A highway traffic control method and system in a low-visibility environment
By building a high-reliable perception matrix and motion detection model, combining multiple algorithms for vehicle trajectory monitoring and early warning, the accuracy and efficiency of highway traffic control in low-visibility environments are solved, and intelligent control and safety warning are achieved.
Patent Information
- Application Number
- CN202411009212.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-26
AI Technical Summary
In a low visibility environment, it is difficult for the existing technology to effectively control highway traffic, resulting in traffic safety and traffic efficiency problems, mainly due to poor monitoring effects and hardware equipment limitations.
Using high-reliable perception matrix partition coding, motion detection and trajectory calculation prediction models, combined with edge detection and random forest algorithm, object detection algorithm, spectral clustering algorithm, and BP neural network algorithm, a low-visibility highway traffic control model is built, and real-time monitoring and control are carried out. Vehicle movement trajectory, providing emergency safety warning.
It improves the accuracy and accuracy of highway traffic control in low-visibility environments, optimizes resource utilization, and realizes intelligent control to adapt to traffic needs in different harsh environments.
Smart Images

Figure CN118736892B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent transportation, and in particular to a method and system for controlling highway traffic in a low-visibility environment. Background Art
[0002] With the continuous expansion of the transportation network and the continuous growth of the number of vehicles, the issue of highway traffic safety has received increasing attention from all walks of life. Among them, low-visibility environments, such as fog, rain, snow and other adverse weather conditions, pose a threat to highway traffic safety. Especially in low-visibility environments, the driver's vision is limited, the reaction time is prolonged, and accidents such as rear-end collisions and lane deviations are very likely to occur.
[0003] At present, speed limits and traffic restrictions will affect the efficiency of highway traffic and increase the risk of traffic congestion; and although the setting of warning signs can remind drivers to pay attention to safety to a certain extent, in the case of extremely low visibility, drivers may not be able to find warning signs in time, leading to accidents. Some progress has been made in the research on highway traffic control in low visibility environments at home and abroad.
[0004] However, the existing technologies still have some shortcomings. On the one hand, most technologies focus on monitoring and warning of traffic status, but lack in-depth research on traffic control; on the other hand, the existing technologies may be limited by hardware equipment during application, such as insufficient sensor accuracy and limited data processing capabilities, resulting in poor monitoring effects. Therefore, Kou is looking to develop a more effective and intelligent method for highway traffic control in low-visibility environments to improve highway traffic efficiency and safety. Summary of the invention
[0005] The purpose of the present invention is to provide a method for controlling highway traffic in a low-visibility environment.
[0006] To achieve the above object, the present invention is implemented according to the following technical solutions:
[0007] The present invention comprises the following steps:
[0008] Collecting traffic data and image data of a highway in a low-visibility environment, and preprocessing the traffic data and the image data;
[0009] Constructing a high-reliability perception matrix according to the traffic data, performing partition coding on the high-reliability perception matrix according to highway lanes to obtain regional coding, and performing positioning recognition on the image data to obtain positioning data;
[0010] According to the similarity, the perception network is used to perform motion detection on the area code to obtain motion data, and the positioning data is used to correct the motion data to obtain correction data, and the correction data is input into the trajectory calculation prediction model to obtain the driving motion trajectory; including:
[0011]
[0012] The sensing angle of the sth motion data is θ s , the number of positioning data is The measured angle of the sth positioning data is θ o , the sth motion data is k s , the sth impact factor is The sth motion data at the tth moment is k s,t , the sth motion data at the t+1th moment is k s,t+1 , the upper limit of the observation time is T, and the sth deviation weight is δ s , the sth positioning data is L s , the sth corrected data is ζ s , the regulatory factor is ε;
[0013] Edge detection is used to obtain vehicle-to-vehicle data according to the driving motion trajectory, a low-visibility highway traffic control model is constructed according to the vehicle-to-vehicle data, the data to be controlled is input into the low-visibility highway traffic control model, and the control result is output.
[0014] Furthermore, the method of performing partition coding on the high-reliability perception matrix according to highway lanes to obtain regional coding includes:
[0015] Divide the high-reliability perception matrix into multiple perception areas by the sensor locations on the road;
[0016] Calculate the upper limit of the error probability between the perception area and the coded data:
[0017]
[0018] The channel matrix is R, the precoding matrix is K, the error probability is H(·), the sensing area is C, and the coded data of the sensing area is The error rate is η, and the norm function is ‖·‖;
[0019] Calculate the channel probability density:
[0020]
[0021] The probability density of channel R is The channel mean is R τ , the transmission covariance is V t , the inverse transmission matrix is V -1 , the rank of the matrix is tr[·], the number of channel vectors is n, the determinant is dt(·), and the number of the f-th signal vector is n f ;
[0022] Compute the minimum objective function:
[0023]
[0024] tr(KK R )=1
[0025] The objective function is The inverse matrix of the transmission covariance is The significance weight is λ, and the precoding matrix of channel R is K R , when the precoding matrix is optimal, the code distance of the matrix is μ o ;
[0026] Calculate significance weights:
[0027]
[0028] The significant constants are l, the natural constant is e, and the average saliency of the current encoded image is The vth coding region U v The average significance of K(U v );
[0029] Given the constraint objective function, the expression is:
[0030]
[0031] The Lagrange coefficient is Channel adjustment matrix Z and Lagrangian coefficients The constraint objective function is The precoding matrix is
[0032] Calculate the optimal encoding matrix:
[0033]
[0034] The optimal encoding matrix is K o , adjust the parameters to The right singular vector is X, the left singular vector is Y, and the region code is output according to the optimal coding matrix.
[0035] Furthermore, the method of obtaining motion data by performing motion detection on the region coding using a perceptual network according to the similarity includes:
[0036] A regional image is obtained according to the coding region, and a convolutional neural network is used to extract the time-varying features of the regional image;
[0037] The time-varying features are input into the perception network, and the input feature layer channels are globally averaged to compress the spatial feature information into the channel dimension. The expression is:
[0038]
[0039] The ath statistic is The height of the region image is S, the width of the region image is D, and the ath feature layer is E a , the current position value of the feature map with the cth height and yth width is (c,y);
[0040] The correlation between channels is fitted through two full connection operations, and the importance of channels is determined based on the correlation. The expression is:
[0041]
[0042] The scaling parameter of the regional image is β, the translation parameter of the regional image is r1, the output weight is ψ, the natural constant is e, and the offset is δ;
[0043] Prune the perception network, the expression is:
[0044]
[0045] The output of the a-th batch standard layer is The input of the a-th batch norm layer is v a , the displacement is The scaling factor is ρ, and the mean of the smallest batch of training data is The variance of the smallest batch of training data is σ W , the error factor is τ;
[0046] The loss function of the improved perception network is expressed as:
[0047]
[0048] The regularization coefficient is α, and the feature representation of the bth network layer is The predicted features at the bth network layer are expressed as Displacement The constraint function is
[0049] The moving target is annotated through the perception network and the similarity of the moving target is calculated:
[0050]
[0051] The cosine weight is χ2, the shape weight is χ1, the cosine function is cos(·), and the i-th shape feature of the p-th moving target is zp,i , the pth moving target appearance feature z p,i The appearance characteristics of the control vehicle are z p,o , the adjustment factor is φ, and the influence degree is The number of shape features is
[0052] The moving targets with similarity greater than 0.718 are regarded as moving vehicles, and the motion data of the moving vehicles are extracted and output as motion data.
[0053] Furthermore, the method of inputting the correction data into a trajectory calculation prediction model to obtain a driving motion trajectory includes:
[0054] The dynamic model and statistical behavior pattern are learned from the corrected data, and a driving mode clustering mechanism based on spectral clustering is introduced. The feature weight distribution that affects the trajectory is obtained through the spatiotemporal attention mechanism. The vehicle's driving trajectory is predicted according to the dynamic model, behavior pattern and feature weight distribution.
[0055] Furthermore, the method of acquiring inter-vehicle data according to the driving motion trajectory by using edge detection includes:
[0056] The image is obtained according to the driving trajectory, and the expression is:
[0057]
[0058] The pixel coordinates of the image are (w, j), the grayscale image is G(w, j), the convolution operator is *, the standard deviation of the Gaussian distribution is ξ, and the deviation constant is The control factor is ω, the horizontal coordinate of the image pixel is w, and the vertical coordinate of the image pixel is j;
[0059] Count the maximum and minimum grayscale values of the image and calculate the initial threshold:
[0060]
[0061] The initial threshold is Q0, and the maximum grayscale value of the image is W max , the minimum gray value of the image is W min ;
[0062] The image with a value greater than the initial threshold is taken as the background, and the image with a value less than the initial threshold is taken as the foreground. The initial threshold is updated according to the foreground grayscale mean and the background grayscale mean. The expression is:
[0063]
[0064] The foreground grayscale mean is The background grayscale mean is The updated threshold is Q1;
[0065] According to the updated threshold, the image is segmented, the moving vehicle is located, and the gradient amplitude and phase of the moving vehicle in the image are calculated:
[0066]
[0067] The gradient amplitude is Qr, the phase is u, and the horizontal component of the image gradient is Qr w , the vertical component of the image gradient is Qr j ;
[0068] The distance between the front and rear vehicles in the lane where the vehicle is traveling is calculated based on the gradient amplitude and phase, and the distance between the vehicles is output as inter-vehicle data.
[0069] Furthermore, a method for constructing a low-visibility highway traffic control model based on the vehicle-to-vehicle data includes:
[0070] Compare the distance between vehicles in the vehicle-to-vehicle data with the spacing threshold. When the distance between the front and rear vehicles in the vehicle's driving lane is less than the threshold, an emergency safety warning is issued.
[0071] The low-visibility highway traffic control model includes random forest algorithm, target detection algorithm, spectral clustering algorithm, and BP neural network algorithm;
[0072] The random forest algorithm randomly divides the input data into training data and test data in a ratio of 7:2;
[0073] The target detection algorithm automatically extracts image features from the training data by training a convolutional neural network, and uses the image features to identify and locate the target vehicle in the image;
[0074] The spectral clustering algorithm uses the target vehicle's driving habit similarity matrix to construct the graph's Laplacian matrix, and clusters the driving feature vectors to obtain habit clusters;
[0075] In low-visibility highway traffic monitoring, the BP neural network algorithm learns the data patterns between vehicles, performs nonlinear mapping on habit clustering, and identifies and predicts traffic conditions for traffic monitoring;
[0076] The test data is input into the low-visibility highway traffic control model to obtain the predicted vehicle-to-vehicle data. The hyperparameters of the low-visibility highway traffic control model are adjusted according to the mean square error between the predicted vehicle-to-vehicle data and the actual predicted vehicle-to-vehicle data until the mean square error is less than 0.11.
[0077] In the second aspect, a highway traffic control system in a low visibility environment includes:
[0078] Data acquisition module: used to collect traffic data and image data of roads in low visibility environments, and pre-process the traffic data and image data;
[0079] Data processing module: used for constructing a high-reliability perception matrix according to the traffic data, performing partition coding on the high-reliability perception matrix according to highway lanes to obtain regional coding, and performing positioning identification on the image data to obtain positioning data;
[0080] Trajectory calculation module: used to perform motion detection on the area coding using a perception network according to similarity to obtain motion data, then use the positioning data to correct the motion data to obtain correction data, and input the correction data into a trajectory calculation prediction model to obtain a driving motion trajectory; including:
[0081]
[0082] The sensing angle of the sth motion data is θ s , the number of positioning data is The measured angle of the sth positioning data is θ o , the sth motion data is k s , the sth impact factor is The sth motion data at the tth moment is k s,t , the sth motion data at the t+1th moment is k s,t+1 , the upper limit of the observation time is T, and the sth deviation weight is δ s , the sth positioning data is L s , the sth corrected data is ζ s , the regulatory factor is ε;
[0083] Vehicle control module: used to obtain vehicle-to-vehicle data according to the driving trajectory using edge detection, build a low-visibility highway traffic control model based on the vehicle-to-vehicle data, input the data to be controlled into the low-visibility highway traffic control model, and output the control results.
[0084] The beneficial effects of the present invention are:
[0085] The present invention is a method and system for controlling highway traffic in a low-visibility environment. Compared with the prior art, the present invention has the following technical effects:
[0086] The present invention can improve the accuracy of highway traffic control in low-visibility environments through preprocessing, partition coding, positioning identification, motion detection, data correction, trajectory calculation, edge detection and model construction steps, thereby improving the precision of highway traffic control in low-visibility environments, and optimizing the positioning of highway traffic control in low-visibility environments, which can greatly save resources and improve work efficiency, and can realize intelligent control of highway traffic in low-visibility environments, and perform positioning correction and trajectory calculation for highway traffic control in low-visibility environments in real time, which is of great significance to highway traffic control in low-visibility environments, can adapt to highway traffic control in multiple low-visibility environments of different standards, and highway traffic control in different low-visibility environments, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS
[0087] Figure 1 The present invention is a flowchart of the steps of a method for controlling highway traffic in a low-visibility environment. DETAILED DESCRIPTION
[0088] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.
[0089] The present invention provides a method and system for controlling highway traffic in a low-visibility environment, comprising the following steps:
[0090] like Figure 1 As shown, in this embodiment, the following steps are included:
[0091] Collecting traffic data and image data of a highway in a low-visibility environment, and preprocessing the traffic data and the image data;
[0092] In the actual evaluation, traffic data includes vehicle type, location, speed, visibility, driving motion trajectory, traffic flow, and road conditions;
[0093] Constructing a high-reliability perception matrix according to the traffic data, performing partition coding on the high-reliability perception matrix according to highway lanes to obtain regional coding, and performing positioning recognition on the image data to obtain positioning data;
[0094] In the actual evaluation, two types of sensors are used to build a high-reliability perception matrix for highway traffic in low-visibility environments. One type is non-contact sensors, such as video, microwave radar, lidar and other sensors deployed along the highway; the other type is contact sensors, such as fiber grating sensors, quartz coil pressure sensors and other sensors embedded in the road structure. The way to build a high-reliability perception matrix for highway traffic in low-visibility environments is to divide each lane of the highway into perception areas through the continuous deployment of sensors on the highway, and perform regional coding on the perception areas;
[0095] In the on-site highway environment, according to the sensing characteristics of various sensors, embedded contact sensors such as fiber grating sensors and quartz coil pressure sensors are deployed, or non-contact sensors such as video, microwave radar, and lidar are deployed, so that the sensor's sensing range covers the entire road environment. The constructed sensor network can detect the motion data of traffic participants in the highway environment; the motion data includes the type of traffic participants, movement speed, movement trajectory and position; the on-site highway environment is position-coded according to lanes and driving directions, and a perception matrix is constructed, with each area given a unique area code;
[0096] The high-reliability perception matrix coding setting method for highway traffic takes the driving direction of the rightmost lane of the highway as the coding starting direction, and the coding starting point value is K01+001. According to the driving direction of the lane, the coding of the next perception matrix area is K01+002. When it exceeds 1km, the coding changes to K11+001 according to the kilometer number, and so on; when encoding the second lane, the coding starting point value is K02+001. According to the driving direction of the lane, the coding of the next perception matrix area is K02+002. When it exceeds 1km, the coding changes to K12+001 according to the kilometer number, and so on, and the coding is set for each lane of the highway;
[0097] According to the similarity, the perception network is used to perform motion detection on the area code to obtain motion data, and the positioning data is used to correct the motion data to obtain correction data, and the correction data is input into the trajectory calculation prediction model to obtain the driving motion trajectory; including:
[0098]
[0099] The sensing angle of the sth motion data is θ s , the number of positioning data is The measured angle of the sth positioning data is θ o , the sth motion data is k s , the sth impact factor is The sth motion data at the tth moment is k s,t , the sth motion data at the t+1th moment is k s,t+1 , the upper limit of the observation time is T, and the sth deviation weight is δ s , the sth positioning data is L s , the sth corrected data is ζ s , the regulatory factor is ε;
[0100] In the actual evaluation, whether there is a moving vehicle in the coded perception matrix area; when it is judged that there is a moving vehicle in the perception matrix area, the adjacent coded perception matrix areas are arranged and calculated according to time, the moving trajectory of the moving vehicle on the entire highway is calculated, and the moving speed of the moving vehicle is inferred; when the perception matrix of the entire highway judges that there are multiple moving vehicles, according to the encoding method of the perception matrix, the moving speed and distance of two adjacent traffic participants are inferred, and the horizontal and vertical information of multiple traffic participants in the perception matrix is established;
[0101] Using edge detection to obtain inter-vehicle data according to the driving motion trajectory, constructing a low-visibility highway traffic control model according to the inter-vehicle data, inputting the data to be controlled into the low-visibility highway traffic control model, and outputting the control result;
[0102] In the actual evaluation, the traffic operation status in low visibility environment is evaluated:
[0103] 1) The lane in which the vehicle is traveling;
[0104] 2) Whether there are other vehicles in front of or behind the lane where the vehicle is traveling;
[0105] 3) The speed of the vehicles in front and behind the lane where the vehicle is traveling, and the distance between the vehicle and the vehicle;
[0106] 4) Whether there are other vehicles in front of or behind the adjacent lanes of the vehicle's lane;
[0107] 5) The speed of vehicles in the lanes adjacent to the vehicle’s lane and the distance between the vehicle and the vehicle;
[0108] 6) When the distance between the vehicle in front and behind the vehicle in the lane is less than the threshold, an emergency safety warning is issued.
[0109] In this embodiment, the method of performing partition coding on the high-reliability perception matrix according to highway lanes to obtain area coding includes:
[0110] Divide the high-reliability perception matrix into multiple perception areas by the sensor locations on the road;
[0111] Calculate the upper limit of the error probability between the perception area and the coded data:
[0112]
[0113] The channel matrix is R, the precoding matrix is K, the error probability is H(·), the sensing area is C, and the coded data of the sensing area is The error rate is η, and the norm function is ‖·‖;
[0114] Calculate the channel probability density:
[0115]
[0116] The probability density of channel R is The channel mean is R τ , the transmission covariance is V t , the inverse transmission matrix is V -1 , the rank of the matrix is tr[·], the number of channel vectors is n, the determinant is dt(·), and the number of the f-th signal vector is n f ;
[0117] Compute the minimum objective function:
[0118]
[0119] tr(KK R )=1
[0120] The objective function is The inverse matrix of the transmission covariance is The significance weight is λ, and the precoding matrix of channel R is K R , when the precoding matrix is optimal, the code distance of the matrix is μ o ;
[0121] Calculate significance weights:
[0122]
[0123] The significant constants are l, the natural constant is e, and the average saliency of the current encoded image is The vth coding region U v The average significance of K(U v );
[0124] Given the constraint objective function, the expression is:
[0125]
[0126] The Lagrange coefficient is Channel adjustment matrix Z and Lagrangian coefficients The constraint objective function is The precoding matrix is
[0127] Calculate the optimal encoding matrix:
[0128]
[0129] The optimal encoding matrix is K o , adjust the parameters to The right singular vector is X, the left singular vector is Y, and the region code is output according to the optimal coding matrix.
[0130] In this embodiment, the method of obtaining motion data by performing motion detection on the region coding using a perceptual network according to the similarity includes:
[0131] A regional image is obtained according to the coding region, and a convolutional neural network is used to extract the time-varying features of the regional image;
[0132] The time-varying features are input into the perception network, and the input feature layer channels are globally averaged to compress the spatial feature information into the channel dimension. The expression is:
[0133]
[0134] The ath statistic is The height of the region image is S, the width of the region image is D, and the ath feature layer is E a , the current position value of the feature map with the cth height and yth width is (c,y);
[0135] The correlation between channels is fitted through two full connection operations, and the importance of channels is determined based on the correlation. The expression is:
[0136]
[0137] The scaling parameter of the regional image is β, the translation parameter of the regional image is r1, the output weight is ψ, the natural constant is e, and the offset is δ;
[0138] Prune the perception network, the expression is:
[0139]
[0140] The output of the a-th batch standard layer is The input of the a-th batch norm layer is v a , the displacement is The scaling factor is ρ, and the mean of the smallest batch of training data is The variance of the smallest batch of training data is σ W , the error factor is τ;
[0141] The loss function of the improved perception network is expressed as:
[0142]
[0143] The regularization coefficient is α, and the feature representation of the bth network layer is The predicted features at the bth network layer are expressed as Displacement The constraint function is
[0144] The moving target is annotated through the perception network and the similarity of the moving target is calculated:
[0145]
[0146] The cosine weight is χ2, the shape weight is χ1, the cosine function is cos(·), and the i-th shape feature of the p-th moving target is z p,i , the pth moving target appearance feature z p,i The appearance characteristics of the control vehicle are z p,o , the adjustment factor is φ, and the influence degree is The number of shape features is
[0147] The moving targets with similarity greater than 0.718 are regarded as moving vehicles, and the motion data of the moving vehicles are extracted and output as motion data.
[0148] In this embodiment, the method of inputting the correction data into the trajectory calculation prediction model to obtain the driving motion trajectory includes:
[0149] The dynamic model and statistical behavior pattern are learned from the corrected data, and a driving mode clustering mechanism based on spectral clustering is introduced. The feature weight distribution that affects the trajectory is obtained through the spatiotemporal attention mechanism. The vehicle's driving trajectory is predicted according to the dynamic model, behavior pattern and feature weight distribution.
[0150] In this embodiment, the method of acquiring vehicle-to-vehicle data according to the driving motion trajectory using edge detection includes:
[0151] The image is obtained according to the driving trajectory, and the expression is:
[0152]
[0153] The pixel coordinates of the image are (w, j), the grayscale image is G(w, j), the convolution operator is *, the standard deviation of the Gaussian distribution is ξ, and the deviation constant is The control factor is ω, the horizontal coordinate of the image pixel is w, and the vertical coordinate of the image pixel is j;
[0154] Count the maximum and minimum grayscale values of the image and calculate the initial threshold:
[0155]
[0156] The initial threshold is Q0, and the maximum grayscale value of the image is W max , the minimum gray value of the image is W min ;
[0157] The image with a value greater than the initial threshold is taken as the background, and the image with a value less than the initial threshold is taken as the foreground. The initial threshold is updated according to the foreground grayscale mean and the background grayscale mean. The expression is:
[0158]
[0159] The foreground grayscale mean is The background grayscale mean is The updated threshold is Q1;
[0160] According to the updated threshold, the image is segmented, the moving vehicle is located, and the gradient amplitude and phase of the moving vehicle in the image are calculated:
[0161]
[0162] The gradient amplitude is Qr, the phase is u, and the horizontal component of the image gradient is Qr w , the vertical component of the image gradient is Qr j ;
[0163] The distance between the front and rear vehicles in the lane where the vehicle is traveling is calculated based on the gradient amplitude and phase, and the distance between the vehicles is output as inter-vehicle data.
[0164] In this embodiment, the method for constructing a low-visibility highway traffic control model based on the vehicle-to-vehicle data includes:
[0165] Compare the distance between vehicles in the vehicle-to-vehicle data with the spacing threshold. When the distance between the front and rear vehicles in the vehicle's driving lane is less than the threshold, an emergency safety warning is issued.
[0166] The low-visibility highway traffic control model includes random forest algorithm, target detection algorithm, spectral clustering algorithm, and BP neural network algorithm;
[0167] The random forest algorithm randomly divides the input data into training data and test data in a ratio of 7:2;
[0168] The target detection algorithm automatically extracts image features from the training data by training a convolutional neural network, and uses the image features to identify and locate the target vehicle in the image;
[0169] The spectral clustering algorithm uses the target vehicle's driving habit similarity matrix to construct the graph's Laplacian matrix, and clusters the driving feature vectors to obtain habit clusters;
[0170] In low-visibility highway traffic monitoring, the BP neural network algorithm learns the data patterns between vehicles, performs nonlinear mapping on habit clustering, and identifies and predicts traffic conditions for traffic monitoring;
[0171] The test data is input into the low-visibility highway traffic control model to obtain the predicted vehicle-to-vehicle data. The hyperparameters of the low-visibility highway traffic control model are adjusted according to the mean square error between the predicted vehicle-to-vehicle data and the actual predicted vehicle-to-vehicle data until the mean square error is less than 0.11.
[0172] In the second aspect, a highway traffic control system in a low visibility environment includes:
[0173] Data acquisition module: used to collect traffic data and image data of roads in low visibility environments, and pre-process the traffic data and image data;
[0174] Data processing module: used for constructing a high-reliability perception matrix according to the traffic data, performing partition coding on the high-reliability perception matrix according to highway lanes to obtain regional coding, and performing positioning identification on the image data to obtain positioning data;
[0175] Trajectory calculation module: used to perform motion detection on the area coding using a perception network according to similarity to obtain motion data, then use the positioning data to correct the motion data to obtain correction data, and input the correction data into a trajectory calculation prediction model to obtain a driving motion trajectory; including:
[0176]
[0177] The sensing angle of the sth motion data is θ s , the number of positioning data is The measured angle of the sth positioning data is θ o , the sth motion data is k s , the sth impact factor is The sth motion data at the tth moment is k s,t , the sth motion data at the t+1th moment is k s,t+1 , the upper limit of the observation time is T, and the sth deviation weight is δ s , the sth positioning data is L s , the sth corrected data is ζ s , the regulatory factor is ε;
[0178] Vehicle control module: used to obtain vehicle-to-vehicle data according to the driving trajectory using edge detection, build a low-visibility highway traffic control model based on the vehicle-to-vehicle data, input the data to be controlled into the low-visibility highway traffic control model, and output the control results.
Claims
1. A method for controlling highway traffic in a low-visibility environment, characterized in that: The following steps are involved: Collecting traffic data and image data of a highway in a low-visibility environment, and preprocessing the traffic data and the image data; Constructing a high-reliability perception matrix according to the traffic data, performing partition coding on the high-reliability perception matrix according to highway lanes to obtain regional coding, and performing positioning recognition on the image data to obtain positioning data; According to the similarity, a perception network is used to perform motion detection on the area code to obtain motion data, the positioning data is used to correct the motion data to obtain correction data, and the correction data is input into a trajectory calculation prediction model to obtain a driving motion trajectory; include: The sensing angle of the sth motion data is , the number of positioning data is , the measured angle of the sth positioning data is , the sth motion data is , No. The impact factor is , the sth motion data at the tth moment is , the sth motion data at the t+1th moment is , the upper limit of the observation time is T, and the s-th deviation weight is , the sth positioning data is , the sth correction data is The regulatory factor is ; The method comprises: obtaining inter-vehicle data according to the driving motion trajectory by using edge detection, constructing a low-visibility highway traffic control model according to the inter-vehicle data, inputting the data to be controlled into the low-visibility highway traffic control model, and outputting the control result; comprising: Compare the distance between vehicles in the vehicle-to-vehicle data with the spacing threshold. When the distance between the front and rear vehicles in the vehicle's driving lane is less than the threshold, an emergency safety warning is issued. The low-visibility highway traffic control model includes random forest algorithm, target detection algorithm, spectral clustering algorithm, and BP neural network algorithm; The random forest algorithm randomly divides the input data into training data and test data in a ratio of 7:2; The target detection algorithm automatically extracts image features from the training data by training a convolutional neural network, and uses the image features to identify and locate the target vehicle in the image; The spectral clustering algorithm uses the target vehicle's driving habit similarity matrix to construct the graph's Laplacian matrix, and clusters the driving feature vectors to obtain habit clusters; In low-visibility highway traffic monitoring, the BP neural network algorithm learns the data patterns between vehicles, performs nonlinear mapping on habit clustering, and identifies and predicts traffic conditions for traffic monitoring; The test data is input into the low-visibility highway traffic control model to obtain the predicted vehicle-to-vehicle data. The hyperparameters of the low-visibility highway traffic control model are adjusted according to the mean square error between the predicted vehicle-to-vehicle data and the actual predicted vehicle-to-vehicle data until the mean square error is less than 0.
11.
2. A method for controlling highway traffic in a low-visibility environment according to claim 1, characterized in that: The method of performing partition coding on the high-reliability perception matrix according to highway lanes to obtain area coding comprises: Divide the high-reliability perception matrix into multiple perception areas by the sensor locations on the road; Calculate the upper limit of the error probability between the perception area and the coded data: The channel matrix is R, the precoding matrix is K, and the error probability is , the sensing area is , the encoding data of the perception area is , the error rate is , the norm function is ; Calculate the channel probability density: The probability density of channel R is , the channel mean is , the transmission covariance is , the inverse transmission matrix is , the rank of the matrix is , the number of channel vectors is n, and the determinant is , the number of the fth signal vector is ; Compute the minimum objective function: The objective function is , the inverse matrix of the transmission covariance is , the significance weight is , the precoding matrix of channel R is , when the precoding matrix is optimal, the code distance of the matrix is ; Calculate significance weights: ; The significant constants are , , , the natural constant is e, and the average saliency of the current encoded image is , the vth coding region The average significance of ; Given the constraint objective function, the expression is: The Lagrange coefficient is , channel adjustment matrix Z and Lagrangian coefficients The constraint objective function is , the precoding matrix is ; Calculate the optimal encoding matrix: The optimal encoding matrix is , adjust the parameters to , the right singular vector is X, the left singular vector is Y, and the region code is output according to the optimal coding matrix.
3. The method for controlling highway traffic in a low-visibility environment according to claim 1, characterized in that: The method of obtaining motion data by performing motion detection on the region encoding using a perception network according to the similarity includes: A regional image is obtained according to the coding region, and a convolutional neural network is used to extract the time-varying features of the regional image; The time-varying features are input into the perception network, and the input feature layer channels are globally averaged to compress the spatial feature information into the channel dimension. The expression is: The ath statistic is , the height of the region image is S, the width of the region image is D, and the ath feature layer is , the current position value of the feature map with the cth height and yth width is ; The correlation between channels is fitted through two full connection operations, and the importance of channels is determined based on the correlation. The expression is: The scaling parameter of the region image is , the translation parameter of the regional image is , the output weight is , the natural constant is e, and the offset is ; Prune the perception network, the expression is: The output of the a-th batch standard layer is , the input of the a-th batch norm layer is , the displacement is , the scaling factor is , the mean of the smallest batch of training data is , the variance of the smallest batch of training data is , the error factor is ; The loss function of the improved perception network is expressed as: The regularity coefficient is , the feature representation of the bth network layer is , the predicted features at the bth network layer are expressed as , displacement The constraint function is ; The moving targets are labeled through the perception network and the similarity of the moving targets is calculated: The cosine weight is , the shape weight is , the cosine function is , the i-th appearance feature of the p-th moving target is , the shape feature of the pth moving target The appearance characteristics of the control vehicle are , the adjustment factor is , the impact degree is , the number of appearance features is ; The moving targets with similarity greater than 0.718 are regarded as moving vehicles, and the motion data of the moving vehicles are extracted and output as motion data.
4. The method for controlling highway traffic in a low-visibility environment according to claim 1, characterized in that: The method of inputting the correction data into a trajectory calculation prediction model to obtain a driving motion trajectory comprises: The dynamic model and statistical behavior pattern are learned from the corrected data, and a driving mode clustering mechanism based on spectral clustering is introduced. The feature weight distribution that affects the trajectory is obtained through the spatiotemporal attention mechanism. The vehicle's driving trajectory is predicted according to the dynamic model, behavior pattern and feature weight distribution.
5. The method for controlling highway traffic in a low-visibility environment according to claim 1, characterized in that: The method for acquiring inter-vehicle data according to the driving motion trajectory by using edge detection comprises: The image is obtained according to the driving trajectory, and the expression is: The pixel coordinates of the image are , the grayscale image is , the convolution operator is , the standard deviation of the Gaussian distribution is , the deviation constant is , the control factor is , the horizontal coordinate of the image pixel is , the vertical coordinate of the image pixel is j; Count the maximum and minimum grayscale values of the image and calculate the initial threshold: The initial threshold is , the maximum gray value of the image is , the minimum gray value of the image is ; The image with a value greater than the initial threshold is taken as the background, and the image with a value less than the initial threshold is taken as the foreground. The initial threshold is updated according to the foreground grayscale mean and the background grayscale mean. The expression is: The foreground grayscale mean is , the background gray mean is , the updated threshold is ; Segment the image according to the updated threshold, locate the moving vehicle, and calculate the gradient amplitude and phase of the moving vehicle in the image: The gradient amplitude is , the phase is , the horizontal component of the image gradient is , the vertical component of the image gradient is ; The distance between the front and rear vehicles in the lane where the vehicle is traveling is calculated based on the gradient amplitude and phase, and the distance between the vehicles is output as inter-vehicle data.
6. A highway traffic control system in a low-visibility environment, used to execute the method according to any one of claims 1 to 5, characterized in that: include: Data acquisition module: used to collect traffic data and image data of roads in low visibility environments, and pre-process the traffic data and image data; Data processing module: used for constructing a high-reliability perception matrix according to the traffic data, performing partition coding on the high-reliability perception matrix according to highway lanes to obtain regional coding, and performing positioning identification on the image data to obtain positioning data; Trajectory calculation module: used for performing motion detection on the area coding using a perception network according to similarity to obtain motion data, correcting the motion data using the positioning data to obtain correction data, and inputting the correction data into a trajectory calculation prediction model to obtain a driving motion trajectory; include: The sensing angle of the sth motion data is , the number of positioning data is , the measured angle of the sth positioning data is , the sth motion data is , No. The impact factor is , the sth motion data at the tth moment is , the sth motion data at the t+1th moment is , the upper limit of the observation time is T, and the s-th deviation weight is , the sth positioning data is , the sth correction data is The regulatory factor is ; Vehicle control module: used to obtain vehicle-to-vehicle data according to the driving trajectory using edge detection, build a low-visibility highway traffic control model based on the vehicle-to-vehicle data, input the data to be controlled into the low-visibility highway traffic control model, and output the control results.
7. An electronic device comprising: processor; as well as A memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the method of any one of claims 1 to 5.
8. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, enables the electronic device to execute the method according to any one of claims 1 to 5.
Citation Information
Patent Citations
Multiple object collision avoidance based on centralized coordination of vehicle operations
US20220051566A1
Scene flow digital twin method and system based on dynamic trajectory flow
WO2023207437A1
