Emergency lane dynamic management method and system

Abstract

The invention provides an emergency lane dynamic management method and system, and the method comprises the steps: obtaining historical monitoring videos of all monitoring points, and carrying out the video processing of the historical monitoring videos; extracting vehicle information in the processed video, and redefining and calculating traffic flow parameters according to the extracted vehicle information; according to the traffic flow parameters, constructing a traffic flow interruption probability model, and determining a final threshold value of emergency lane opening; based on the traffic flow parameters and the final threshold, training the multiple classification models; training a traffic jam early warning model according to the traffic flow parameters; a monitoring video of a target monitoring point in a target time period is obtained, video processing and data extraction are carried out, based on the calculated real-time traffic flow parameters, a traffic jam early warning model is used for processing, and traffic flow density and an early warning point are predicted; and according to the prediction result, using multiple classification models to decide whether to open the emergency lane, and fusing all decision results. According to the invention, the use of the emergency lane is optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent traffic optimization, and in particular relates to a dynamic management method and system for emergency lanes. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] There are many reasons for highway congestion. In addition to traffic accidents, the most typical reason is the bottleneck phenomenon on some sections of the road. The main reason is the gathering of vehicles, and the congestion is easy to spread. Some specific sections of highways are prone to forming congestion points, such as ramp entrances and exits or bridge entrances. Of course, some highways are affected by various factors, and the limited lanes and limited traffic capacity are also one of the main factors.

[0004] When constructing a highway, an emergency lane is often added on the right side to meet the needs of emergency vehicles such as engineering rescue, fire rescue, and medical rescue. Normally, the emergency lane is a life-saving channel and cannot be occupied at will. However, if it is used reasonably, for example, if it is found through monitoring of the upstream, midstream, and downstream traffic flow on a certain section that congestion is likely to occur, and there is no accident on the section, the emergency lane is allowed to be used, and the traffic density is reduced in time, it is likely to avoid a major congestion. The highway management department has been exploring how to use the emergency lane reasonably. In most cases, managers often use video surveillance of multiple sections and rely on experience to decide whether to allow temporary use of the emergency lane. On the one hand, this approach is prone to controversy due to the lack of theoretical basis, and on the other hand, it is difficult to evaluate the effect of alleviating congestion. Therefore, it is of great theoretical and practical significance to establish a mathematical model to explore the conditions for the impending congestion of a specific section of the highway and evaluate the role of the temporary use of the emergency lane in alleviating road congestion. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a dynamic management method and system for emergency lanes. The present invention establishes a traffic congestion warning model to warn of the congestion status of two adjacent monitoring video points in real time, provide decision support for traffic management departments, and provide decision makers with a theoretical basis for temporarily activating emergency lanes by integrating the decision results of multiple classification models, thereby optimizing the use of emergency lanes.

[0006] According to some embodiments, the present invention adopts the following technical solutions:

[0007] A dynamic management method for an emergency lane comprises the following steps:

[0008] Obtain historical surveillance videos of each monitoring point and perform video processing on the historical surveillance videos;

[0009] Extracting vehicle information from the processed video, determining the number and speed of each type of vehicle based on the extracted vehicle information, redefining and calculating traffic flow parameters, the traffic flow parameters including vehicle flow density, speed and traffic flow, and determining the changing trend of the traffic flow parameters over time;

[0010] According to the traffic flow parameters, the initial speed threshold for the opening of the emergency lane is analyzed, and the parameters of the congestion probability distribution are estimated to build a traffic flow interruption probability model, calculate the probability that the road capacity is less than the observed flow, and determine the final threshold for the opening of the emergency lane;

[0011] Based on traffic flow parameters and final thresholds, multiple classification models are trained;

[0012] According to the traffic flow parameters, the traffic congestion warning model is trained;

[0013] Obtain surveillance video of the target monitoring point within the target time period, perform video processing and data extraction, and use the trained traffic congestion warning model to process based on the calculated real-time traffic flow parameters to predict traffic density and warning points;

[0014] Based on the prediction results, a variety of pre-trained classification models are used to decide whether to open the emergency lane. The results of each decision are integrated to obtain the final judgment result on whether the emergency lane needs to be opened.

[0015] As an optional implementation, the process of video processing of historical surveillance videos includes: extracting image data using stratified sampling, selecting one frame as a representative every set time; merging data, calibrating timestamps, and using the 3σ principle and box plots to detect missing values ​​and outliers, and performing Lagrange interpolation on outliers and missing values.

[0016] As an optional implementation, the process of extracting vehicle information from the processed video includes: extracting vehicle information from the processed video using a target detection algorithm, and introducing a Kalman filter algorithm to improve recognition accuracy and target tracking accuracy.

[0017] As an optional implementation, the process of redefining and calculating traffic flow parameters includes: the traffic density is the total number of vehicles appearing in one frame;

[0018] The speed calculation process includes: calculating the speed of the same vehicle by tracking the position change in consecutive frames, that is, assigning a unique ID to each vehicle and recording its position in consecutive frames, and obtaining the speed of the vehicle by calculating the displacement of the vehicle between consecutive frames and dividing the time interval;

[0019] The traffic flow is the number of vehicles passing through the corresponding monitoring point per unit time.

[0020] As an optional implementation, the warning process of the traffic congestion warning model includes: evaluating whether the current traffic density is close to the critical density based on the traffic density data recorded by the relevant monitoring points, and issuing a traffic congestion alarm if the relevant monitoring points all reach the adjacent interval of the critical density and the density growth exceeds the set value;

[0021] By calculating the rate of change of the flow at relevant monitoring points, if the flow rate drops by more than a threshold and the density increases, the risk of congestion increases, and a traffic congestion alarm is issued;

[0022] Set the prediction time, and combine it with the current traffic density to determine whether congestion will occur within the set prediction time. If so, issue a traffic congestion alarm.

[0023] As an optional implementation method, the traffic congestion warning model is an LWR model. After the traffic congestion warning model is trained, cross-validation is performed using the TPI-LSTM method. The specific process includes: based on dynamic vehicle position information, the operating speeds of roads with different functional levels are obtained, and the weight of the road in the entire network is calculated according to the different road functions and traffic data. Finally, the index value is converted through people's perception and judgment of congestion. Based on the calculation method of the sliding window, the index value is divided to obtain training samples. Based on the training samples, the long short-term memory network neural network model is used in combination with historical data to predict whether a continuous warning state will occur, and multiple prediction results are obtained.

[0024] The prediction results of the LWR model can be used for short-term predictions, and the results of the LSTM model can be used for long-term predictions.

[0025] As an optional implementation, the process of determining the final threshold for opening the emergency lane includes: establishing a congestion index model based on traffic flow parameters, analyzing the initial speed threshold for opening and activating the emergency lane, estimating the parameters of the congestion probability distribution through the maximum likelihood estimation function, thereby constructing a traffic flow interruption probability model, calculating the probability that the road capacity is less than the observed flow, and visualizing the relationship between the congestion probability and the traffic flow, so as to determine the traffic flow for activating the emergency lane.

[0026] As a further step, the process of establishing a congestion index model includes: grading the traffic status and road operation status according to the traffic density, traffic flow and speed, determining the traffic congestion index, calculating the index indicator value, and drawing a visual line graph of its change over time.

[0027] As an optional implementation, a traffic flow interruption probability model is constructed to calculate the probability that the road capacity is less than the observed flow, and the process of visualizing the relationship between the congestion probability and the traffic flow includes: establishing a traffic flow interruption probability model:

[0028]

[0029] Where q is the number of vehicles passing through each lane within a set time, c is the capacity, i.e., the traffic capacity; P(c, q) is the probability that the capacity is less than the observed traffic; q i is the traffic flow observed at interval i, which represents the traffic flow before the speed decreases; ki refers to q>q i The number of time intervals; d i It means that the traffic flow reaches q i The number of time intervals when each congestion is considered separately, d i =1; B{is the set of congestion intervals B1, B2, ...};

[0030] The distribution parameters are estimated by applying the maximum likelihood function;

[0031] Based on historical data, the relationship between traffic flow parameters is determined and a scatter plot is drawn. Based on the relationship, the speed threshold is preliminarily estimated, and the traffic flow interruption probability model is solved to calculate the congestion probability. A scatter plot of the relationship between the congestion probability and the flow rate is drawn. Based on the relationship scatter plot, the threshold for activating the emergency lane is determined.

[0032] As an optional implementation, the process of training multiple classification models includes:

[0033] The final threshold determined by the traffic flow interruption probability model is used as the training set and as the input of the KNN model, SVM model and LWR traffic flow model for decision training.

[0034] As an optional implementation method, the process of fusing the decision results includes: using the idea of ​​random forest voting, fusing the decision results of the KNN model, the SVM model and the LWR traffic flow model, to complete the judgment of whether to open the emergency lane.

[0035] An emergency lane dynamic management system, comprising:

[0036] The video acquisition module is configured to acquire the historical surveillance videos of each surveillance point and perform video processing on the historical surveillance videos;

[0037] a traffic flow parameter calculation module, configured to extract vehicle information from the processed video, determine the number and speed of each type of vehicle based on the extracted vehicle information, redefine and calculate traffic flow parameters, the traffic flow parameters including vehicle density, speed and traffic flow, and determine the change trend of the traffic flow parameters over time;

[0038] an opening threshold calculation module, configured to analyze the initial speed threshold for the opening of the emergency lane according to the traffic flow parameters, estimate the parameters of the congestion probability distribution, thereby constructing a traffic flow interruption probability model, calculating the probability that the road capacity is less than the observed flow, and determining the final threshold for the opening of the emergency lane;

[0039] A decision training module, configured to train multiple classification models based on traffic flow parameters and final thresholds;

[0040] A congestion warning training module is configured to train a traffic congestion warning model according to traffic flow parameters;

[0041] The traffic congestion prediction module is configured to obtain surveillance videos of target surveillance points within a target time period, perform video processing and data extraction, and process based on the calculated real-time traffic flow parameters using the trained traffic congestion warning model to predict traffic density and warning points;

[0042] The decision fusion module is configured to decide whether to open the emergency lane based on the prediction results using a variety of pre-trained classification models, and to fuse the decision results to obtain a final judgment result on whether the emergency lane needs to be opened.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention realizes vehicle detection, tracking, counting and speed measurement by adopting YOLOv8 algorithm and Kalman filtering technology in combination with visual recognition technology, improves recognition accuracy and accurately extracts traffic flow parameters.

[0045] By establishing a traffic congestion model, the present invention can warn of the congestion status of two adjacent monitoring video points in real time, provide decision support for traffic management departments, construct a model for rationally activating emergency lanes on highways, provide decision makers with a theoretical basis for temporarily activating emergency lanes, and optimize the use of emergency lanes.

[0046] The present invention designs a KNN, SVM, and LWR model fusion decision algorithm based on random forest optimization, which improves the accuracy, recall rate, and F1 score of the decision. Through quantitative analysis, the average speed of the highway increased by 7%, the traffic density increased by 14%, the traffic flow increased by 21.4%, the congestion time decreased by 43.46 minutes, and the number of congestion events decreased by 24.14%.

[0047] The present invention uses visualization to display the changes of vehicle speed, total vehicle flow, vehicle flow density, traffic flow and traffic performance parameters (TPI) over time, making the change trend of traffic flow parameters more intuitive and easy to understand.

[0048] In summary, the present invention can effectively reduce congestion incidents and improve road utilization efficiency, especially during peak traffic hours.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0051] Figure 1 is a schematic diagram of an outlier processing process of an embodiment;

[0052] Figure 2 is a schematic diagram of a process flow of video processing and data extraction of an embodiment;

[0053] Figure 3 is a schematic diagram of an LSTM unit of an embodiment;

[0054] Figure 4 is a flow chart of traffic congestion prediction according to an embodiment;

[0055] Figure 5 is a flow chart of a highway emergency lane model in an embodiment;

[0056] Figure 6 1 is a schematic diagram of model decision making of an embodiment; (a) is the KNN model result of monitoring point 3; (b) is the SVM model result of monitoring point 3;

[0057] Figure 7 is a schematic diagram of a fusion decision process of an embodiment;

[0058] Figure 8 is a schematic diagram of a random forest combined with voting mechanism in one embodiment;

[0059] Fig. 9 is a schematic diagram of traffic density before and after the activation decision of an embodiment;

[0060] Fig.10 is a schematic diagram of average pixel speed before and after enabling decision of an embodiment;

[0061] Fig.11 is a schematic diagram of traffic volume before and after decision making in an embodiment;

[0062] Fig.12 is a schematic diagram of a decision model before optimization of an embodiment;

[0063] Fig.13The figure is a schematic diagram of an optimized decision model of an embodiment. DETAILED DESCRIPTION

[0064] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0065] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0066] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0067] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0068] Embodiment 1

[0069] A dynamic management method for an emergency lane comprises the following steps:

[0070] Obtain historical surveillance videos of each monitoring point and perform video processing on the historical surveillance videos;

[0071] Extracting vehicle information from the processed video, determining the number and speed of each type of vehicle based on the extracted vehicle information, redefining and calculating traffic flow parameters, the traffic flow parameters including vehicle flow density, speed and traffic flow, and determining the changing trend of the traffic flow parameters over time;

[0072] According to the traffic flow parameters, the initial speed threshold for the opening of the emergency lane is analyzed, and the parameters of the congestion probability distribution are estimated to build a traffic flow interruption probability model, calculate the probability that the road capacity is less than the observed flow, and determine the final threshold for the opening of the emergency lane;

[0073] Based on traffic flow parameters and final thresholds, multiple classification models are trained;

[0074] According to the traffic flow parameters, the traffic congestion warning model is trained;

[0075] Obtain surveillance video of the target monitoring point within the target time period, perform video processing and data extraction, and use the trained traffic congestion warning model to process based on the calculated real-time traffic flow parameters to predict traffic density and warning points;

[0076] Based on the prediction results, a variety of pre-trained classification models are used to decide whether to open the emergency lane. The results of each decision are integrated to obtain the final judgment result on whether the emergency lane needs to be opened.

[0077] The following is a detailed introduction:

[0078] In order to quickly and accurately extract video data such as vehicle type recognition, speed measurement, and vehicle counting, this embodiment selects the YOLOv8 algorithm for video processing and data extraction. The main advantages of YOLOv8 in vehicle counting, speed detection, and vehicle recognition include its real-time performance, high accuracy, and easy integration. YOLOv8 is designed for real-time applications and can quickly detect vehicles in video streams, which provides rapid response capabilities for traffic monitoring and management.

[0079] Since the span of highway video events is long and the video quality decreases after acceleration, the Kalman filter algorithm is introduced to improve the recognition accuracy and target tracking accuracy. During the data acquisition stage, the frame rate of the video and the time interval of each frame are accurately obtained, and the recognition area is clarified to avoid misidentifying the traffic flow in the opposite lane. In addition, the flow and density are redefined, and a duplicate elimination algorithm is added.

[0080] Due to factors such as video acceleration, near objects appear larger than far objects, etc., it is impossible to accurately measure speed through video. We redefined speed, and the measured speed has a coefficient relationship with the real speed. Therefore, we recorded the number and speed of each vehicle in detail, and exported it to an Excel file together with the time information.

[0081] a. Definition and calculation of traffic density

[0082] Redefine traffic density: the total number of vehicles appearing in one frame.

[0083]

[0084] Where ρ represents the traffic density, M represents the number of vehicles in this area, and length represents the length of a frame of road captured by the surveillance camera.

[0085] b. Definition and transformation of speed

[0086] The program that measures vehicle speed calculates its speed by tracking the position change of the same vehicle in consecutive frames. The program assigns a unique ID to each vehicle and records its position in consecutive frames. The speed of the vehicle can be obtained by calculating the position of the vehicle between consecutive frames and dividing it by the time interval. This method assumes that the movement of the vehicle between two frames is approximately uniform.

[0087]

[0088] When measuring speed, it involves video acceleration, near-major and far-major as well as image transformation. Therefore, the measured speed and the real speed need to be transformed through a transformation coefficient.

[0089] v z = k·v (3)

[0090] Where k is the transformation coefficient.

[0091] c. Traffic flow calculation

[0092] The number of vehicles passing a certain point per unit time at a certain moment is redefined and measured as the total number of vehicles within 5 seconds after this frame.

[0093] Q=ρ·v(4)

[0094] Among them, Q represents the traffic flow.

[0095] Since drawing images in frames appears to be very dense, a stratified sampling method is used to select one frame as a representative every second. After comparison, the trends before and after sampling are roughly the same.

[0096] Next, if Figure 2 As shown in the figure, the collected data was processed in detail, including data merging, time stamp calibration (one row of data is one second, and the data is re-assigned with a time stamp), and speed conversion calculation. Figure 1 As shown in the figure, the 3σ principle and box plot are used to detect missing values ​​and handle outliers in the data, and Lagrange interpolation is performed on outliers and missing values ​​to ensure the accuracy and completeness of the data.

[0097] A detailed visualization analysis of vehicle speed, traffic volume, traffic density and traffic flow is performed to intuitively show the changing trends of traffic flow parameters over time.

[0098] In order to establish a real-time warning model for traffic congestion, the LWR model and LSTM model were combined to establish a traffic congestion model. First, based on the LWR model and actual observations, the main influencing factors in the traffic flow model were analyzed, and the data observed at four points were considered in the traffic flow model from high to low importance. The LWR calculation method and LSTM machine learning methods were used respectively, combined with traffic density and flow data, to establish a continuous congestion state warning model, predict the possible continuous congestion state 10 minutes in advance, and compare the accuracy of the two methods.

[0099] In specific predictions, the LWR model makes predictions based on the congestion index fluctuations in a shorter time series before the prediction time. It is sensitive to short-term fluctuations. When the time series is shorter, the LWR model has higher judgment accuracy, faster speed, and is more economical. The LSTM model is trained by inputting data samples in a longer time series. When the time series is longer, the LSTM long time series prediction ability is better reflected, and the LSTM model is used at this time.

[0100] Based on the obtained video data, the video data is processed to obtain the actual number of congestion events, and the time point when the real-time warning is issued is displayed by means of visualization. In addition, the accuracy of the model is calculated by comparing it with the congestion events warned, and the effectiveness of the congestion warning model is verified.

[0101] The LWR model assumes that the flow of vehicles is similar to that of a fluid. The equations of the model are as follows:

[0102]

[0103] The meanings of the physical quantities in the formula are as follows:

[0104] ρ: vehicle flow density at the observation point, (n is the number of vehicles, l is the road length), unit: vehicle / km.

[0105] q: traffic flow at the observation point, (t is time), unit: vehicle / hour

[0106] t: the time when the observation point observed the data

[0107] x: The location of the observation point

[0108] In order to better fit the actual situation, the vehicle speed v is introduced. It is generally believed that the vehicle speed v is a function of the traffic density, and the vehicle speed and traffic density satisfy the following relationship:

[0109]

[0110] where v max is the maximum vehicle speed, ρ max is the maximum traffic density.

[0111] Combining the above two equations, the flow rate can be expressed as a quadratic function of density:

[0112]

[0113] It can be seen from the formula that this is a parabola opening downwards, with low flow rates at low and high densities and maximum flow rates at medium density. That is:

[0114]

[0115] The density ρ crit This is the critical density of traffic.

[0116] An important feature of the LWR model is its ability to predict the formation and dissipation of traffic jams. In a traffic flow, an increase in vehicle density can cause a decrease in speed, which can lead to the formation of jams. The LWR model helps understand and predict changes in traffic flow by simulating these phenomena.

[0117] Congestion warning model establishment

[0118] According to the theory of traffic flow stability, when the traffic density is close to the critical density, the traffic flow is prone to change from a stable state to an unstable state, causing congestion. Therefore, the following method is adopted to establish a congestion warning model, taking monitoring points 3 and 4 as examples for explanation:

[0119] (1) Based on the traffic density data ρ3(t) and ρ4(t) recorded at monitoring points 3 and 4, evaluate whether the current traffic density is close to the critical density ρ crit If both ρ3(t) and ρ4(t) reach the critical density interval and the density increases rapidly, a traffic congestion alarm is issued. The threshold of the critical density interval is usually set to ρ crit A percentage of, for example, 90%.

[0120] (2) By calculating the change rates of flow Δq3(t) and Δq4(t), if the flow rate decreases significantly and the density increases, the congestion risk increases. The calculation formula of Δq3(t) is as follows:

[0121] Δq3(t)=q3(t)-q3(t-Δt)(9)

[0122] If Δq3(t)<0 and the density increases, an early warning is issued.

[0123] (3) Set the prediction time T pre (e.g. 10 minutes), combined with the current density T pre Will there be congestion during the time period?

[0124]

[0125] In addition to the LWR model, a TPI-LSTM method can also be considered for cross-validation.

[0126] TPI is an indicator that reflects the degree of traffic congestion based on in-depth processing of dynamic vehicle location information (floating vehicle data). By processing vehicle location data, the operating speed of roads with different functional levels can be obtained, and then the weight of the road in the entire network is calculated based on the different road functions and traffic data. Finally, based on people's perception of congestion, an index value converted to 0-10 is given. The calculation formula is as follows:

[0127]

[0128] Where γ p (t) is the traffic congestion index of the corresponding time period at time t at location p; occ p (t) is the road time occupancy rate at time t at location p; Q p (t) is the traffic flow at time t at location p; v p (t) is the location velocity at time t at location p.

[0129] Road time occupancy refers to the percentage of the cumulative time that vehicles pass through a certain section of the road in a certain period of time. The calculation formula for road time occupancy is as follows:

[0130]

[0131] Where R t represents the lane time occupancy rate; t i represents the occupancy time of the i-th vehicle; T represents the total observation time; n is the number of vehicles on the road section.

[0132] In the process of solving this problem, we first calculate the lane time occupancy rate based on the number of vehicles and passing time obtained through image processing in the first question, and then calculate the traffic congestion index at each sampling time point according to formula (11) for subsequent prediction and verification.

[0133] At the same time, this embodiment also introduces a long short-term memory network (LSTM), which enhances the long-term dependency processing capability through a special memory unit, the LSTM unit. This design enables the network to maintain information over a long time span, thereby effectively retaining and utilizing long-term historical information.

[0134] like Figure 3 As shown in Figure 1, the Long Short-Term Memory Network (LSTM) is a derivative type of Recurrent Neural Network (RNN) that uses built-in memory units to capture long-term and short-term patterns in sequence data. Figure 3The core of the LSTM unit shown includes a memory cell ct for storing information. There are three key control mechanisms inside the LSTM: input gate it, forget gate ft, and output gate ot. The regulatory effects of these gates are represented by the blue circles in the figure. This ingenious design enables LSTM to effectively decide which information should be retained, forgotten, or output when processing sequence information, thereby optimizing its performance in sequence tasks.

[0135] i t =σ(W hi h t-1 +W xi X t +b i )(13)

[0136] f t =σ(W hf h t-1 +W xf X t +b f )(14)

[0137] o t =σ(W ho h t-1 +W xo X t +b o )(15)

[0138]

[0139] Where X t is the input vector at the tth time step, c t candidate memory state, σ is the sigmoid activation function, tanh represents the hyperbolic tangent activation function, W hi , W xi , W ho , W xo , W hc , W xc is the weight matrix, b i , b o , b c is the deviation term, represents the point-wise multiplication of two vectors. Specifically, the input vector is processed by the input gate and the forget gate to determine how to update the old memory state c t-1 .

[0140] Then, the normalized c is filtered by using the output gate t , emits the output state (also called the “hidden state”) h tThe LSTM network is a recurrent neural network composed of LSTM cells that can learn fluctuation patterns in long-term time series and then make predictions.

[0141] Prediction based on LSTM time series:

[0142] 1) Training

[0143] Based on the calculation method of sliding window, we analyze TPI. We select a time length of n as a sliding window, select a step size of 1, and analyze the TPI within the selected n time lengths as a training sample. If the sum of twice the TPI in the first half of this period is less than the sum of TPI in the second half of the period, then output "warning state", otherwise output "non-warning state". We can define "warning state" as "1" and "non-warning state" as 0. In this way, every time the time window slides forward, we get a new training sample. Finally, for the total number of l TPI changes over time, we can get nl training samples and a new set of data sets. We can use this set of data sets to train the LSTM network.

[0144] 2) Prediction

[0145] Using the LSTM neural network model obtained in the previous section, we can combine historical data to predict whether a continuous warning state will occur, and we can get nl prediction results.

[0146] Constructing a highway emergency lane model requires finding a reasonable threshold for enabling the emergency lane, which is closely related to traffic flow parameters (including average vehicle speed, traffic flow, and vehicle density). In addition, the activation of the emergency lane must not affect the passage of emergency vehicles. When there are emergency vehicles on the lane, the emergency lane must be closed in a timely manner. In severe weather conditions, the emergency lane can also be enabled to ensure road safety. In peak traffic hours such as holidays, the emergency lane can also be enabled to relieve traffic pressure. Since emergency vehicles, severe weather, and peak traffic hours are special cases, we do not consider them.

[0147] like Figure 4 As shown in the figure, the basic traffic flow parameters (including traffic density, traffic flow, and vehicle speed) obtained by extracting and processing the video data are further processed. First, a congestion index model is established, and then the initial speed threshold for opening and activating the emergency lane is analyzed and the parameters of the congestion probability distribution are estimated through the maximum likelihood estimation function, thereby constructing a traffic flow interruption probability model, calculating the probability that the road capacity is less than the observed flow, and then visualizing the relationship between the congestion probability and the traffic flow, so as to decide at what traffic flow the emergency lane should be activated, providing a theoretical basis for the decision to temporarily activate the emergency lane.

[0148] Traffic congestion index model establishment

[0149] The basic traffic flow parameters (including traffic density, traffic flow, and vehicle speed) obtained through video data extraction and processing are further processed to establish a congestion index model to preliminarily judge the road congestion situation.

[0150] The traffic congestion index ranges from 1 to 10 and is divided into five levels (i.e., "smooth traffic", "basically smooth traffic", "mild congestion", "moderate congestion", and "severe congestion"). The higher the value, the more serious the traffic congestion. As shown in the following table:

[0151] Table 1 Traffic congestion index classification

[0152]

[0153] In the calculation we assume that C i is the number of times the i-th vehicle appears within 5 seconds, then the road time occ at time t at location p p for;

[0154]

[0155] Assuming that each car is about 4 meters long, the length of the road that can be identified by the video is estimated to be 20 meters. So when calculating γ p When (t) is TPI, the formula is rewritten as:

[0156]

[0157] Where γ p (t) is the traffic congestion index of the corresponding time period at time t at location p; Q p (t) is the traffic flow at time t at location p; v p (t) is the location velocity at time t at location p.

[0158] Calculate γ p (t) and draw a visual line graph showing the change of TPI over time.

[0159] The calculation results of the traffic congestion index help us understand the congestion situation of roads. However, although the index calculated initially can provide a rough assessment of traffic conditions, it may be inaccurate. In order to improve the credibility and accuracy of the analysis, we need to continuously monitor traffic conditions and use calculation formulas to obtain more accurate assessment results.

[0160]

[0161] Where: I p(t) is the traffic congestion level evaluation value at time t at location p, where 1 represents a smooth state and 2 represents a congested state; I p (t) is the single traffic status evaluation value at time t at location p, and g is the set number of tests. That is, the reliability of the results is ensured by multiple tests.

[0162] Establish a traffic flow interruption probability model:

[0163]

[0164] Where: q is the number of vehicles passing through each lane per hour, veh / h;

[0165] c is capacity, veh / h, i.e., traffic capacity;

[0166] P(c,,q) is the probability that the capacity is less than the observed flow;

[0167] q i is the traffic flow observed at interval i, which represents the traffic flow before the speed decreases;

[0168] ki means q>q i The number of time intervals;

[0169] d i It means that the traffic flow reaches q i The number of time intervals when each congestion is considered separately, d i =1;

[0170] B{is a set of congestion intervals B1, B2, ...}.

[0171] In order to estimate the parameters of the congestion probability distribution, the function type of the distribution must be determined in advance. The distribution parameters can be estimated by applying the maximum likelihood technique. For capacity analysis, the likelihood function is:

[0172]

[0173] Where: f c (q i ) is the statistical density function of capacity c;

[0174] F c (q i ) is the cumulative distribution function of capacity c;

[0175] n represents the number of time intervals;

[0176] δ i = 1, the time interval i includes unreviewed objects;

[0177] δ i=0, the time interval i does not include unexamined objects.

[0178] Solution to the traffic flow interruption probability model

[0179] In order to solve the congestion probability of the traffic flow interruption model, the relationship between the three major traffic flow parameters is obtained by analyzing and processing historical data, including the relationship between average speed and density, the relationship between flow and density, and the relationship between flow and average speed, and a scatter plot is drawn.

[0180] From the relationship diagram of the above three parameters, it can be concluded that when the average speed is higher than a certain value V m When the traffic density is small, when the average speed is lower than V m When the average speed is higher than V m When the flow rate is small, when the average speed is lower than V m When the flow rate is large, the speed threshold V m .

[0181] First, we need to solve k i and d i . k i means q≥q i The number of time intervals; d i It means that the traffic flow reaches q i The number of time intervals when each congestion is considered separately, d i =1; According to the definition, we can get k i The formula is:

[0182] K=∑C i (twenty four)

[0183] Assume V ij is the speed of the i-th vehicle at the j-th second. The speed threshold V can be preliminarily estimated from the relationship between the above three parameters. m , so we can solve d i :

[0184] d=∑C i (Vij>Vmkm / h) (25)

[0185] Substituting the above calculation formulas of K and d into the interruption probability model (2) yields the solution formula for the probability P:

[0186]

[0187] The above formula can be used to calculate the congestion probability, and a scatter plot of the relationship between the congestion probability and the flow rate can be drawn. It can be seen intuitively from the scatter plot of the relationship between the congestion probability and the flow rate that when the flow rate is a certain threshold value Q m The emergency lane should be activated.

[0188] like Figure 5 As shown, real-time decision is made whether to enable the emergency lane:

[0189] In order to ensure the differences between models and improve the generalization ability of model decision results, we use different congestion judgment criteria when training different classification models, on the one hand to improve the reliability of model prediction, and on the other hand to ensure accuracy.

[0190] like Figure 6 As shown in (a), the KNN model trains the congestion state judgment criteria:

[0191] Congestion judgment criteria: Before training the KNN model, we need to introduce a threshold-based congestion probability assessment criterion.

[0192] According to the time-speed image, based on the traffic flow speed and traffic volume in the current time interval, the probability of traffic flow interruption (traffic congestion) in the next time interval is determined. The traffic flow interruption probability model is as follows:

[0193]

[0194] Where q is the number of vehicles passing through each lane per hour, veh / (h·ln); c is the capacity, veh / (h·ln), i.e., the traffic capacity; P(c≤q) is the probability that the capacity is less than the observed flow; qi is the traffic flow observed at interval i, which represents the traffic flow before the speed drops; ki refers to the number of time intervals when q≥qi; di refers to the number of time intervals when the traffic flow reaches qi, and when each congestion is considered separately, di=1; B is the set of congestion intervals {B1, B2, …}.

[0195] Based on this model, we can determine whether there is congestion at a certain moment. We use this model as a criterion to improve the training set and use it as the input of the KNN model.

[0196] SVM model training congestion status criteria

[0197] Congestion status criterion judgment: In SVM, the criterion based on traffic congestion index is adopted. We can measure the degree of traffic congestion through the traffic congestion index, which is roughly divided into smooth, basically smooth, light congestion, moderate congestion, and severe congestion. The established decision model is as follows:

[0198]

[0199] Where: γp(t) is the traffic congestion index of the corresponding period at time t at location p; occp(t) is the road time occupancy rate at time t at location p; Qp(t) is the traffic flow at time t at location p; vp(t) is the location speed at time t at location p.

[0200] By setting a specific threshold, we can use the traffic congestion index to judge the congestion of the road, that is, by comparing the actual index with this threshold, we can determine whether the road is unobstructed or congested. However, this traffic flow judgment based on a single time point may not be completely accurate and there is a risk of misjudgment. In order to improve the reliability and accuracy of the analysis results, it is recommended to continuously monitor the traffic conditions. By applying formula (27), we can more accurately evaluate the traffic conditions of the road and ensure the stability of the analysis conclusions.

[0201]

[0202] Where: Ip(t) is the traffic congestion level evaluation value at time t at location p, 1 represents a smooth state, and 2 represents a congested state; Ip(t) is the single traffic state evaluation value at time t at location p, 1 represents a smooth state, and 2 represents a congested state; g is the set number of detections.

[0203] The training set can be processed according to the congestion probability assessment model and then used as the input of SVM to complete the model training and achieve the purpose of classification decision, such as Figure 6 as shown in (b).

[0204] LWR traffic flow model: The LWR traffic flow model is a macroscopic traffic flow model that regards traffic flow as a continuous medium and describes the evolution of traffic flow through the relationship between flow, density and speed. The model establishes nonlinear scalar conservation laws for traffic flow and can describe nonlinear hyperbolic wave phenomena such as congestion waves and sparse waves in traffic flow. The LWR model has important theoretical and practical significance in traffic management and signal light control. It is easy to implement calculations and has a good effect on simulating the dynamic characteristics of traffic flow. In addition, the LWR model can also be combined with deep learning to solve challenges in traffic state estimation, such as data scarcity and sensor noise, through physical information deep learning methods.

[0205] In order to correctly establish the model and achieve the purpose of real-time decision-making on whether to enable the emergency lane. A KNN, SVM, and LWR fusion model decision algorithm based on random forest optimization is proposed. First, the congestion status is obtained according to the real-time warning model, and whether the emergency lane needs to be opened is judged according to the congestion status. Secondly, in order to optimize the decision, the KNN and SVM classification models are introduced, and the idea of ​​random forest voting is adopted to complete the judgment of whether to open the emergency lane. The third point monitoring data is used as the training set data and input into the KNN and SVM algorithms for model training and verification of the model accuracy. Then the model is migrated to predict the data of other detection points to determine whether the emergency lane needs to be opened. The three models of KNN, SVM, and LWR will obtain three results. Cross-validation is performed on the three algorithms to determine the accuracy of the model algorithms, such as Figure 7 shown.

[0206] In order to verify the accuracy of the model, we introduce accuracy, recall, and F1 score as evaluation indicators for classification. Through the three-dimensional scatter plot, we can intuitively observe the distribution of data points of different categories (congested and non-congested).

[0207] like Figure 8 As shown in the figure, the random forest voting mechanism is finally used to vote and summarize the results obtained by the three algorithms, and finally determine whether to open the emergency lane. This method effectively reduces the risk of overfitting, because the prediction results of different trees may offset each other, thereby improving the generalization ability of the model. At the same time, compared with the prediction of a single model, it improves the robustness and accuracy of the model.

[0208] In addition, according to the changing trend of traffic flow, speed and vehicle density on the highway after the emergency lane is opened, combined with the decision results of the decision model, the traffic flow, speed and vehicle density of the time period judged as "needing to open the emergency lane" in the decision model are weighted according to the changing trend, and the weighting of traffic flow, speed and vehicle density is cancelled after the variable is reduced to the emergency lane opening threshold. Finally, the weighted traffic volume data is calculated through the LWR model in the first question to calculate the number of congestion after the weighting of the decision mechanism, and compared with the number of congestion of the original data. According to formula (28), the reduction rate of the decision model for congestion is calculated, κ is the reduction rate, Nbf is the original number of congestion, and Naf is the number of congestion after the decision of "whether the emergency lane should be opened".

[0209]

[0210] Figure 9-13 To compare the images of parameters such as average vehicle speed, traffic density, and traffic volume before and after the decision.

[0211] like Figure 9-13 It can be seen that after the decision to open the emergency lane was made, the traffic density between monitoring points 3 and 4 on the highway decreased significantly. After the decision was made, the average speed of vehicles increased in all time periods. After the decision was made, the traffic volume on the highway increased to a certain extent.

[0212] Traffic congestion event images before and after the decision is made, where the blue line represents the non-congested period and the red line represents the congested period.

[0213] According to data calculations, among the total 2,187 event points selected, 814 points had traffic jams, which was 528 fewer event points than the 1,342 event points where emergency decisions were not made, equivalent to a 24.142% reduction in congestion time.

[0214] Table 2 Decision effect quantification table

[0215]

[0216] In summary, it is proved that the proposed decision-making model is effective.

[0217] Embodiment 2

[0218] An emergency lane dynamic management system, characterized in that it includes:

[0219] The video acquisition module is configured to acquire the historical surveillance videos of each surveillance point and perform video processing on the historical surveillance videos;

[0220] a traffic flow parameter calculation module, configured to extract vehicle information from the processed video, determine the number and speed of each type of vehicle based on the extracted vehicle information, redefine and calculate traffic flow parameters, the traffic flow parameters including vehicle density, speed and traffic flow, and determine the change trend of the traffic flow parameters over time;

[0221] an opening threshold calculation module, configured to analyze the initial speed threshold for the opening of the emergency lane according to the traffic flow parameters, estimate the parameters of the congestion probability distribution, thereby constructing a traffic flow interruption probability model, calculating the probability that the road capacity is less than the observed flow, and determining the final threshold for the opening of the emergency lane;

[0222] A decision training module, configured to train multiple classification models based on traffic flow parameters and final thresholds;

[0223] A congestion warning training module is configured to train a traffic congestion warning model according to traffic flow parameters;

[0224] The traffic congestion prediction module is configured to obtain surveillance videos of target surveillance points within a target time period, perform video processing and data extraction, and process based on the calculated real-time traffic flow parameters using the trained traffic congestion warning model to predict traffic density and warning points;

[0225] The decision fusion module is configured to decide whether to open the emergency lane based on the prediction results using a variety of pre-trained classification models, and to fuse the decision results to obtain a final judgment result on whether the emergency lane needs to be opened.

[0226] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention without creative labor shall be included in the protection scope of the present invention.

Claims

1. A method for dynamic management of emergency lanes, characterized in that: The following steps are involved: Obtain historical surveillance videos of each monitoring point and perform video processing on the historical surveillance videos; Extracting vehicle information from the processed video, determining the number and speed of each type of vehicle based on the extracted vehicle information, redefining and calculating traffic flow parameters, the traffic flow parameters including vehicle flow density, speed and traffic flow, and determining the changing trend of the traffic flow parameters over time; According to the traffic flow parameters, the initial speed threshold for the opening of the emergency lane is analyzed, and the parameters of the congestion probability distribution are estimated to build a traffic flow interruption probability model, calculate the probability that the road capacity is less than the observed flow, and determine the final threshold for the opening of the emergency lane; Based on traffic flow parameters and final thresholds, multiple classification models are trained; According to the traffic flow parameters, the traffic congestion warning model is trained; Obtain surveillance video of the target monitoring point within the target time period, perform video processing and data extraction, and use the trained traffic congestion warning model to process based on the calculated real-time traffic flow parameters to predict traffic density and warning points; Based on the prediction results, a variety of pre-trained classification models are used to decide whether to open the emergency lane. The results of each decision are integrated to obtain the final judgment result on whether the emergency lane needs to be opened.

2. The method for dynamic management of emergency lanes according to claim 1, characterized in that: The process of video processing for historical surveillance videos includes: extracting image data using stratified sampling, selecting one frame as a representative at set intervals; merging data, calibrating timestamps, and using the 3σ principle and box plots to detect missing values ​​and outliers, and performing Lagrange interpolation on outliers and missing values.

3. The method for dynamic management of emergency lanes as claimed in claim 1, characterized in that: The process of extracting vehicle information from the processed video includes: extracting vehicle information from the processed video using a target detection algorithm, and introducing a Kalman filter algorithm to improve recognition accuracy and target tracking accuracy.

4. The method for dynamic management of emergency lanes according to claim 1, characterized in that: The process of redefining and calculating traffic flow parameters includes: the traffic flow density is the total number of vehicles appearing in one frame; The speed calculation process includes: calculating the speed of the same vehicle by tracking the position change in consecutive frames, that is, assigning a unique ID to each vehicle and recording its position in consecutive frames, and obtaining the speed of the vehicle by calculating the displacement of the vehicle between consecutive frames and dividing the time interval; The traffic flow is the number of vehicles passing through the corresponding monitoring point per unit time.

5. The method for dynamic management of emergency lanes as claimed in claim 1, characterized in that: The warning process of the traffic congestion warning model includes: evaluating whether the current traffic density is close to the critical density based on the traffic density data recorded by the relevant monitoring points, and issuing a traffic congestion alarm if the relevant monitoring points all reach the critical density proximity interval and the density growth exceeds the set value; By calculating the rate of change of the flow at relevant monitoring points, if the flow rate drops by more than a threshold and the density increases, the risk of congestion increases, and a traffic congestion alarm is issued; Set the prediction time, and combine it with the current traffic density to determine whether congestion will occur within the set prediction time. If so, issue a traffic congestion alarm.

6. The method for dynamic management of emergency lanes as claimed in claim 1, characterized in that: The traffic congestion warning model is an LWR model. After the traffic congestion warning model is trained, it is cross-validated using the TPI-LSTM method. The specific process includes: based on dynamic vehicle position information, the operating speed of roads with different functional levels is obtained, and the weight of the road in the entire network is calculated according to the different road functions and traffic data. Finally, through people's perception and judgment of congestion, the index indicator value is converted, and based on the calculation method of the sliding window, the index indicator value is divided to obtain training samples. Based on the training samples, the long short-term memory network neural network model is used in combination with historical data to predict whether a continuous warning state will occur, and multiple prediction results are obtained.

7. The method for dynamic management of emergency lanes as claimed in claim 1, characterized in that: The process of determining the final threshold for opening the emergency lane includes: establishing a congestion index model based on traffic flow parameters, analyzing the initial speed threshold for opening and activating the emergency lane, estimating the parameters of the congestion probability distribution through the maximum likelihood estimation function, thereby constructing a traffic flow interruption probability model, calculating the probability that the road capacity is less than the observed flow, and visualizing the relationship between the congestion probability and the traffic flow, so as to determine the traffic flow for activating the emergency lane; The process of establishing a congestion index model includes: grading the traffic status and road operation status according to traffic density, traffic flow and speed, determining the traffic congestion index, calculating the index indicator value, and drawing a visual line graph of its changes over time.

8. The method for dynamic management of emergency lanes as claimed in claim 1, characterized in that: The process of constructing a traffic flow interruption probability model, calculating the probability that the road capacity is less than the observed flow, and visualizing the relationship between congestion probability and traffic flow includes: Establishing a traffic flow interruption probability model: Where q is the number of vehicles passing through each lane within a set time, c is the capacity, i.e., the traffic capacity; P(c, q) is the probability that the capacity is less than the observed traffic; q i is the traffic flow observed at interval i, which represents the traffic flow before the speed decreases; ki refers to q>q i The number of time intervals; d i It means that the traffic flow reaches q i The number of time intervals when each congestion is considered separately, d i =1; B{is the set of congestion intervals B1, B2, ...}; The distribution parameters are estimated by applying the maximum likelihood function; Based on historical data, the relationship between traffic flow parameters is determined and a scatter plot is drawn. Based on the relationship, the speed threshold is preliminarily estimated, and the traffic flow interruption probability model is solved to calculate the congestion probability. A scatter plot of the relationship between the congestion probability and the flow rate is drawn. Based on the relationship scatter plot, the threshold for activating the emergency lane is determined.

9. The method for dynamic management of emergency lanes as claimed in claim 1, characterized in that: The process of training multiple classification models includes: The final threshold determined by the traffic flow interruption probability model is used as the training set and as the input of the KNN model, SVM model and LWR traffic flow model for decision training; The process of fusing the decision results includes: using the idea of ​​random forest voting, fusing the decision results of the KNN model, SVM model and LWR traffic flow model, and completing the judgment of whether to open the emergency lane.

10. An emergency lane dynamic management system, characterized in that: include: The video acquisition module is configured to acquire the historical surveillance videos of each surveillance point and perform video processing on the historical surveillance videos; a traffic flow parameter calculation module, configured to extract vehicle information from the processed video, determine the number and speed of each type of vehicle based on the extracted vehicle information, redefine and calculate traffic flow parameters, the traffic flow parameters including vehicle density, speed and traffic flow, and determine the change trend of the traffic flow parameters over time; an opening threshold calculation module, configured to analyze the initial speed threshold for the opening of the emergency lane according to the traffic flow parameters, estimate the parameters of the congestion probability distribution, thereby constructing a traffic flow interruption probability model, calculating the probability that the road capacity is less than the observed flow, and determining the final threshold for the opening of the emergency lane; A decision training module, configured to train multiple classification models based on traffic flow parameters and final thresholds; A congestion warning training module is configured to train a traffic congestion warning model according to traffic flow parameters; The traffic congestion prediction module is configured to obtain surveillance videos of target surveillance points within a target time period, perform video processing and data extraction, and process based on the calculated real-time traffic flow parameters using the trained traffic congestion warning model to predict traffic density and warning points; The decision fusion module is configured to decide whether to open the emergency lane based on the prediction results using a variety of pre-trained classification models, and to fuse the decision results to obtain a final judgment result on whether the emergency lane needs to be opened.

Images