spatial position information of the video image during feature processing.

[0046] The pixel height H of the vehicle is obtained by using the local features extracted by CNN and the global features processed by Transformer. pred and type as output, H pred =W H Concat(CNN_features,Transformer_features)+b H , where W H is the regression weight matrix, b H is the bias term; the vehicle type is classified, the number of categories is C, and the output probability distribution is, P(y=c)=softmax(W c Concat(CNN_features,Transformer_features)+b c ), where W c is the classification weight matrix, b c The vehicle pixel height estimation uses the mean square error loss function, and the type classification uses the cross entropy loss. The total loss function L is the weighted sum of the two, L = α × MSE (H true ,H pred )+β×∑ c=1 C y c logP(y=c), where α and β are weight factors for height estimation and category classification, H true is the true height, H pred is the predicted height, y cis the indicator variable of the true category. A large-scale labeled dataset is used for model training, including video image data in different environments, angles, and vehicle types, to ensure that the model is robust to diverse scenarios. Based on the predicted vehicle pixel height, combined with reference objects around the vehicle, such as height limit poles, the actual height of the vehicle is inferred. The precise height, H, is obtained through the proportional relationship in the video image and the pre-set scale model. vehicle =H image S real / S pixel , where H vehicle is the actual height of the vehicle, H image is the pixel height of the vehicle in the video image, S pixel is the pixel value of the reference scale in the video image, S real is the actual height of the reference ruler.

[0047] As an optional solution of this application, the license plate recognition module uses the vehicle video image captured by the camera to identify the vehicle's license plate number and compares it with the backend database to determine whether the vehicle has special access rights. Specifically, the license plate video image is converted into a grayscale video image to reduce the amount of calculation and improve the contrast in subsequent processing. gray (x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y), where R(x,y), G(x,y), B(x,y) are the pixel values ​​of the video image in the red, green, and blue channels, respectively, and Igray(x,y) is the grayscale video image. Adaptive histogram equalization CLAHE is used to enhance the contrast of the video image, especially in low-light environments, to ensure that the license plate area is clearly visible. Gaussian filtering is used to smooth the video image and remove noise interference, making the license plate edge clearer. smooth (x,y)=∑ i=-k k ∑ j=-k k G(i,j)I gray (x+i,y+j), where G(i,j) is the Gaussian kernel function, I gray (x, y) is a grayscale video image. A convolutional neural network (CNN) is used to detect the specific location of the license plate. The network includes multiple convolutional layers and pooling layers to gradually extract video image features. Finally, a fully connected layer outputs the coordinates of the license plate area. The fully connected layer outputs the bounding box B of the license plate. plate , which is in the form of, B plate=(x1,y1,x2,y2), (x1,y1) and (x2,y2) are the coordinates of the upper left corner and lower right corner of the license plate area respectively. After the license plate area is determined, the characters in the license plate are segmented and recognized. The characters in the license plate video image are segmented separately by vertical projection method. The segmentation points between characters are found by calculating the sum of the gray values ​​of each column of pixels. P(x)=∑ y= y1 y2 I plate (x, y), where P(x) is the pixel value projection of the license plate video image on column x, I plate (x, y) is the video image of the license plate area. The OCR optical character recognition technology in the deep learning model is used to recognize the characters in the license plate based on the convolutional neural network. Each character is processed by CNN and the output is the probability distribution of the character category. The final output license plate is the character sequence S plate After recognizing the license plate number, the license plate number is compared with the vehicle record in the database to determine the special access rights of the vehicle. The vehicle database is queried to determine whether it is a special vehicle, such as a fire truck, ambulance, or a vehicle authorized to pass through the height-restricted area. If the match is successful, the height-restricted pole is automatically adjusted to allow the vehicle to pass. License plate area detection loss uses bounding box regression loss to optimize license plate area detection, L plate =∑ i=1 N ||B pred i -B true i || 2 , where B pred i is the predicted license plate bounding box in the i-th video image, B true i is the true bounding box. The character recognition loss uses the cross entropy loss to optimize the character recognition model, L char =-∑ i=1 M ∑ c=1 C y c i logP(y=c|X i ), where y c i is the true category of the i-th character, P(y=c|X i ) is the character category probability predicted by the model. By processing the video image preprocessing, license plate detection, and character recognition steps in parallel on the edge computing device, license plate recognition can be completed in a short time.

[0048] As an optional solution of the present application, the detour guidance module obtains relevant information of super-high vehicles from the license plate recognition module and the vehicle height recognition module, and generates the optimal detour path based on real-time road conditions, traffic flow, height-restricted area location and other information. The detour path planning is implemented based on graph theory and dynamic programming, and the goal is to generate an optimal path for super-high vehicles to bypass the height-restricted area. The road network graph is represented as a weighted directed graph G=(V,E), where V is a set of nodes representing intersections or key points of the road, and E is an edge set representing the roads connecting these nodes. Each edge e∈E has a weight w(e), w(e)=α×d(e)+β×t(e)+γ×c(e).

[0049] Among them, d(e) is the physical distance of edge e, t(e) is the travel time under current traffic conditions, c(e) represents the degree of traffic congestion on the road section, calculated using traffic flow data, and α, β, γ are weight coefficients of different factors. The Dijkstra algorithm is used to calculate the shortest path from the current position of the vehicle to the destination. The current position of the vehicle is v0, and the entrance to the height restriction area is v L , the destination of the vehicle is v f , Dijkstra algorithm starts from the current position of the vehicle v0, traverses the nodes in the road network one by one, and gradually finds the minimum path cost until the shortest path to bypass the height restriction area is found. The starting point v0 is assigned d(v0)=0, and all other nodes are assigned d(v)=∞; find the node u closest to v0 among the unprocessed nodes and mark it as processed; for each neighboring node v, update its distance d(v)=min(d(v),d(u)+w(u,v)); repeat the steps to find the node closest to the unprocessed node and update the distance of the neighboring nodes until the path to the destination v is found. f The shortest path. During the path planning process, the path is adjusted dynamically in combination with real-time traffic flow data. If a sudden congestion occurs on a section e on the current path, the travel time t(e) increases, the shortest path is recalculated and a new detour route is generated, w′(e)=w(e)+δc′(e), where c′(e) is the real-time congestion level of the current section, and δ is the congestion penalty factor. The path planning algorithm will recalculate the detour path based on the new weight. V2X communication technology is used to transmit the generated detour path to the vehicle's on-board terminal or the driver's smartphone. By communicating with the road infrastructure, the system can push detour suggestions to the vehicle within milliseconds.

[0050] As an optional solution of the present application, the automatic violation recording module records the violation of the overheight vehicle, automatically records the vehicle information through the license plate recognition system, and uploads the records of vehicles with multiple violations to the traffic management department. Specifically, based on the results of the vehicle height recognition module, it is determined whether the vehicle is overheight or violates the height limit regulations. The license plate number of the vehicle is obtained through the license plate recognition module and compared with the historical data to determine whether it is a multiple violation. If the vehicle has been recorded in the database, it is marked as a multiple violation vehicle. For a vehicle with a single violation, the system will record the violation of the vehicle and generate a warning notification. The notification content includes vehicle information, violation time, and height limit area location. For vehicles with multiple violations, a penalty notice or restriction instruction is generated and automatically sent to the driver via email, text message or the traffic management department system.

[0051] Please refer to Figure 2 , provides a vehicle height limit warning method based on intelligent video recognition technology, including the following steps.

[0052] S1. Monitor the ambient light intensity in real time through the data acquisition module and automatically adjust the camera parameters according to the light intensity.

[0053] S2. Use the image enhancement module to perform video image extraction and video image enhancement processing on the collected vehicle video data to improve the video image quality.

[0054] S3, the vehicle height recognition module recognizes the vehicle type, height, and roof attachments through CNN and Transformer architecture.

[0055] S4. The license plate recognition module uses OCR and deep learning technology to identify the vehicle's license plate number and compares it with the backend database to determine the vehicle's special access rights.

[0056] S5. When the detour guidance module identifies an overheight vehicle, it uses navigation data or V2X communication technology to generate a detour route and directly sends it to the driver or the vehicle system to help bypass the height-restricted area.

[0057] S6. The automatic violation recording module records the license plate information of the overheight vehicle and compares it with the vehicle records in the database. For vehicles that violate the regulations multiple times, the system automatically generates a penalty notice or a restriction order and handles it through relevant departments.

[0058] S7, the multi-channel warning module accurately delivers warning information to the driver through V2X communication, LED display screen, and on-board terminal. Especially for over-high vehicles, it provides detour suggestions through the display screen and navigation system.

[0059] In step S1 of the present application, the data acquisition module monitors the ambient light intensity in real time and automatically adjusts the camera parameters according to the light intensity: it can automatically adapt to various lighting environments to ensure that the camera always captures clear images, reduce manual intervention, and improve the response speed and accuracy of the system.

[0060] In step S2 of the present application, the image enhancement module is used to extract and enhance the collected vehicle video data to improve the quality of the video image: improve the clarity and details of the video image, so that the subsequent recognition processing is more accurate. Higher quality images help improve the accuracy of vehicle and license plate recognition.

[0061] In step S3 of this application, the vehicle height recognition module uses CNN and Transformer architecture to identify vehicle type, height and roof attachments: using deep learning algorithms, it can accurately identify different types of vehicles and their heights, even including roof attachments. Rapidly process and analyze video data to achieve real-time warning and recognition.

[0062] In step S4 of this application, the license plate recognition module uses OCR and deep learning technology to identify the vehicle's license plate number, and compares it with the background database to determine the vehicle's special access rights: combining OCR and deep learning technology, it can efficiently and accurately identify the license plate number. By comparing with the database, it can effectively manage and identify vehicles with special access rights.

[0063] In step S5 of this application, when the detour guidance module identifies an overheight vehicle, it uses navigation data or V2X communication technology to generate a detour route and directly sends it to the driver or the vehicle system to help detour the height-restricted area: providing an accurate detour path to prevent overheight vehicles from entering the height-restricted area. Directly sending detour information to the driver or the vehicle system improves driving safety and convenience.

[0064] In step S6 of this application, the automatic violation recording module records the license plate information of the overheight vehicle and compares it with the vehicle records in the database. For vehicles that violate the rules multiple times, the system automatically generates a penalty notice or a restriction order and handles it through relevant departments: reducing manual intervention, and automatically recording and processing the information of the violating vehicle through the system. Improve the efficiency of management and punishment of illegal vehicles.

[0065] In step S7 of this application, the multi-channel warning module accurately delivers warning information to the driver through V2X communication, LED display screen, and vehicle-mounted terminal. Especially for over-height vehicles, detour suggestions are provided through the display screen and navigation system: warning information is transmitted through multiple channels to ensure that the driver can receive it in time. Specific detour suggestions are provided for over-height vehicles to help drivers avoid height-restricted areas and improve overall traffic safety.

[0066] Please refer to Figure 3 , provides a vehicle height limit warning device based on intelligent video recognition technology, including the following components.

[0067] Smart camera component, the smart camera has built-in light detection sensor and automatic adjustment module, which can sense the ambient light intensity in real time and automatically adjust the shooting parameters to adapt to different lighting environments. The camera is equipped with high-definition lens, wide-angle field of view and night vision function to ensure the capture of clear video images and videos around the clock. It is installed on both sides of the road and on the elevated structure in front of the height limit pole, covering the entire height limit area and access roads.

[0068] The data processing and storage components use high-performance GPU processors, support parallel computing, have built-in image processing units, and are equipped with large-capacity storage units to store historical traffic flow and video data collected in real time. They are centrally located in road management centers or cloud servers near height-restricted areas and connect to various terminal devices in the system for data processing and transmission.

[0069] The multi-channel warning component, including a large-screen LED display, an on-board communication module and a voice prompt system, can exchange data with the driver's on-board device or mobile phone to ensure the accurate delivery of warning information. The LED display is located within the visible range in front of the height-restricted area, and the on-board communication module is paired with the vehicle's on-board terminal.

[0070] The detour guidance component includes a high-precision GPS module, a navigation calculation unit and a communication module. It can generate the optimal detour route based on the real-time conditions of traffic flow, road conditions and height-restricted areas. It is located in the system's background server and performs calculations and information transmission. Some hardware is integrated into the driver's on-board terminal device.

[0071] The violation recording component is equipped with an automatic recording module, which combines with the license plate recognition system to update the information of illegal vehicles in real time, links with relevant departments through the network, sends penalty notices or restriction instructions, is integrated into the data processing and storage device, and is directly connected to the background system of the traffic management department.

Claims

1. A vehicle height limit warning system based on intelligent video recognition technology, characterized in that: The system includes the following modules: The data acquisition module uses a light detection sensor to sense the ambient light intensity in real time, automatically adjusts the camera's capture parameters for different lighting environments, and obtains vehicle video data on connected roads; An image enhancement module extracts video images from the collected vehicle video data and performs adaptive video image enhancement processing; The vehicle height recognition module uses CNN and Transformer architecture to identify the type, height, and roof attachments of vehicles approaching the height-restricted area; The license plate recognition module uses OCR and deep learning technology to identify the vehicle's license plate number and compare it with the backend database to determine the vehicle's special access rights and whether it is an overheight vehicle. The detour guidance module identifies over-height vehicles and uses navigation data or V2X communication technology to generate detour routes and send them directly to the driver or the vehicle system to help bypass height-restricted areas; The automatic violation recording module records the license plate information of overheight vehicles and compares it with the vehicle records in the database. For vehicles that violate the regulations multiple times, the system automatically generates a penalty notice or a traffic restriction order and handles it through relevant departments. The multi-channel warning module accurately delivers warning information to the driver through V2X communication, LED display screen, and on-board terminal. For over-high vehicles, it provides detour suggestions through the display screen and navigation system.

2. According to claim 1, a vehicle height limit warning system based on intelligent video recognition technology is characterized in that: Data acquisition module: The data collection module includes light detection sensors, environmental perception sensors, cameras and historical traffic flow collection equipment; The light detection sensor is used to monitor the ambient light intensity in real time and provide feedback to the data acquisition module for automatic adjustment of the camera ISO, exposure and shutter speed to cope with different lighting environments; Cameras are installed on both sides of the height-restricted area to capture video data of vehicles entering the height-restricted area. The cameras support high-definition resolution, wide-angle field of view, and night vision to ensure that every vehicle passing through the height-restricted area can be fully recorded; After adjusting parameters through the light detection sensor, the camera can capture video data of vehicles approaching the height-restricted area. The vehicle video data contains detailed information on the vehicle's appearance and outline.

3. According to claim 1, a vehicle height limit warning system based on intelligent video recognition technology is characterized in that: Image enhancement module: When a vehicle passes through a height-restricted area, the most representative video image is extracted through the optical flow method to ensure recognition. The optical flow method calculates the video image by comparing the motion changes between adjacent frames; Enhance the video image and use the ACE algorithm to dynamically adjust the contrast of the video image under changing ambient light conditions; In low-light environments, Gamma correction is used to adjust the brightness and contrast of video images; CLAHE algorithm is used to enhance the contrast and details of video images; Use SRGAN to improve the resolution of low-resolution video images. SRGAN generates high-resolution video images from low-resolution video images through a generative adversarial network architecture to enhance the details of vehicle video images. The loss function is, L SRGAN =E IHR [logD(I HR )]+E ILR [log(1-D(G(I LR )))]Among them, I HR is a high-resolution video image, I LR is a low-resolution video image, G is the generator, and D is the discriminator; The enhanced video image is passed to the vehicle height recognition module and the license plate recognition module for subsequent vehicle type, license plate and height analysis; The enhancement effect is evaluated based on PSNR peak signal-to-noise ratio and SSIM structural similarity. The quality of the enhanced video image is evaluated by PSNR and SSIM to ensure that the video image has high detail retention and structural similarity after enhancement.

4. According to claim 1, a vehicle height limit warning system based on intelligent video recognition technology is characterized in that: Vehicle height recognition module: The vehicle height recognition module uses a deep learning model based on the convolutional neural network (CNN) and the Transformer architecture to accurately identify the height and type of the vehicle. It extracts the structural features of the vehicle through video images, identifies the vehicle type and calculates its height. CNN is used to extract local features in video images. Through the stacking of convolutional layers and pooling layers, CNN can learn the edge, shape and spatial structure information of the vehicle; Transformer is mainly used to capture long-distance dependencies and global features in video images, ensuring that the model can accurately identify the height and type of vehicles even when the perspective changes and the vehicle angles are diverse.

5. The vehicle height limit warning system based on intelligent video recognition technology according to claim 4 is characterized in that: Extract local features from video images: The enhanced video image is first normalized and resized to ensure that the size of the video image input to CNN is consistent. The input video image is I(x,y). The video image is first normalized to the range of [0,1] and adjusted to a fixed size, I′(x,y)=[I(x,y)-min(I)] / [max(I)-min(I)]. The normalized video image I′(x,y) can accelerate the convergence of the model and ensure that the vehicle video images of different sizes are consistently input into CNN. Through multiple convolutional layers, the low-level features of the vehicle are extracted. For the input video image I(x, y), the convolution kernel size is K×K, and the convolution output is O(x, y), O(x, y)=∑ i=-K / 2 K / 2 ∑ j=-K / 2 K / 2 W(i,j)I(x+i,y+j)+b, where W(i,j) is the weight of the convolution kernel and b is the bias term. After multiple convolution layers are superimposed, CNN can gradually extract the local features of the vehicle; After the convolutional layer, the pooling layer is used to reduce the size of the video image and retain important features. pool (x,y)=max{O(x+i,y+j)},∀i,j∈pool_window, maximum pooling is used to retain important edge features in video images while reducing the amount of computation.

6. The vehicle height limit warning system based on intelligent video recognition technology according to claim 4 is characterized in that: Capturing long-range dependencies and global features in video images: After being processed, the local features extracted by the convolutional layer enter the Transformer model. The Transformer processes the global contextual relationship in the video image through the self-attention mechanism to help identify the overall shape and height characteristics of the vehicle. The multi-head self-attention mechanism divides the video image features into multiple attention heads, and processes different parts of the video image through different heads to capture the global information of the vehicle height. The input is the feature vector X, and the self-attention output of each head is, head i =Attention(Q i ,K i ,V i ), where Q i =W i Q X,K i =W i K X,V i =W i V X, W i Q ,W i K ,W i V The weight matrices of different heads are used to process multiple video image regions in parallel through the multi-head attention mechanism; Adding position encoding helps understand the position information of pixels in the video image, PE(pos,2i)=sin(pos / 10000 2i / dmodel ),PE(pos,2i+1)=cos(pos / 10000 2i / dmodel ), where pos is the position and dmodel is the dimension of the feature vector. Position encoding helps the Transformer model retain the spatial position information of the video image during feature processing.

7. The vehicle height limit warning system based on intelligent video recognition technology according to claim 4 is characterized in that: Identify the vehicle type and calculate its height: pass The local features extracted by CNN and the global features processed by Transformer are used to transform the pixel height H of the vehicle pred The vehicle type is classified using the and type as output. The number of categories is C and the output probability distribution is P(y=c); The vehicle pixel height estimation uses the mean square error loss function, and the type classification uses the cross entropy loss. The total loss function L is the weighted sum of the two, L = α × MSE (H true ,H pred )+β×∑ c=1 C y c logP(y=c), where α and β are weight factors for height estimation and category classification, H true is the true height, H pred is the predicted height, y c is an indicator variable of the true category; Use large-scale annotated datasets for model training, including video image data in different environments, angles, and vehicle types, to ensure that the model is robust to diverse scenarios; According to the predicted vehicle pixel height, combined with the actual height of the reference objects around the vehicle, the precise height is obtained through the proportional relationship in the video image and the pre-set scale model, H vehicle =H image S real / S pixel , where H vehicle is the actual height of the vehicle, H image is the pixel height of the vehicle in the video image, S pixel is the pixel value of the reference scale in the video image, S real is the actual height of the reference ruler.

8. The vehicle height limit warning system based on intelligent video recognition technology according to claim 1 is characterized in that: License Plate Recognition Module: Convert the license plate video image into a grayscale video image; Use adaptive histogram equalization CLAHE; Use Gaussian filtering to smooth the video image and remove noise interference; A convolutional neural network (CNN) is used to detect the specific location of the license plate. The network includes multiple convolutional layers and pooling layers to gradually extract video image features. Finally, the coordinates of the license plate area are output through a fully connected layer. The fully connected layer outputs the bounding box B of the license plate. plate , B plate =(x1,y1,x2,y2), (x1,y1) and (x2,y2) are the coordinates of the upper left corner and lower right corner of the license plate area respectively; Segment and recognize the characters in the license plate. Use the vertical projection method to separate the characters in the license plate video image. Find the segmentation point between the characters by calculating the sum of the grayscale values ​​of each column of pixels. P(x)=∑ y=y1 y2 I plate (x, y), where P(x) is the pixel value projection of the license plate video image on column x, I plate (x, y) is the video image of the license plate area; The OCR optical character recognition technology in the deep learning model is used to identify the characters in the license plate based on the convolutional neural network. Each character is processed by CNN and the output is the probability distribution of the character category. The final output license plate is the character sequence S plate ; Compare the license plate number with the vehicle records in the database to determine the special access rights of the vehicle, query the vehicle database to determine whether it is a special vehicle or a vehicle authorized to pass through the height-restricted area. If a match is successful, the height-restricted pole is automatically adjusted to allow the vehicle to pass; The license plate area detection loss uses the bounding box regression loss to optimize the license plate area detection, L plate =∑ i=1 N ||B pred i -B true i || 2 , where B pred i is the predicted license plate bounding box in the i-th video image, B true i is the true bounding box; Character recognition loss uses cross entropy loss to optimize the character recognition model, L char =-∑ i=1 M ∑ c=1 C y c i logP(y=c|X i ), where y c i is the true category of the i-th character, P(y=c|X i ) is the character category probability predicted by the model.

9. The vehicle height limit warning system based on intelligent video recognition technology according to claim 1 is characterized in that: Detour guidance module: Generate the optimal detour route based on real-time road conditions, traffic flow, and height-restricted area locations. Detour route planning is based on graph theory and dynamic programming, generating an optimal route for super-height vehicles to bypass height-restricted areas. The road network graph is represented as a weighted directed graph G=(V,E), where V is a set of nodes, representing intersections or key points of roads, and E is a set of edges, representing roads connecting these nodes; each edge e∈E has a weight w(e); Use Dijkstra algorithm to calculate the shortest path from the current position of the vehicle to the destination. The current position of the vehicle is v0 and the entrance of the height-restricted area is v L , the destination of the vehicle is v f , starting from the current position of the vehicle v0, traverse the nodes in the road network one by one, and gradually find the minimum path cost until the shortest path to bypass the height restriction area is found. Assign d(v0)=0 to the starting point v0 and d(v)=∞ to all other nodes; find the node u closest to v0 among the unprocessed nodes and mark it as processed; for each neighboring node v, update its distance d(v)=min(d(v),d(u)+w(u,v)); repeat the steps to find the node closest to the unprocessed node and update the distance of the neighboring nodes until the path to the destination v is found. f The shortest path; During the route planning process, the route is adjusted dynamically based on real-time traffic flow data. If a section of the current route is suddenly congested and the travel time increases, the shortest route is recalculated and a new detour is generated. V2X communication technology is used to transmit the generated detour route to the vehicle's on-board terminal or the driver's smartphone. By communicating with the road infrastructure, the system can push detour suggestions to the vehicle within milliseconds.

10. The device used in the vehicle height limit warning system based on intelligent video recognition technology according to claim 1, characterized in that: The device comprises the following components; Smart camera assembly, with built-in light detection sensors and automatic adjustment modules, installed on both sides of the road and on the elevated structure in front of the height limit pole, covering the entire height limit area and access roads; Data processing and storage components use high-performance GPU processors, support parallel computing, have built-in image processing units, and are equipped with large-capacity storage units. They are centrally located in road management centers or cloud servers near height-restricted areas and connect to various terminal devices in the system for data processing and transmission. The multi-channel warning component includes a large-screen LED display, an on-board communication module and a voice prompt system. It can exchange data with the driver's on-board device or mobile phone to ensure the accurate delivery of warning information. The LED display is located in the visible range in front of the height-restricted area, and the on-board communication module is paired with the vehicle's on-board terminal. The detour guidance component includes a high-precision GPS module, a navigation calculation unit and a communication module. It can generate the optimal detour route based on the real-time situation of traffic flow, road conditions and height-restricted areas. It is located in the system's backend server and performs calculations and information transmission. Some of the hardware is integrated into the driver's in-vehicle terminal device. The violation recording component is equipped with an automatic recording module, which combines with the license plate recognition system to update the information of illegal vehicles in real time, links with relevant departments through the network, sends penalty notices or restriction instructions, is integrated into the data processing and storage device, and is directly connected to the background system of the traffic management department.

Images