Dynamic road toll charging method and system based on vehicle type and vehicle driving path

Abstract

The invention discloses a dynamic road toll charging method and system based on vehicle types and vehicle driving paths. The method comprises the steps that 1, a target vehicle owner inputs vehicle information into a database in advance; 2, related information of a target vehicle is collected at an entrance of a toll road, and after a camera at the entrance of the road captures a vehicle image, the vehicle image is uploaded to a rear-end processing system through a 5G network and a communication line, and accurate recognition and processing of a license plate number of the target vehicle are achieved; 3, according to cameras installed at important nodes in the toll road, the cameras capture vehicle images in real time when the target vehicle passes by, and the driving path of the vehicle is calculated in combination with known camera information; and 4, the driving path of the target vehicle is calculated, and the system carries out real-time fee deduction through a vehicle-mounted payment device or a mobile payment platform. According to the invention, license plate recognition and path tracking are carried out on a driving automobile, the driving mileage is calculated, and dynamic road toll collection is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of road toll systems, and in particular to a dynamic road toll charging method and system based on vehicle type and vehicle travel path. Background Art

[0002] In the prior art, path-based tracking and charging methods have been studied to a certain extent. For example, "The Optimal Road Congestion Charging Model in a Mixed Manual and Automatic Driving Environment" (Li Hao, 2023) explores the charging scheme under the condition of coexistence of autonomous driving vehicles and manually driven vehicles based on road capacity, and Publication No. CN117540890A (November 2023) proposes a path optimization method based on the taboo search algorithm, which can find the optimal solution in a global range and optimize the path charging process.

[0003] However, these technologies focus more on solving theoretical path optimization solutions and autonomous driving charging solutions. However, in broader application scenarios, they do not discuss the problem of insufficient highway tolls caused by electric vehicles not paying fuel taxes because they do not use fuel, and there are still deficiencies in practical applications. Summary of the invention

[0004] In order to overcome the defects of the above-mentioned prior art, the present invention proposes a dynamic road toll collection method and system based on vehicle type and vehicle driving path, which realizes dynamic road charging by performing license plate recognition and path tracking on the moving car and calculating its mileage.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A dynamic road toll collection method based on vehicle type and vehicle travel path comprises the following steps;

[0007] Step 1: The owner of the target vehicle enters the vehicle information into the database in advance;

[0008] Step 2: Set up a camera at the entrance of the toll road to collect relevant information of the target vehicle. After the camera at the entrance captures the vehicle image, it is uploaded to the back-end processing system through the 5G network and communication line. The system uses automatic license plate recognition technology (ANPR) and optical character recognition technology (OCR), combined with the recognition of the license plate color, to accurately identify and process the target vehicle's license plate number;

[0009] Step 3: Cameras are installed at important nodes inside the toll road. The cameras capture vehicle images in real time when the target vehicle passes by, and the vehicle's driving path is calculated based on known camera information.

[0010] Step 4: The target vehicle's driving path is calculated, and the system calculates the fee based on the pre-set charging standards (unit: yuan / km) for different vehicle types, i.e. new energy vehicles, fuel vehicles, and trucks. Finally, the system deducts the fee in real time through the on-board payment device (such as ETC device) or mobile payment platform (such as Alipay, WeChat Pay).

[0011] In step 1, the vehicle information includes vehicle color, license plate number, vehicle type and vehicle energy type.

[0012] In step 2, the relevant information includes target vehicle information.

[0013] After the camera at the entrance of the road captures a high-resolution image of the target vehicle, the automatic license plate recognition technology (ANPR) uses the YOLO model to process the high-resolution image. The model divides the processed high-resolution image into grids and performs border prediction and classification regression on each grid to accurately segment the license plate image and license plate type.

[0014] Optical character recognition (OCR) technology recognizes the segmented license plate images. OCR technology uses a deep learning model that combines convolutional neural network (CNN) and recurrent neural network (RNN) to match and classify license plate images and extract the digital and letter information on the license plate. It uses a multi-layer convolutional neural network to extract features in the image, and then uses a recurrent neural network to process and identify the contextual relationship between characters to ensure the correctness of the character order.

[0015] The vehicle information pre-entered in the database is matched with the collected information to be judged. When the identified license plate number and license plate color of the target vehicle are consistent with the entered vehicle information, the road entrance is released.

[0016] Furthermore, the specific steps of the automatic license plate recognition technology (ANPR) using the YOLO model to process the high-resolution vehicle image captured by the camera are as follows:

[0017] The YOLO model first normalizes the input high-resolution image, adjusts the pixel values ​​to the [0, 1] interval, and uniformly scales it to a fixed size;

[0018] Subsequently, the model divides the high-resolution image into S×S grids. Each grid is responsible for detecting the location of the license plate. When the center point of the license plate falls into a certain grid, the grid is responsible for predicting the bounding box of the license plate. The model predicts B bounding boxes in each grid by regression. The prediction parameters of each bounding box include the center point position (b x ,b y )、width and height(b w ,b h), confidence score c Each grid also outputs a class probability vector P (Class i );

[0019] Among them (b x ,b y ) represents the normalized coordinates of the center of the bounding box relative to the grid cell, (b w ,b h ) represents the normalized ratio of the width and height of the bounding box to the image size. The confidence score c is composed of two parts: one is the probability P(Object) of the existence of the target object, i.e., the license plate, and the other is the overlap between the predicted box and the real box using the intersection over union (IoU) pred,truth Indicates that P(Class i ) represents the predicted probability of each category, i∈{1,2,3} corresponds to the categories of green, blue and white license plates. The category probability is output by the classification head of the model and normalized by the Softmax activation function.

[0020] In the above process, the model determines whether the license plate exists and the degree of overlap between the predicted box and the real box by calculating the confidence score c of each grid. The calculation formula is:

[0021] c=P(Object)×IoU pred,truth

[0022] Among them, P(Object) represents the probability that the target object, i.e., the license plate, exists in the grid, and IoU pred,truth It is the intersection-over-union ratio of the predicted bounding box and the true bounding box; a grid with a high confidence score indicates that there is a high possibility that a license plate exists. At the same time, if IoU pred,truth A higher value indicates that the predicted box and the true box overlap more, and the prediction is more accurate;

[0023] Finally, the YOLO model uses the non-maximum suppression (NMS) algorithm to remove redundant boxes and retain only the box with the highest confidence as the final detection result, that is, the license plate image. The category label of each detection box is obtained by selecting the maximum category probability P (Class i ) is determined by the corresponding category.

[0024] During the training process, the YOLO model optimizes the network parameters by minimizing the loss function, which consists of three parts: position loss, confidence loss, and classification loss. The position loss is used to measure the gap between the prediction and the real box, and is calculated using the mean square error (MSE). The formula is:

[0025]

[0026] in, and are the coordinate parameters of the predicted box and the real box, respectively, coord is the weighting coefficient. The confidence loss is usually calculated by the difference between the predicted confidence and the true confidence, and the formula is:

[0027]

[0028] Among them, c pred and c true are the predicted confidence and the true confidence values ​​respectively. The classification loss usually uses the cross entropy loss function to measure the difference between the predicted and true categories of the license plate category. The formula is:

[0029]

[0030] Among them, y i is the one-hot encoding of the true category, P(Class i ) is the category probability output by the model. Finally, the total loss function of the YOLO model is:

[0031] Loss total =λ coord ·Loss loc +Loss conf +λ class ·Loss class

[0032] By minimizing the total loss function, the model is able to gradually optimize the accuracy of license plate position, confidence, and category prediction.

[0033] The specific steps of the optical character recognition technology (OCR) for recognizing the segmented license plate image are as follows:

[0034] The input image is a license plate image processed by automatic license plate recognition technology. After normalization, it is input into a deep learning model that combines convolutional neural network (CNN) and recurrent neural network (RNN) to complete the process of feature extraction, sequence modeling and character recognition in sequence.

[0035] First, the input license plate image is recorded as Where H and W represent the height and width of the image, respectively, and C is the number of channels. The convolutional neural network CNN performs multi-layer convolution and pooling operations on the image to extract local and global features from the license plate image. The feature map of each layer of convolution is expressed as:

[0036] F (l) =σ(W (l) *F (l-1) +b (l) )

[0037] Among them, F(l-1) is the input feature map of the l-1th layer (X for the first layer), W (l) and b (l) are the weight and bias of the lth layer convolution kernel respectively, σ is a nonlinear activation function such as ReLU activation function. After multi-layer convolution and downsampling, a feature map F is generated, which contains important shape and structure information of the license plate characters;

[0038] Next, the feature map F is expanded into a feature vector {f1,f2,...,f T}, where each Represents the features of a time step. The sequence feature vector is input into a bidirectional recurrent neural network (Bi-RNN), such as a bidirectional long short-term memory network (Bi-LSTM) to capture the contextual dependencies between characters. The forward and backward calculations of Bi-LSTM at each time step t are:

[0039]

[0040] Ultimately, the output at each time step is represented as the concatenation of the forward and backward hidden states:

[0041]

[0042] In order to accurately identify the characters in the feature sequence and maintain their order, the connection temporal classification (CTC) loss function is used. CTC predicts the probability of the character sequence in an unaligned manner, specifically:

[0043]

[0044] Among them, S represents the target character sequence, s t is the predicted character at time step t, F is the input feature map, and the loss function of CTC is defined as:

[0045]

[0046] Where Alignments(S) is the set of all possible character alignments, and π is a specific alignment path. CTC sums all possible paths to ensure that model training does not require explicit alignment annotations;

[0047] Finally, combined with the beam search decoding algorithm, the output character probability distribution is converted into the final character sequence, successfully extracting the digital and letter information in the license plate image and segmenting the license plate number.

[0048] In step 3, cameras are installed at important nodes inside the toll road, such as main line lane diversion points, service area entrances and exits, bridge entrances and exits, tunnel entrances and exits, ramp entrances and exits, dangerous sections (such as continuous curves, steep slopes), road intersections or emergency lanes at certain intervals (such as 10 km).

[0049] When the target vehicle passes each camera, the system extracts the license plate number and the location of the vehicle in the image through automatic license plate recognition technology and optical character recognition technology. The geographical location and angle parameters of each camera installation are known in the system.

[0050] When a vehicle passes through multiple cameras, the system calculates the vehicle's driving path between different nodes by linking the positioning data between the cameras; in addition, the system will correct the camera node distance according to the curvature of the road and the actual driving trajectory of the vehicle, especially when there are bends or uphill and downhill roads. In this case, the system obtains the road's curvature radius, slope, and vertical and horizontal bending angles based on the digital map of the road. When the vehicle is driving on a bend, the system uses a curve distance calculation method. When driving on a ramp, the actual length of the path is obtained by expanding the Pythagoras theorem into a calculation method in three-dimensional space. The actual mileage of the target vehicle on the toll section is calculated by accumulating the distances between the path nodes.

[0051] In the above curved road section, the actual driving path length of the vehicle is determined by the road curvature radius R and the horizontal bending angle θ. According to the road geometry, the actual driving distance L of the vehicle on the curve is curve The calculation formula is:

[0052] L curve =R·θ

[0053] Where R is the radius of curvature of the road in meters, and θ is the horizontal curvature angle of the road in radians.

[0054] For the above slope section, the actual path length of the vehicle needs to take into account the vertical height change h and the horizontal projection length L of the road. horizontal Based on the calculation principle of path length in three-dimensional space, the actual driving path L of the vehicle slope By extending Pythagoras' theorem, we can get:

[0055]

[0056] Where h is the vertical height difference of the ramp, L horizontal is the horizontal distance of the ramp. By correcting the actual driving distance on curves and ramps, the driving path error caused by the geometric characteristics of the road is reduced.

[0057] A dynamic road toll collection system based on vehicle type and vehicle travel path, the system includes a database storage module, a camera, a sensor inside the vehicle, a computing platform, an image processing and recognition module, a path calculation module and a 5G network;

[0058] The database storage module provides a stable data management platform for the entire system, which is used to store and manage the pre-entered information of the vehicle, including relevant data such as vehicle color, vehicle type, license plate number, etc.;

[0059] The camera is responsible for capturing the target vehicle passing by and using a high-resolution camera to provide vehicle images. The vehicle images provide a basis for subsequent license plate recognition and path calculation. The vehicle images are transmitted to the back-end computing platform via the 5G network.

[0060] The image processing and recognition module uses automatic license plate recognition (ANPR) and optical character recognition technology to process the image transmitted to the computing platform to identify the license plate number and vehicle type of the target vehicle;

[0061] The path calculation module uses the known camera position and license plate recognition information, combined with navigation information, to calculate the vehicle's precise mileage through a dynamic path correction method;

[0062] The computing platform receives information from the database storage module, cameras, and sensors, and performs path calculation, vehicle information comparison, and fee calculation to ensure real-time performance and data accuracy;

[0063] The 5G network ensures real-time data transmission between various system components, and ensures efficient operation of the system through low-latency, high-bandwidth data transmission between high-speed wireless networks, cameras and computing platforms.

[0064] Beneficial effects of the present invention:

[0065] The system and method realize dynamic charging based on vehicle type and actual mileage, effectively solving the problem that electric vehicles cannot fairly pay road use fees and the problem of insufficient road use charging rates, and effectively alleviates the shortage of funds for highway maintenance. The entire system realizes automated management from image capture, vehicle identification, driving path calculation and final fee settlement when the vehicle passes through the toll station.

[0066] The present invention can effectively track the driving path of the vehicle and calculate the mileage, and charge fairly according to the actual path of the vehicle, thereby ensuring the fairness of charging for electric vehicles and fuel vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1It is a schematic diagram of a dynamic road toll charging method based on vehicle type and vehicle travel path provided by the present invention.

[0068] Figure 2 It is a schematic diagram of a dynamic road toll collection system based on vehicle type and vehicle travel path provided by the present invention.

[0069] Figure 3 It is a schematic diagram of the road camera structure of the present invention. DETAILED DESCRIPTION

[0070] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0071] like Figure 1-Figure 3 As shown, a dynamic road toll collection method based on vehicle type and vehicle travel path includes the following steps:

[0072] Step 1: The owner of the target vehicle enters the vehicle information into the database in advance;

[0073] Step 2: Collect relevant information of the target vehicle at the entrance of the toll road. After the camera at the entrance captures the vehicle image, it is uploaded to the back-end processing system through the 5G network and communication line. The system uses automatic license plate recognition technology (ANPR) and optical character recognition technology (OCR), combined with the license plate color recognition method, to accurately identify and process the target vehicle's license plate number;

[0074] Step 3: Based on the cameras installed at important nodes inside the toll road, the cameras capture the vehicle image in real time when the target vehicle passes by, and calculate the vehicle's driving path in combination with the known camera information;

[0075] Step 4: The target vehicle's driving path is calculated, and the system calculates the fee based on the pre-set charging standards (unit: yuan / km) for different vehicle types, i.e. new energy vehicles, fuel vehicles, and trucks. Finally, the system deducts the fee in real time through the on-board payment device (such as ETC device) or mobile payment platform (such as Alipay, WeChat Pay).

[0076] In step 1, the vehicle information includes vehicle color, license plate number, vehicle type and vehicle energy type.

[0077] In step 2, the relevant information includes target vehicle information.

[0078] After the target vehicle is captured by a high-resolution image at the entrance of the road, the automatic license plate recognition technology (ANPR) uses the YOLO model to process the image. The model divides the image into grids and performs border prediction and classification regression on each grid to accurately segment the license plate image and license plate type;

[0079] Optical character recognition (OCR) technology recognizes the segmented license plate images. OCR technology uses a deep learning model that combines convolutional neural network (CNN) and recurrent neural network (RNN) to match and classify license plate images and extract the digital and letter information on the license plate. It uses a multi-layer convolutional neural network to extract features in the image, and then uses a recurrent neural network to process and identify the contextual relationship between characters to ensure the correctness of the character order.

[0080] The vehicle information pre-entered in the database is matched with the collected information to be judged. When the identified license plate number and license plate color of the target vehicle are consistent with the entered vehicle information, the road entrance is released.

[0081] Furthermore, the specific steps of the automatic license plate recognition technology (ANPR) using the YOLO model to process the high-resolution vehicle image captured by the camera are as follows:

[0082] The YOLO model first normalizes the input image, adjusts the pixel values ​​to the [0, 1] interval, and uniformly scales it to a fixed size, such as 416×416 pixels;

[0083] Subsequently, the model divides the image into S×S grids, for example, 19×19. Each grid is responsible for detecting the location of the license plate. When the center point of the license plate falls into a certain grid, the grid is responsible for predicting the bounding box of the license plate. The model predicts B bounding boxes in each grid by regression. The prediction parameters of each bounding box include the center point position (b x ,b y )、width and height(b w ,b h ), confidence score c Each grid also outputs a class probability vector P (Class i );

[0084] Where (b x ,b y ) represents the normalized coordinates of the center of the bounding box relative to the grid cell, (b w ,b h ) represents the normalized ratio of the width and height of the bounding box to the image size, P(Class i ) represents the predicted probability of each category, i∈{1,2,3} corresponds to the categories of green, blue and white license plates. The category probability is output by the classification head of the model and normalized by the Softmax activation function.

[0085] In the inference stage, the model determines whether the license plate exists and the degree of overlap between the predicted box and the true box by calculating the confidence score c of each grid. The calculation formula is:

[0086] c=P(Object)×IoU pred,truth

[0087] Among them, P(Object) represents the probability that the target object, i.e., the license plate, exists in the grid, and IoU pred,truth It is the intersection-over-union ratio of the predicted bounding box and the true bounding box.

[0088] The YOLO model uses the non-maximum suppression (NMS) algorithm to remove redundant boxes and retain only the box with the highest confidence as the final detection result, i.e., the license plate image. The category label of each detection box is obtained by selecting the maximum category probability P (Class i ) is determined by the corresponding category.

[0089] During the training process, the YOLO model optimizes the network parameters by minimizing the loss function, which consists of three parts: position loss, confidence loss, and classification loss. The position loss is used to measure the gap between the prediction and the real box, and is calculated using the mean square error (MSE). The formula is:

[0090]

[0091] in, and are the coordinate parameters of the predicted box and the real box, respectively, coord is the weighting coefficient. The confidence loss is usually calculated by the difference between the predicted confidence and the true confidence, and the formula is:

[0092]

[0093] Among them, c pred and c true are the predicted confidence and the true confidence values ​​respectively. The classification loss usually uses the cross entropy loss function to measure the difference between the predicted and true categories of the license plate category. The formula is:

[0094]

[0095] Among them, y i is the one-hot encoding of the true category, P(Class i ) is the category probability output by the model. Finally, the total loss function of the YOLO model is:

[0096] Loss total =λ coord ·Loss loc +Loss conf +λ class ·Loss class

[0097] By minimizing the total loss function, the model is able to gradually optimize the accuracy of license plate location, confidence, and category prediction.

[0098] The specific steps of the optical character recognition technology (OCR) for recognizing the segmented license plate image are as follows:

[0099] The input image is a license plate image processed by automatic license plate recognition technology. After normalization, it is input into a deep learning model that combines convolutional neural network (CNN) and recurrent neural network (RNN) to complete the process of feature extraction, sequence modeling and character recognition in sequence.

[0100] First, the input license plate image is recorded as Where H and W represent the height and width of the image, respectively, and C is the number of channels. The convolutional neural network CNN performs multi-layer convolution and pooling operations on the image to extract local and global features from the license plate image. The feature map of each layer of convolution is expressed as:

[0101] F (l) =σ(W (l) *F (l-1) +b (l) )

[0102] Among them, F (l-1) is the input feature map of the l-1th layer (X for the first layer), W (l) and b (l) are the weight and bias of the lth layer convolution kernel respectively, σ is a nonlinear activation function such as ReLU activation function. After multi-layer convolution and downsampling, a feature map F is generated, which contains important shape and structure information of the license plate characters;

[0103] Next, the feature map F is expanded into a feature vector {f1,f2,...,f T}, where each Represents the features of a time step. The sequence is input to a bidirectional recurrent neural network (Bi-RNN), such as a bidirectional long short-term memory network (Bi-LSTM) to capture the contextual dependencies between characters. The forward and backward calculations of Bi-LSTM at each time step t are:

[0104]

[0105] Ultimately, the output at each time step is represented as the concatenation of the forward and backward hidden states:

[0106]

[0107] In order to accurately identify the characters in the feature sequence and maintain their order, the connection temporal classification (CTC) loss function is used. CTC predicts the probability of the character sequence in an unaligned manner, specifically:

[0108]

[0109] Among them, S represents the target character sequence, s t is the predicted character at time step t, F is the input feature map, and the loss function of CTC is defined as:

[0110]

[0111] Where Alignments(S) is the set of all possible character alignments, and π is a specific alignment path. CTC sums all possible paths to ensure that model training does not require explicit alignment annotations;

[0112] Finally, combined with the beam search decoding algorithm, the output character probability distribution is converted into the final character sequence, successfully extracting the digital and letter information in the license plate image and segmenting the license plate number.

[0113] In the step 3, cameras are installed at important nodes inside the toll road, such as main line lane diversion points, service area entrances and exits, bridge entrances and exits, tunnel entrances and exits, ramp entrances and exits, dangerous sections (such as continuous curves, steep slopes), road intersections or emergency lanes every 10 km.

[0114] In step 3, when the target vehicle passes each camera, the system extracts the license plate number and the position of the vehicle in the image through automatic license plate recognition technology and optical character recognition technology. The geographical location and angle parameters of each camera installation are known in the system;

[0115] When a vehicle passes through multiple cameras, the system calculates the vehicle's driving path between different nodes by linking the positioning data between the cameras; in addition, the system will correct the camera node distance according to the curvature of the road and the actual driving trajectory of the vehicle, especially when there are bends or uphill and downhill roads. In this case, the system obtains the road's curvature radius, slope, and vertical and horizontal bending angles based on the digital map of the road. When the vehicle is driving on a bend, the system uses a curve distance calculation method. When driving on a ramp, the actual length of the path is obtained by expanding the Pythagoras theorem into a calculation method in three-dimensional space. The actual mileage of the target vehicle on the toll section is calculated by accumulating the distances between the path nodes.

[0116] In the above curved road section, the actual driving path length of the vehicle is determined by the road curvature radius R and the horizontal bending angle θ. According to the road geometry, the actual driving distance L of the vehicle on the curve is curve The calculation formula is:

[0117] L curve =R·θ

[0118] Where R is the radius of curvature of the road in meters, and θ is the horizontal curvature angle of the road in radians.

[0119] For the above slope section, the actual path length of the vehicle needs to take into account the vertical height change h and the horizontal projection length L of the road. horizontal Based on the calculation principle of path length in three-dimensional space, the actual driving path L of the vehicle slope By extending Pythagoras' theorem, we can get:

[0120]

[0121] Where h is the vertical height difference of the ramp, L horizontal is the horizontal distance of the ramp. By correcting the actual driving distance on curves and ramps, the driving path error caused by the geometric characteristics of the road is reduced.

[0122] A dynamic road toll collection system based on vehicle type and vehicle driving path. The toll collection system refers to the entire license plate recognition and path calculation automated management platform, which aims to achieve intelligent vehicle monitoring, path calculation and toll management. The system completes the entire process from vehicle information collection to toll settlement through the close integration of software, hardware and communication network;

[0123] Specifically, the system includes a database storage module, a camera, sensors inside the vehicle, a computing platform, an image processing and recognition module, a path calculation module, and a 5G network;

[0124] The database storage module provides a stable data management platform for the entire system, which is used to store and manage the pre-entered information of the vehicle, including relevant data such as vehicle color, vehicle type, license plate number, etc.;

[0125] When the camera captures, uploads and extracts the license plate number, the system will query the relevant data in the database in real time. When the license plate number used as an identifier is consistent with the vehicle information stored in the database, the vehicle can pass.

[0126] The camera is responsible for capturing the target vehicle passing by and using a high-resolution camera to provide vehicle images. The vehicle images provide a basis for subsequent license plate recognition and path calculation. The vehicle images are transmitted to the back-end computing platform via the 5G network.

[0127] The image processing and recognition module uses automatic license plate recognition (ANPR) and optical character recognition technology to process the image transmitted to the computing platform to identify the license plate number and vehicle type of the target vehicle;

[0128] The path calculation module uses the known camera positions and license plate recognition information, combined with navigation information, to calculate the vehicle's precise mileage through a dynamic path correction method, avoiding the calculation of straight-line distances based solely on the positioning data between cameras, thereby ensuring the accuracy of the vehicle's path calculation;

[0129] The computing platform receives information from the database storage module, the camera, and the sensor, and performs path calculation, vehicle information comparison, and fee calculation to ensure real-time performance and data accuracy.

[0130] The computing platform can be based on cloud servers or local servers, supports high-concurrency data processing, can quickly respond to the information collection of a large number of vehicles, and settle fees according to the actual mileage of different vehicle types according to the predetermined charging standards. The platform sends the settlement results and driving path information to the payment account pre-set by the car owner to realize automatic charging and settlement.

[0131] The 5G network ensures real-time data transmission between various system components, and ensures efficient operation of the system through low-latency, high-bandwidth data transmission between high-speed wireless networks, cameras and computing platforms.

[0132] Embodiment 1:

[0133] When the target vehicle arrives at the entrance of the toll road, the camera captures the vehicle image and collects the vehicle's information to be determined, namely the license plate number and vehicle type, through the image processing and recognition module. According to the vehicle information pre-entered in the system, when the information to be determined matches the pre-entered vehicle information, the vehicle passes. When the target vehicle passes through the toll road section, the camera installed at the important nodes inside the road captures the vehicle image and transmits it to the path calculation module through the communication line to calculate the actual path of the vehicle. When the vehicle reaches the exit of the toll road, the system calculates the target fee based on the calculated actual path of the vehicle and the pre-set charging standards for different vehicles, and deducts it through the on-board payment device or mobile payment platform.

[0134] Embodiment 2:

[0135] Example 2 shows a schematic diagram of a dynamic road toll collection system based on vehicle type and vehicle driving path proposed by the present invention. It is mainly divided into hardware equipment, namely cameras at road entrances and exits and important internal nodes; software equipment, namely the 5G network, a communication network that provides the entire system with high bandwidth and low latency efficiency; an image processing and recognition module that processes images captured by cameras to recognize vehicle license plate numbers and vehicle types; a path calculation module that calculates the actual driving path of the vehicle on the road; a data processing and calculation platform consisting of three modules: a database storage module that provides storage for pre-entered vehicle information, a matching module that matches the image processing and recognition module with the database storage information, and a fee calculation module that performs the final calculation.

[0136] Embodiment 3:

[0137] Example 3 shows a schematic diagram of the road camera structure after the target vehicle enters the road network.

[0138] When a vehicle enters the road network, it first passes through camera 1. The camera captures the vehicle image and transmits it to the image processing and recognition module for processing. Then the license plate number and the camera location are transmitted to the path calculation module. When the target vehicle passes by cameras 2, 3, and 4 in turn, the same steps are performed. When the vehicle finally reaches camera 5 and leaves the road network, the path calculation module will transmit all the returned camera information and the actual mileage of the calculated vehicle to the fee calculation part of the data processing and calculation platform. The platform calculates the vehicle toll based on the vehicle type (new energy vehicles, fuel vehicles, trucks, etc.) and the pre-set charging standards for different vehicle types, and then transmits the vehicle path information and fee information to the vehicle owner's bound vehicle payment device or mobile payment platform.

Claims

1. A dynamic road toll collection method based on vehicle type and vehicle travel path, characterized in that: The steps include: Step 1: The owner of the target vehicle enters the vehicle information into the database in advance; Step 2: Set up a camera at the entrance of the toll road to collect relevant information of the target vehicle. After the camera at the entrance captures the vehicle image, it is uploaded to the back-end processing system through the 5G network and communication line. The system uses automatic license plate recognition technology and optical character recognition technology, combined with license plate color recognition, to accurately identify and process the target vehicle's license plate number; Step 3: Cameras are installed at important nodes inside the toll road. The cameras capture vehicle images in real time when the target vehicle passes by, and the vehicle's driving path is calculated based on known camera information. Step 4: The target vehicle's driving path is calculated, and the system calculates the fee based on the pre-set charging standards for different vehicle types, namely new energy vehicles, fuel vehicles, and trucks. Finally, the system deducts the fee in real time through the on-board payment device or mobile payment platform.

2. A dynamic road toll collection method based on vehicle type and vehicle travel path according to claim 1, characterized in that: In step 1, the vehicle information includes vehicle color, license plate number, vehicle type and vehicle energy type.

3. A dynamic road toll collection method based on vehicle type and vehicle travel path according to claim 1, characterized in that: In the step 2, the relevant information includes target vehicle information; After the camera at the entrance of the road captures a high-resolution image of the target vehicle, the automatic license plate recognition technology uses the YOLO model to process the high-resolution image. The model divides the processed high-resolution image into grids and performs border prediction and classification regression on each grid to accurately segment the license plate image and license plate type. Optical character recognition technology recognizes the segmented license plate images. OCR technology uses a deep learning model that combines convolutional neural networks and recurrent neural networks to match and classify license plate images, extract the number and letter information on the license plate, use multi-layer convolutional neural networks to extract features in the image, and then use recurrent neural networks to process and recognize the contextual relationship between characters to ensure the correctness of the character order; The vehicle information pre-entered in the database is matched with the collected information to be judged. When the identified license plate number and license plate color of the target vehicle are consistent with the entered vehicle information, the road entrance is released.

4. A dynamic road toll collection method based on vehicle type and vehicle travel path according to claim 3, characterized in that: The specific steps of the automatic license plate recognition technology using the YOLO model to process the high-resolution vehicle image captured by the camera are as follows: The YOLO model first normalizes the input processed high-resolution image, adjusts the pixel values ​​to the [0,1] interval, and uniformly scales it to a fixed size; Subsequently, the model divides the processed high-resolution image into S×S grids, each of which is responsible for detecting the location of the license plate. When the center point of the license plate falls into a certain grid, the grid is responsible for predicting the bounding box of the license plate. The model predicts B bounding boxes within each grid by regression. The prediction parameters of each bounding box include the center point position (b x ,b y )、width and height(b w ,b h ), confidence score c, each grid also outputs a class probability vector P (Class i ); Where (b x ,b y ) represents the normalized coordinates of the center of the bounding box relative to the grid cell, (b w ,b h ) represents the normalized ratio of the width and height of the bounding box to the image size. The confidence score c is composed of two parts: one is the probability P(Object) of the existence of the target object, i.e., the license plate, and the other is the overlap between the predicted box and the real box using the intersection over union (IoU) pred,truth Indicates that P(Class i ) represents the predicted probability of each category, i∈{1,2,3} corresponds to the categories of green, blue and white license plates. The category probability is output by the classification head of the model and normalized by the Softmax activation function; In the above process, the model determines whether the license plate exists and the degree of overlap between the predicted box and the real box by calculating the confidence score c of each grid. The calculation formula is: c=P(Object)×IoU pred,truth Among them, P(Object) represents the probability that the target object, i.e., the license plate, exists in the grid, and IoU pred,truth It is the intersection-over-union ratio of the predicted bounding box and the true bounding box; a grid with a high confidence score indicates that there is a high possibility that a license plate exists. At the same time, if IoU pred,truth A higher value indicates that the predicted box and the true box overlap more, and the prediction is more accurate; Finally, the YOLO model uses the non-maximum suppression (NMS) algorithm to remove redundant boxes and retain only the box with the highest confidence as the final detection result, that is, the license plate image. The category label of each detection box is obtained by selecting the maximum category probability P (Class i ) is determined by the corresponding category; During the training process, the YOLO model optimizes the network parameters by minimizing the loss function, which consists of three parts: position loss, confidence loss, and classification loss. The position loss is used to measure the gap between the prediction and the real box, and is calculated using the mean square error (MSE). The formula is: in, and are the coordinate parameters of the predicted box and the real box, respectively, coord is a weighted coefficient. The confidence loss is usually calculated by the difference between the predicted confidence and the true confidence. The formula is: Among them, c pred and c true are the predicted confidence and the true confidence values ​​respectively. The classification loss usually uses the cross entropy loss function to measure the difference between the predicted and true categories of the license plate category. The formula is: Among them, y i is the one-hot encoding of the true category, P(Class i ) is the category probability output by the model. Finally, the total loss function of the YOLO model is: Loss total =λ coord ·Loss loc +Loss conf +λ class ·Loss class By minimizing the total loss function, the model is able to gradually optimize the accuracy of license plate position, confidence, and category prediction.

5. A dynamic road toll collection method based on vehicle type and vehicle travel path according to claim 4, characterized in that: The specific steps of the optical character recognition technology for identifying the segmented license plate image are as follows: The input image is a license plate image processed by automatic license plate recognition technology. After normalization, it is input into a deep learning model that combines convolutional neural network and recurrent neural network to complete the process of feature extraction, sequence modeling and character recognition in sequence. First, the input license plate image is recorded as Where H and W represent the height and width of the image, respectively, and C is the number of channels. The convolutional neural network CNN performs multi-layer convolution and pooling operations on the image to extract local and global features from the license plate image. The feature map of each layer of convolution is expressed as: F (l) =σ(W (l) *F (l-1) +b (l) ) Among them, F (l-1) is the input feature map of the l-1th layer (X for the first layer), W (l) and b (l) are the weight and bias of the lth layer convolution kernel respectively, σ is a nonlinear activation function such as ReLU activation function. After multi-layer convolution and downsampling, a feature map F is generated, which contains important shape and structure information of the license plate characters; Next, the feature map F is expanded into a feature vector {f1,f2,...,f T }, where each Represents the features of a time step. The sequence feature vector is input into a bidirectional recurrent neural network (Bi-RNN), such as a bidirectional long short-term memory network (Bi-LSTM) to capture the contextual dependencies between characters. The forward and backward calculations of Bi-LSTM at each time step t are: Ultimately, the output at each time step is represented as the concatenation of the forward and backward hidden states: In order to accurately identify the characters in the feature sequence and maintain their order, the connection temporal classification (CTC) loss function is used. CTC predicts the probability of the character sequence in an unaligned manner, specifically: Among them, S represents the target character sequence, s t is the predicted character at time step t, F is the input feature map, and the loss function of CTC is defined as: Where Alignments(S) is the set of all possible character alignments, and π is a specific alignment path. CTC sums all possible paths to ensure that model training does not require explicit alignment annotations; Finally, combined with the beam search decoding algorithm, the output character probability distribution is converted into the final character sequence, successfully extracting the digital and letter information in the license plate image and segmenting the license plate number.

6. A dynamic road toll collection method based on vehicle type and vehicle travel path according to claim 1, characterized in that: In the step 3, cameras are installed at important nodes inside the toll road, such as main lane diversion points, service area entrances and exits, bridge entrances and exits, tunnel entrances and exits, ramp entrances and exits, dangerous sections, road intersections or emergency lanes at intervals; When the target vehicle passes each camera, the system extracts the license plate number and the location of the vehicle in the image through automatic license plate recognition technology and optical character recognition technology. The geographical location and angle parameters of each camera installation are known in the system. When a vehicle passes through multiple cameras, the system calculates the vehicle's driving path between different nodes by linking the positioning data between the cameras; In addition, the system will correct the camera node distance according to the curvature of the road and the actual driving trajectory of the vehicle, especially when there are bends or uphill and downhill roads. In this case, the system obtains the road's curvature radius, slope, and vertical and horizontal bending angles based on the road's digital map. When the vehicle is driving on a bend, the system uses a curve distance calculation method. When driving on a ramp, the actual length of the path is obtained by expanding the Pythagoras theorem into a calculation method in three-dimensional space. The actual mileage of the target vehicle on the toll section is calculated by accumulating the distances between the path nodes.

7. A dynamic road toll collection method based on vehicle type and vehicle travel path according to claim 6, characterized in that: In the above curved road section, the actual driving path length of the vehicle is determined by the road curvature radius R and the horizontal bending angle θ. According to the road geometry, the actual driving distance L of the vehicle on the curved road is curve The calculation formula is: L curve =R·θ Where R represents the radius of curvature of the road, in meters, and θ is the horizontal curvature angle of the road, in radians; For the above slope section, the actual path length of the vehicle needs to take into account the vertical height change h and the horizontal projection length L of the road. horizontal Based on the calculation principle of path length in three-dimensional space, the actual driving path L of the vehicle slope By extending Pythagoras' theorem we get: Where h is the vertical height difference of the ramp, L horizontal is the horizontal distance of the ramp.

8. A dynamic road toll collection system based on vehicle type and vehicle travel path for implementing any one of the methods of claims 1 to 7, characterized in that: The system includes a database storage module, a camera, sensors inside the vehicle, a computing platform, an image processing and recognition module, a path calculation module, and a 5G network; The database storage module provides a stable data management platform for the entire system, which is used to store and manage the pre-entered information of vehicles, including vehicle color, vehicle type, and license plate number related data; The camera is responsible for capturing the target vehicle passing by and using a high-resolution camera to provide vehicle images. The vehicle images provide a basis for subsequent license plate recognition and path calculation. The vehicle images are transmitted to the back-end computing platform via the 5G network. The image processing and recognition module uses automatic license plate recognition technology and optical character recognition technology to process the image transmitted to the computing platform to recognize the license plate number and vehicle type of the target vehicle; The path calculation module uses the known camera position and license plate recognition information, combined with navigation information, to calculate the vehicle's precise mileage through a dynamic path correction method; The computing platform receives information from the database storage module, cameras, and sensors, and performs path calculation, vehicle information comparison, and fee calculation to ensure real-time performance and data accuracy; The 5G network ensures real-time data transmission between various system components, and ensures efficient operation of the system through low-latency, high-bandwidth data transmission between high-speed wireless networks, cameras and computing platforms.

Images