Expressway emergency lane starting method and device based on video data

Abstract

The invention relates to the technical field of traffic resource allocation, and provides an expressway emergency lane starting method and device based on video data. A traffic flow congestion prediction model is constructed, so that the congestion condition is predicted, and early warning of congestion is realized; on the basis of the output prediction result, an emergency lane starting decision model is constructed; and taking a decision output by the emergency lane starting decision model as a label, constructing a data set in combination with the output of the traffic flow congestion prediction model, and further constructing the emergency lane starting decision model of the interpretable decision tree, thereby controlling the starting of the emergency lane by using the constructed emergency lane starting decision model of the interpretable decision tree. The emergency lane starting decision-making model capable of explaining the decision-making tree can directly display key factors influencing emergency lane starting decision-making, has high interpretability, and provides a clear and visual decision-making path.

Description

Method and device for activating emergency lanes on highways based on video data Technical Field

[0001] This application relates to the field of traffic resource allocation technology, and in particular to a method and device for activating emergency lanes on highways based on video data. Background Technology

[0002] During highway construction, emergency lanes are often added on the right side to accommodate emergency vehicles such as fire trucks, medical personnel, and engineering rescue vehicles. As an important management tool for adjusting traffic flow on highways, the proper use of emergency lanes can effectively alleviate highway congestion and improve highway safety to a certain extent.

[0003] However, there are still many challenges in the use of emergency lanes. How to effectively identify and predict highway traffic congestion, scientifically evaluate the role of temporary use of emergency lanes in alleviating road congestion, and formulate scientific emergency lane usage standards to improve highway operating efficiency have become urgent problems to be solved in the field of transportation research. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and device for activating emergency lanes on highways based on video data to address the aforementioned technical problems.

[0005] A method for activating emergency lanes on highways based on video data includes the following steps: collecting highway video data from preset monitoring points and extracting state and speed information for different lanes; calculating traffic flow parameter data based on the state and speed information of different lanes, and using the traffic flow parameter data as input to construct a traffic flow congestion prediction model, outputting predicted future traffic flow parameter data, traffic congestion index, and travel time, and issuing a real-time warning before persistent congestion occurs; the traffic flow parameter data includes traffic volume, road occupancy, and average vehicle speed; calculating congestion severity indicators using the predicted traffic congestion index and travel time, and constructing an emergency lane activation decision model with the objective of minimizing congestion queue length by combining the highway accident rate, congestion duration, and congestion queue length calculated from the predicted traffic flow parameter data; using the decisions output by the emergency lane activation decision model as labels, and constructing a dataset by combining the output of the traffic flow congestion prediction model; using the dataset as input to construct an interpretable decision tree-based emergency lane activation decision model, and activating emergency lanes at corresponding road segment locations and times according to the output decisions of the interpretable decision tree-based emergency lane activation decision model.

[0006] In one embodiment, highway video data is collected from preset monitoring points, and state and speed information of different lanes is extracted. This includes: collecting highway video data from preset monitoring points, detecting vehicles in the video data using a deep learning-based target detection algorithm, tracking vehicles in the video data using the DeepSORT algorithm, and extracting state and speed information of different lanes. The state information includes the number of vehicles in the lane, road occupancy, and traffic flow. The speed information includes the average speed of the lane and the average speed of the vehicle.

[0007] In one embodiment, highway video data from preset monitoring points is collected, and status and speed information of different lanes are extracted. The method also includes: preprocessing the extracted status and speed information of different lanes, wherein the preprocessing includes filtering, clearing, repairing and standardizing; and generating a heat map based on the preprocessed status and speed information of different lanes for visualization.

[0008] In one embodiment, traffic flow parameter data is calculated based on the state and speed information of different lanes. Using this traffic flow parameter data as input, a traffic congestion prediction model is constructed, outputting predicted future traffic flow parameter data, a traffic congestion index, and travel time. A real-time warning is issued before persistent congestion occurs. This includes: calculating traffic flow parameter data based on the state and speed information of different lanes and performing data signal filtering; dividing the filtered traffic flow parameter data into a training set and a test set; constructing a traffic congestion prediction model using the training set; using the test set as model input for prediction; and optimizing model performance by adjusting parameters; outputting predicted future traffic flow parameter data, a traffic congestion index, and travel time using the trained traffic congestion prediction model, and issuing a real-time warning before persistent congestion occurs.

[0009] In one embodiment, the emergency lane activation decision model is: ; ; ;in, For decision variables; This refers to the average length of the vehicle. Traffic flow downstream of the congestion point; For the traffic capacity of all expressways; To ensure the traffic capacity of the emergency lane; Highway accident rate; The threshold for highway accident rate; As an indicator of congestion level; Duration of congestion; The threshold for congestion duration; This refers to the length of the congestion queue. This is the threshold for congestion queue length; For vehicle speed; This is the maximum speed limit on highways.

[0010] In one embodiment, using the dataset as input, an emergency lane activation decision model with an interpretable decision tree is constructed, including: inputting the dataset, which includes a training set D and an attribute set A; generating an empty node as the root of the current subtree, denoted as Node; traversing the samples in the training set, and if the samples in the training set belong to the same category C, then marking Node as a category C node to complete the category branch construction; calculating the attribute information entropy Ent for each attribute in the attribute set, and using the attribute with the smallest Ent as the feature to divide the attribute. The training set is filtered to obtain a subset. ; Determine subset Is it an empty set? If the set is empty, then mark the branch node as a leaf node and label it with the class that has the most samples in D; if If it is not an empty set, then remove it from attribute set A. The updated attribute set is obtained. Call the recursive function The returned subtree is used as the child node of the current branch; the decision tree with Node as the root node is output, and an interpretable decision tree is constructed for the emergency lane activation decision model.

[0011] In one embodiment, using the dataset as input, constructing an interpretable decision tree for an emergency lane activation decision model further includes: weighting the samples in the dataset. ; ; ;in, The sample weights for activating emergency lane G; The number of vehicles passing through the cross section within the observation period; The number of vehicles passing through section G of the emergency lane during the observation period; The sample weights for G where the emergency lane was not activated; This refers to the number of vehicles passing through sections other than emergency lane G during the observation period.

[0012] In one embodiment, the likelihood of each node in the decision tree is calculated according to the following formula: ;in, Let n be the likelihood of node n; Let n be the number of vehicles that passed through the emergency lane G section within the observation time corresponding to node n. Let n be the number of vehicles that passed through sections other than emergency lane G during the observation period corresponding to node n.

[0013] In one embodiment, the method further includes: constructing a multi-objective optimization model for monitoring layout with the objectives of minimizing the setup cost of monitoring equipment and maximizing effective coverage, and generating an initial set of monitoring equipment locations; the multi-objective optimization model for monitoring layout is as follows: ; ; ; ;in, Set costs for individual monitoring devices; For control variables; Number of monitoring devices; The number of locations to be covered; The number of nodes effectively covered; denoted as the total number of nodes to be covered; ECR as the effective coverage rate; S as the set of candidate locations for monitoring devices; T as the set of locations of monitoring points to be covered; r as the radius threshold of the effective coverage area; using the initial set of monitoring device locations as the initial population, the tDEA-CPBI algorithm is used for optimization, and the monitoring device locations are output.

[0014] A highway emergency lane activation device based on video data includes: a data sampling module for collecting highway video data from preset monitoring points and extracting state and speed information for different lanes; a congestion prediction module for calculating traffic flow parameter data based on the state and speed information of different lanes, constructing a traffic flow congestion prediction model using the traffic flow parameter data as input, outputting predicted future traffic flow parameter data, a traffic congestion index, and travel time, and issuing a real-time warning before persistent congestion occurs; the traffic flow parameter data includes traffic volume, road occupancy, and average vehicle speed; and a decision model construction module for using the predicted traffic congestion index... The system calculates congestion levels based on traffic flow data and travel time, and combines this with highway accident rates, congestion duration, and queue length calculated from predicted traffic flow parameters to construct an emergency lane activation decision model aimed at minimizing queue length. A dataset construction module is used to build a dataset using the decisions output by the emergency lane activation decision model as labels, combined with the output of a traffic flow congestion prediction model. An emergency lane activation decision module is used to construct an interpretable decision tree-based emergency lane activation decision model using the dataset as input, and to activate emergency lanes at corresponding road segment locations and times based on the output decisions of the interpretable decision tree-based emergency lane activation decision model.

[0015] The aforementioned method and device for activating emergency lanes on highways based on video data predicts congestion by constructing a traffic flow congestion prediction model, thereby providing real-time warnings before persistent congestion occurs. Based on the predicted results, an emergency lane activation decision model is then constructed. Using the decisions output by this model as labels, and combining them with the output of the traffic flow congestion prediction model, a dataset is built to further construct an interpretable decision tree-based emergency lane activation decision model. This model controls the location and opening time of emergency lanes. The interpretable decision tree-based emergency lane activation decision model directly displays the key factors influencing emergency lane activation decisions, exhibiting high interpretability and providing a clear and intuitive decision-making path. Attached Figure Description

[0016] Figure 1 is a flowchart illustrating a method for activating emergency lanes on highways based on video data, provided in one embodiment; Figure 2 is a heatmap provided in one embodiment; Figure 3 is a decision tree structure diagram provided in one embodiment; Figure 4 is a comparison chart of test set model evaluation metrics provided in one embodiment; Figure 5 is a comparison chart of PR curves provided in one embodiment; Figure 6 is a comparison chart of model inference time provided in one embodiment; Figure 7 is a comparison chart of total congestion time for emergency lanes with and without activation, provided in one embodiment; Figure 8 is a comparison chart of queue lengths for emergency lanes with and without activation, provided in one embodiment; Figure 9 is a structural block diagram of a highway emergency lane activation device based on video data, provided in one embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] In one embodiment, as shown in Figure 1, a method for activating emergency lanes on highways based on video data is provided, including the following steps: Step 201, collecting highway video data from preset monitoring points and extracting status information and speed information of different lanes.

[0019] Step 202: Calculate traffic flow parameter data based on the state and speed information of different lanes, and construct a traffic flow congestion prediction model using the traffic flow parameter data as input. Output the predicted future traffic flow parameter data, traffic congestion index, and travel time, and issue a real-time warning before persistent congestion occurs. The traffic flow parameter data includes traffic flow, road occupancy, and average vehicle speed.

[0020] Step 203: Calculate the congestion level index using the predicted traffic congestion index and travel time, and combine it with the highway accident rate, congestion duration and congestion queue length calculated using the predicted traffic flow parameter data to construct an emergency lane activation decision model with the goal of minimizing the congestion queue length.

[0021] Step 204: Using the decision output by the emergency lane activation decision model as a label, construct a dataset by combining the output of the traffic flow congestion prediction model.

[0022] Step 205: Using the dataset as input, construct an interpretable decision tree-based emergency lane activation decision model, and activate the emergency lane at the corresponding road segment location and time according to the output decision of the interpretable decision tree-based emergency lane activation decision model.

[0023] The aforementioned method for activating emergency lanes on highways based on video data involves constructing a traffic flow congestion prediction model to predict congestion and provide real-time warnings before persistent congestion occurs. Based on the prediction results, an emergency lane activation decision model is then built. Using the decisions from this model as labels, and combining them with the output of the traffic flow congestion prediction model, a dataset is constructed. This dataset is then used to further develop an interpretable decision tree-based emergency lane activation decision model, which controls the location and opening time of emergency lanes. This interpretable decision tree model directly demonstrates the key factors influencing emergency lane activation decisions, exhibiting high interpretability and providing a clear and intuitive decision-making path.

[0024] In one embodiment, step 201 involves collecting highway video data from preset monitoring points and extracting state and speed information for different lanes. This includes: collecting highway video data from preset monitoring points; detecting vehicles in the video data using a deep learning-based target detection algorithm; tracking vehicles in the video data using the DeepSORT algorithm; and extracting state and speed information for different lanes. The state information includes the number of vehicles in each lane, road occupancy, and traffic flow. The speed information includes the average speed of the lane and the average speed of the vehicles.

[0025] In this embodiment, the fast detection capability of the deep learning-based target detection algorithm is used to ensure that vehicles in the video can be identified in a timely and accurate manner. The DeepSORT algorithm is used to effectively associate vehicle information between different frames, thereby achieving stable tracking in the case of dense vehicles or occlusion. By combining the two, the effectiveness of the acquired highway video data is guaranteed.

[0026] In one embodiment, step 201 involves collecting highway video data from preset monitoring points and extracting status and speed information for different lanes. It also includes preprocessing the extracted status and speed information for different lanes, wherein the preprocessing includes filtering, clearing, repairing, and standardizing. Based on the preprocessed status and speed information for different lanes, a heat map is generated for visualization.

[0027] In this embodiment, the extracted state and speed information of different lanes is preprocessed to filter and remove abnormal data and fill in data gaps, thereby providing an effective data source for subsequent analysis and ensuring the accuracy of the analysis. Simultaneously, standardization is performed to eliminate the impact of differences in data units. A heatmap is generated for visualization, thus helping to effectively analyze the correlation between different data parameters.

[0028] In one embodiment, four monitoring points were set up on a highway: at the beginning of the road segment, at the starting gantry of the large section at 1 kilometer, at 2 kilometers, and at the end of the road segment. Referring to Figure 2, which shows the heat map of the first monitoring point, it can be seen that the number of vehicles in lane 1 and lane 2 are closely related, which is consistent with the situation in reality where vehicles change lanes while driving; the number of vehicles in lane 1 and lane 2 are both closely related to traffic flow, which is consistent with the actual situation; road occupancy and traffic flow, as well as the average vehicle speed and emergency lane speed, all show a high correlation.

[0029] In one embodiment, step 201 involves calculating traffic flow parameter data based on the state and speed information of different lanes, and using the traffic flow parameter data as input to construct a traffic flow congestion prediction model. The model outputs predicted future traffic flow parameter data, a traffic congestion index, and travel time, and issues a real-time warning before persistent congestion occurs. This includes: calculating traffic flow parameter data based on the state and speed information of different lanes and performing data signal filtering; dividing the filtered traffic flow parameter data into a training set and a test set; constructing a traffic flow congestion prediction model using the training set; using the test set as model input for prediction; and optimizing model performance by adjusting parameters; outputting predicted future traffic flow parameter data, a traffic congestion index, and travel time using the trained traffic flow congestion prediction model, and issuing a real-time warning before persistent congestion occurs.

[0030] In this embodiment, data noise is removed and individual outliers are suppressed by filtering the data signal, thereby making the data quality more stable. The constructed model outputs predicted future traffic flow parameters, traffic congestion index, and travel time, thereby predicting locations prone to congestion and issuing real-time warnings before persistent congestion occurs, thus achieving real-time early warning before persistent congestion occurs.

[0031] It should be noted that traffic flow parameter data includes traffic volume, road occupancy, and average vehicle speed. Traffic volume refers to the number of traffic entities passing through a cross-section per unit time; road occupancy approximately represents vehicle density. Traffic volume Q is calculated using the following formula: Q = N / B, where N is the number of vehicles passing through the cross-section during the observation time; B is the preset observation duration. Road occupancy K is calculated using the following formula: Where S is the length of the target travel segment; l y Let y be the length of road occupied by the y-th vehicle. Average vehicle speed. Calculate using the following formula: ;in, Let y be the instantaneous speed of the y-th vehicle as it passes the observation section.

[0032] The Traffic Congestion Index (TPI) is a conceptual numerical value that comprehensively reflects whether a road network is smooth or congested. The value ranges from 0 to 10, with each pair of numbers representing a level. These levels correspond to "smooth", "mostly smooth", "lightly congested", "moderately congested", and "severely congested". The higher the value, the more severe the traffic congestion.

[0033] Travel time (T) is the time it takes for a vehicle to travel through a unit road segment under free-flowing conditions.

[0034] In one embodiment, the traffic flow congestion prediction model is a LightGBM regression model, and the loss function of LightGBM includes a regularization term.

[0035] It's important to note that the LightGBM regression model doesn't change its core structure in each iteration. Instead, it optimizes the model by adding new functions to make the predictions closer to the true values. By incorporating a regularization term into the LightGBM loss function, the risk of overfitting due to the complexity of the tree structure is reduced.

[0036] In one embodiment, the emergency lane activation decision model is: ; ; ;in, For decision variables; This refers to the average length of the vehicle. Traffic flow downstream of the congestion point; For the traffic capacity of all expressways; To ensure the traffic capacity of the emergency lane; Highway accident rate; The threshold for highway accident rate; As an indicator of congestion level; Duration of congestion; The threshold for congestion duration; This refers to the length of the congestion queue. This is the threshold for congestion queue length; For vehicle speed; This is the maximum speed limit on highways.

[0037] In one embodiment, using the dataset as input, an emergency lane activation decision model with an interpretable decision tree is constructed, including: inputting the dataset, which includes a training set D and an attribute set A; generating an empty node as the root of the current subtree, denoted as Node; traversing the samples in the training set, and if the samples in the training set belong to the same category C, then marking Node as a category C node to complete the category branch construction; calculating the attribute information entropy Ent for each attribute in the attribute set, and using the attribute with the smallest Ent as the feature to divide the attribute. The training set is filtered to obtain a subset. ; Determine subset Is it an empty set? If the set is empty, then mark the branch node as a leaf node and label it with the class that has the most samples in D; if If it is not an empty set, then remove it from attribute set A. The updated attribute set is obtained. Call the recursive function The returned subtree is used as the child node of the current branch; the decision tree with Node as the root node is output, and an interpretable decision tree is constructed for the emergency lane activation decision model.

[0038] In this embodiment, by constructing an interpretable decision tree, the user can intuitively understand the model's impact on input features and the judgment process, providing a more accurate basis for decision-making.

[0039] Understandably, in an interpretable decision tree, each node represents a test on a specific attribute, each branch represents a test output, and each leaf node represents a category or decision result.

[0040] In one embodiment, the constructed decision tree is shown in Figure 3, where Congestion_Range is the congestion queue length; samples is the number of vehicles; Congestion_Duration is the congestion queue length; Traffic_State is the traffic flow state; P_safety is the safety emergency rescue requirement; class is the decision; gini is the Gini coefficient; and value is the sample class count, value = [number of samples of the inactive class, number of samples of the active class].

[0041] In one embodiment, using the dataset as input, constructing an interpretable decision tree for an emergency lane activation decision model further includes: weighting the samples in the dataset. ; ; ;in, The sample weights for activating emergency lane G; The number of vehicles passing through the cross section within the observation period; The number of vehicles passing through section G of the emergency lane during the observation period; The sample weights for G where the emergency lane was not activated; This refers to the number of vehicles passing through sections other than emergency lane G during the observation period.

[0042] In this embodiment, by weighting the samples in the dataset, the number of samples is balanced during the construction of the emergency lane activation decision model with an interpretable decision tree, so that all information is fully utilized during the construction process, thereby improving the prediction accuracy of the model.

[0043] In one embodiment, the likelihood of each node in the decision tree is calculated according to the following formula: ;in, Let n be the likelihood of node n; Let n be the number of vehicles that passed through the emergency lane G section within the observation time corresponding to node n. Let n be the number of vehicles that passed through sections other than emergency lane G during the observation period corresponding to node n.

[0044] In one embodiment, the method further includes: constructing a multi-objective optimization model for monitoring layout with the objectives of minimizing the setup cost of monitoring equipment and maximizing effective coverage, and generating an initial set of monitoring equipment locations; the multi-objective optimization model for monitoring layout is: ; ; ; ;in, Set costs for individual monitoring devices; For control variables; Number of monitoring devices; The number of locations to be covered; The number of nodes effectively covered; denoted as the total number of nodes to be covered; ECR as the effective coverage rate; S as the set of candidate locations for monitoring devices; T as the set of locations of monitoring points to be covered; r as the radius threshold of the effective coverage area; using the initial set of monitoring device locations as the initial population, the tDEA-CPBI algorithm is used for optimization, and the monitoring device locations are output.

[0045] In this embodiment, by optimizing the location of monitoring points, the optimal balance between cost-effectiveness and coverage is intelligently found. Utilizing an efficient search and iteration mechanism, the optimal layout scheme is quickly explored and converged. This effectively solves the trade-off between cost and coverage, providing scientific and efficient decision support for the planning and design of practical monitoring networks.

[0046] It should be understood that although the steps in the flowchart of Figure 1 are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in Figure 1 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0047] To verify the effectiveness of the method described in this invention, predictions were made using the present invention, an emergency lane activation decision prediction model based on Long Short-Term Memory (LSTM), Naive Bayes, an emergency lane activation decision prediction model based on Random Forest, and an emergency lane activation decision prediction model based on SVM. The results are shown in Figure 4. It can be seen that for accuracy (ACC), the Decision Tree (the present invention) performed best, reaching 0.9142, followed by LSTM and Random Forest, with accuracies of 0.8878 and 0.8829, respectively. Naive Bayes performed relatively poorly, with an accuracy of only 0.7744. SVM had a slightly lower accuracy of 0.8419. Next, the classification ability of the models was evaluated using the AUC (area of ​​origin, ROC). In this test set, the Decision Tree performed best in AUC, reaching 0.9172, followed by Random Forest and LSTM, with AUCs of 0.8854 and 0.8863, respectively. Naive Bayes had a low AUC of 0.7670, indicating relatively weak classification ability. SVM (Support Vector Machine) performed similarly to Naive Bayes on this metric, with an AUC of 0.8402. Next, the F1-Score combines the model's precision and recall, evaluating its ability to classify positive and negative samples. The results show that DecisionTree and LSTM achieved the highest F1-Scores, at 0.9138 and 0.8859, respectively. Random Forest and Naive Bayes had slightly lower F1-Scores, at 0.8831 and 0.7629, respectively. SVM performed relatively poorly on this metric, with a score of 0.8379.

[0048] As can be seen, compared to the Naive Bayes model, decision trees, LSTM, and random forests all show improvements in accuracy, area under the ROC curve, F1 score, and Matthews coefficient (Matthews correlation coefficient, referred to as Matthews coefficient in Figure 4), with decision trees performing the best. Support vector machines performed relatively poorly, with accuracy, area under the ROC curve, F1 score, and Matthews coefficient all lagging behind the other models.

[0049] Referring to Figure 5, which shows a comparison of the PR curves of the five models, it can be seen that Decision Tree performs best overall, followed by LSTM and Random Forest, while SVM performs relatively weakly on this test set.

[0050] Referring to Figure 6, it can be seen that the average inference time of this invention is optimal at 0.0022s, and in terms of error, the standard error of this invention is only 0.0001s, which is the smallest. This indicates that the inference speed of this invention is relatively stable and the inference accuracy is high.

[0051] By using congestion time and queue length as indicators, the scenarios of activating and deactivating emergency lanes in this invention are evaluated and compared separately. This quantifies the effectiveness of the evaluation model. The results are shown in Figures 7 and 8, demonstrating that activating emergency lanes at each location segment significantly improves highway traffic efficiency and alleviates traffic congestion.

[0052] In one embodiment, as shown in Figure 9, a highway emergency lane activation device based on video data is provided, including: a data sampling module 901, used to collect highway video data from preset monitoring points and extract status information and speed information of different lanes.

[0053] The congestion prediction module 902 is used to calculate traffic flow parameter data based on the state and speed information of different lanes, and to construct a traffic flow congestion prediction model with the traffic flow parameter data as input. It outputs the predicted future traffic flow parameter data, traffic congestion index and travel time, and issues a real-time warning before the occurrence of persistent congestion. The traffic flow parameter data includes traffic flow, road occupancy and average vehicle speed.

[0054] The decision model building module 903 is used to calculate the congestion level index based on the predicted traffic congestion index and travel time, and to build an emergency lane activation decision model with the goal of minimizing the congestion queue length by combining the highway accident rate, congestion duration and congestion queue length calculated based on the predicted traffic flow parameter data.

[0055] Dataset building module 904 is used to build a dataset by labeling the decisions output by the emergency lane activation decision model and combining them with the output of the traffic flow congestion prediction model.

[0056] The emergency lane activation decision module 905 is used to construct an interpretable decision tree emergency lane activation decision model with the dataset as input, and to activate the emergency lane at the corresponding road segment location and time according to the output decision of the interpretable decision tree emergency lane activation decision model.

[0057] Specific limitations regarding the video data-based highway emergency lane activation device can be found in the above-mentioned limitations on the video data-based highway emergency lane activation method, and will not be repeated here. Each module in the aforementioned video data-based highway emergency lane activation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0058] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0059] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for activating emergency lanes on highways based on video data, characterized in that, Includes the following steps: Collect highway video data from preset monitoring points and extract status and speed information for different lanes; Traffic flow parameter data is calculated based on the state and speed information of different lanes. Using the traffic flow parameter data as input, a traffic congestion prediction model is constructed, which outputs predicted future traffic flow parameter data, traffic congestion index, and travel time, and issues real-time warnings before persistent congestion occurs. The traffic flow parameter data includes traffic volume, road occupancy, and average vehicle speed. The congestion level index is calculated based on the predicted traffic congestion index and travel time. Combined with the highway accident rate, congestion duration, and congestion queue length calculated from the predicted traffic flow parameter data, an emergency lane activation decision model is constructed with the goal of minimizing congestion queue length. Using the decisions output by the emergency lane activation decision model as labels, a dataset is constructed by combining the output of the traffic flow congestion prediction model; using the dataset as input, an interpretable decision tree-based emergency lane activation decision model is constructed, and emergency lanes are activated at corresponding road segment locations and times based on the output decisions of the interpretable decision tree-based emergency lane activation decision model.

2. The method for activating emergency lanes on highways based on video data according to claim 1, characterized in that, Collect highway video data from preset monitoring points and extract status and speed information for different lanes. This includes: collecting highway video data from preset monitoring points; detecting vehicles in the video data using a deep learning-based target detection algorithm; tracking vehicles in the video data using the DeepSORT algorithm; and extracting status and speed information for different lanes. The status information includes the number of vehicles in each lane, road occupancy, and traffic flow. The speed information includes the average speed of each lane and the average speed of each vehicle.

3. The method for activating emergency lanes on highways based on video data according to claim 2, characterized in that, The process involves collecting highway video data from preset monitoring points, extracting status and speed information for different lanes, and preprocessing the extracted status and speed information for different lanes, including filtering, clearing, repairing, and standardizing. Based on the preprocessed status and speed information for different lanes, a heat map is generated for visualization.

4. The method for activating emergency lanes on highways based on video data according to claim 1, characterized in that, Traffic flow parameters are calculated based on the state and speed information of different lanes. Using this traffic flow parameter data as input, a traffic congestion prediction model is constructed. The model outputs predicted future traffic flow parameters, a traffic congestion index, and travel time, and issues real-time warnings before persistent congestion occurs. The process includes: calculating traffic flow parameters based on the state and speed information of different lanes and performing data signal filtering; dividing the filtered traffic flow parameter data into training and testing sets; constructing a traffic congestion prediction model using the training set and using the testing set as model input for prediction, optimizing model performance by adjusting parameters; and outputting predicted future traffic flow parameters, a traffic congestion index, and travel time from the trained traffic congestion prediction model, while issuing real-time warnings before persistent congestion occurs.

5. The method for activating emergency lanes on highways based on video data according to claim 1, characterized in that, The decision-making model for activating the emergency lane is as follows: in, For decision variables; This refers to the average length of the vehicle. Traffic flow downstream of the congestion point; For the traffic capacity of all expressways; To ensure the traffic capacity of the emergency lane; Highway accident rate; The threshold for highway accident rate; As an indicator of congestion level; Duration of congestion; The threshold for congestion duration; This refers to the length of the congestion queue. This is the threshold for congestion queue length; For vehicle speed; This is the maximum speed limit on highways.

6. The method for activating emergency lanes on highways based on video data according to claim 1, characterized in that, Using the dataset as input, an emergency lane activation decision model with an interpretable decision tree is constructed, including: inputting the dataset, which includes a training set D and an attribute set A; generating an empty node as the root of the current subtree, denoted as Node; traversing the samples in the training set, if the samples in the training set belong to the same category C, then marking Node as a category C node, completing the category branch construction; calculating the attribute information entropy Ent for each attribute in the attribute set, and using the attribute with the smallest Ent as the feature to divide the attribute. The training set is filtered to obtain a subset. ; Determine subset Is it an empty set? If the set is empty, then mark the branch node as a leaf node and label it with the class that has the most samples in D; if If it is not an empty set, then remove it from attribute set A. The updated attribute set is obtained. Call the recursive function The returned subtree is used as the child node of the current branch; the decision tree with Node as the root node is output, and an interpretable decision tree is constructed for the emergency lane activation decision model.

7. The method for activating emergency lanes on highways based on video data according to claim 6, characterized in that, Using the dataset as input, a decision model for emergency lane activation that can be interpreted as an emergency lane activation model is constructed, which further includes: weighting the samples in the dataset. in, The sample weights for activating emergency lane G; The number of vehicles passing through the cross section within the observation period; The number of vehicles passing through section G of the emergency lane during the observation period; The sample weights for G where the emergency lane was not activated; This refers to the number of vehicles passing through sections other than emergency lane G during the observation period.

8. The method for activating emergency lanes on highways based on video data according to claim 7, characterized in that, The likelihood of each node in the decision tree is calculated using the following formula: in, Let n be the likelihood of node n; Let n be the number of vehicles that passed through the emergency lane G section within the observation time corresponding to node n. Let n be the number of vehicles that passed through sections other than emergency lane G during the observation period corresponding to node n.

9. The method for activating emergency lanes on highways based on video data according to claim 1, characterized in that, Also includes: With the goal of minimizing the setup cost of monitoring equipment and maximizing effective coverage, a multi-objective optimization model for monitoring layout is constructed to generate an initial set of monitoring equipment locations. The multi-objective optimization model for the monitoring layout is as follows: in, Set costs for individual monitoring devices; For control variables; Number of monitoring devices; The number of locations to be covered; The number of nodes to be effectively covered; The total number of nodes to be covered; ECR is the effective coverage rate; S is the set of candidate locations for monitoring devices; T represents the set of locations to be monitored; r represents the radius threshold of the effective monitoring coverage area; using the initial set of monitoring device locations as the initial population, the tDEA-CPBI algorithm is used for optimization to output the monitoring device locations.

10. A highway emergency lane activation device based on video data, characterized in that, include: The data sampling module is used to collect highway video data from preset monitoring points and extract status and speed information for different lanes; The congestion prediction module calculates traffic flow parameter data based on the status and speed information of different lanes. Using this data as input, it constructs a traffic flow congestion prediction model, outputting predicted future traffic flow parameter data, a traffic congestion index, and travel time. It also issues real-time warnings before persistent congestion occurs. The traffic flow parameter data includes traffic volume, road occupancy, and average vehicle speed. The decision model construction module calculates congestion severity indicators using the predicted traffic congestion index and travel time. It then combines these with the highway accident rate, congestion duration, and queue length calculated from the predicted traffic flow parameter data to construct an emergency lane activation decision model aimed at minimizing queue length. The dataset building module is used to build a dataset by labeling the decisions output by the emergency lane activation decision model and combining them with the output of the traffic flow congestion prediction model. The emergency lane activation decision module is used to construct an interpretable decision tree emergency lane activation decision model with the dataset as input, and to activate the emergency lane at the corresponding road segment location and time according to the output of the interpretable decision tree emergency lane activation decision model.

Images