Event allocation method and system based on dynamically changing islands

Abstract

The invention provides an event allocation method and system based on dynamic island change, and the method comprises the steps: obtaining urban road information, and dividing urban roads according to an intelligent traffic network to obtain a plurality of regions; identifying vehicle information in the current dynamic island to obtain a vehicle distribution information graph according to an identification result, and clustering vehicles according to the vehicle distribution information graph to obtain a plurality of vehicle distribution graphs, screening a plurality of vehicle distribution diagrams to obtain a first distribution sub-graph and a second distribution sub-graph of which the density ranks the first two in the same main road according to the vehicle density, and identifying the first distribution sub-graph and the second distribution sub-graph to obtain a sub-graph distance and a lane ratio of the number of lanes occupied by each distribution sub-graph to the total number of lanes of the main road; respectively judging whether the subgraph distance is greater than a distance threshold value and whether the lane proportion is smaller than a proportion threshold value; and if not, analyzing the distribution sub-graph according to the deaggregation model, and pushing an analysis result to a traffic management department. According to the invention, the precision and efficiency of the event detection and distribution technology are improved.

Description

Technical Field

[0001] This invention relates to the field of dynamic island event processing technology, and in particular to an event allocation method and system based on dynamically changing islands. Background Technology

[0002] In the context of intelligent transportation networks, "island" does not refer to an island in the traditional geographical sense, nor to a physical traffic island used to manage traffic in transportation facilities (such as a central island, directional island, or safety island). As a metaphorical term, "island" is usually used to refer to specific events or states in a transportation network, especially those traffic anomalies that are relatively isolated and prominent, similar to islands in the ocean, i.e., "island events."

[0003] Traffic congestion events are called "island events" because, within the overall traffic flow, a congested section of road acts as a relatively isolated area. Vehicles within this area exhibit significantly different driving patterns compared to surrounding normal road sections, much like an island differs from its surrounding ocean environment. It may be caused by a specific reason, such as a traffic accident, road construction, or vehicle malfunction—a single component event—leading to a sudden change in traffic flow, speed, and other indicators of that road segment, forming a relatively isolated area of ​​abnormal traffic conditions.

[0004] Currently, in traffic congestion island event network systems, events can typically only be detected at a broad level, such as locating events to a large area of ​​the network, rather than pinpointing specific units, such as a particular main road. This makes traffic maintenance difficult and time-consuming, causing existing event detection and allocation technologies to fall short of practical requirements in terms of accuracy and efficiency. For example, in a traffic network, when traffic congestion occurs, traditional methods can only identify that a traffic problem has occurred on the island, but cannot quickly and accurately pinpoint which direction, which main road, or other specific traffic equipment (such as traffic lights) is malfunctioning. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide an event allocation method and system based on dynamically changing islands, which solves the technical problem that the event detection and allocation technology in the prior art is difficult to meet the actual needs in terms of accuracy and efficiency.

[0006] This invention provides, in one aspect, an event allocation method based on dynamically changing islands, comprising:

[0007] The system acquires urban road information and divides urban roads into multiple regions based on the intelligent transportation network. Each region corresponds to a dynamic island, and each dynamic island includes the region information of the corresponding region. The region information includes road information, traffic facility information, and vehicle information. The vehicle information includes vehicle location and speed. The traffic facility information includes traffic lights, road sensors, and surveillance cameras.

[0008] Identify vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results. Cluster vehicles according to the vehicle distribution information map to obtain several vehicle distribution maps. Filter several vehicle distribution maps to obtain the first and second distribution sub-maps with the highest density on the same main road based on vehicle density. Identify the first and second distribution sub-maps to obtain the sub-map distance and the lane ratio of the number of lanes occupied by each distribution sub-map to the total number of lanes on the main road.

[0009] Determine whether the sub-image distance is greater than the distance threshold and whether the lane proportion is less than the proportion threshold, respectively.

[0010] If the subgraph distance is less than the distance threshold or the lane ratio is greater than the ratio threshold, the distribution subgraph is analyzed according to the de-aggregation model to obtain the event analysis results, and the event analysis results are pushed to the traffic management department to avoid traffic congestion in the area. The event analysis results include accident scenes and traffic light anomalies.

[0011] The aforementioned event allocation method based on dynamically changing islands obtains vehicle information within a region by dividing the dynamic islands, thereby generating a vehicle distribution information map. Each region corresponds to a dynamic island, and each region includes multiple main roads. Based on the vehicle distribution information map, vehicles are clustered to obtain several vehicle distribution maps, which are then filtered to obtain the first and second distribution sub-maps with the highest density on the same main road. Based on the sub-map distance and lane ratio, the specific areas requiring intervention are further located, and the event analysis results are pushed to the traffic management department to prevent traffic congestion in the area. This enables dynamic detection and analysis, improving the accuracy and efficiency of event detection and allocation technology.

[0012] In addition, the event allocation method based on dynamically changing islands according to the present invention may also have the following additional technical features:

[0013] Furthermore, before the steps of determining whether the subgraph distance is greater than the distance threshold and whether the lane proportion is less than the proportion threshold, the following steps are also included:

[0014] Get the current time;

[0015] Determine whether it is a peak period based on the current time. Peak periods include the morning peak and the evening peak.

[0016] If so, then proceed with the step of analyzing the distribution subgraph based on the de-aggregation model to obtain the event analysis results.

[0017] Furthermore, the step of clustering vehicles based on the vehicle distribution information map to obtain several vehicle distribution maps includes:

[0018] The vehicle speed is used to identify the slow driving area, which is the area where the vehicle speed is lower than the preset speed. The current area range is determined based on the vehicle speed slow driving area, and the number of lanes in the current area range is determined as the number of lanes occupied by slow driving.

[0019] Based on lane information, determine whether the number of lanes occupied by the slow-moving vehicle is greater than the preset number of lanes;

[0020] If so, a vehicle distribution map will be constructed based on the current vehicle distribution within the area.

[0021] Furthermore, the steps of filtering several vehicle distribution maps to obtain the first and second distribution sub-maps with the highest vehicle density on the same main road include:

[0022] The vehicle locations are identified based on the vehicle distribution map, and the distances between adjacent vehicles are obtained based on the vehicle locations. Based on the vehicle distances, vehicles on the same main road are divided and classified to obtain multiple distribution sub-maps.

[0023] Identify the vehicle density in each distribution submap to obtain the first and second distribution submaps with the highest density on the same main road based on vehicle density.

[0024] Furthermore, the step of obtaining the vehicle distance between adjacent vehicles based on the vehicle location, and then dividing and classifying vehicles on the same main road according to the vehicle distance to obtain multiple distribution sub-maps includes:

[0025] Obtain the current vehicle's location, and using this vehicle's location as the origin, identify adjacent vehicles and obtain the vehicle distance between the current vehicle and adjacent vehicles;

[0026] Determine whether the distance to the vehicle is less than a distance threshold;

[0027] If not, the current vehicle is a boundary vehicle on the vehicle distribution map, and the process of identifying adjacent vehicles is stopped.

[0028] If so, then the adjacent vehicles whose distance is less than the distance threshold are classified as secondary vehicles, the secondary vehicles are classified with the current vehicle in the same distribution subgraph, the secondary vehicles are updated to the origin, and the process returns to the step of identifying adjacent vehicles and obtaining the vehicle distance between the current vehicle and the adjacent vehicles, until the vehicle distance is not less than the distance threshold.

[0029] Furthermore, the step of classifying adjacent vehicles whose distance is less than a distance threshold as level two vehicles includes:

[0030] Obtain the relative positional relationship between the secondary vehicle and the current vehicle;

[0031] Based on the relative positional relationship, determine whether the secondary vehicle and the current vehicle are in the same lane;

[0032] If they are not in the same lane, then determine whether there is a physical barrier between the secondary vehicle and the current vehicle, the physical barrier including a fence;

[0033] If a physical barrier exists, the current vehicle is a boundary vehicle on the vehicle distribution map. The currently located secondary vehicle is discarded, and the identification of adjacent vehicles is stopped.

[0034] Furthermore, the steps of identifying vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results include:

[0035] The improved YOLOv5s detection algorithm is obtained by replacing the C3 module in the YOLOv5s model with the GSConv module, introducing the SE attention mechanism at the lateral connection of the neck network in the YOLOv5s model, and designing the D-C3 module in the backbone network.

[0036] The improved YOLOv5s detection algorithm is used to identify vehicle information in the current dynamic island, and a vehicle distribution information map is obtained based on the identification results.

[0037] Furthermore, the SE attention mechanism includes squeezing operations, incentive operations, and weighted operations, among which...

[0038] The extrusion operation includes:

[0039] Global average pooling is used to transform the two-dimensional features of each channel into real numbers. By compressing the features in the spatial dimension, each real number is given a global receptive field, thereby achieving the fusion of global contextual information. The specific expression is as follows:

[0040]

[0041] In the formula, W, H, and C represent the width, height, and channel of the feature map, respectively; u represents the feature map of a channel.

[0042] Incentive operations include:

[0043] Two fully connected layers are used to determine the correlation between different channels, thereby dynamically distributing corresponding weights to each channel according to its importance. The specific expression is as follows:

[0044] s = F ex (z,W)=σ(W2δ(W1z));

[0045] In the formula, z is the output feature vector after the activation operation; W1 and W2 are two fully connected layers; σ is the ReLU activation function; and δ is the Sigmoid activation function.

[0046] Weighted operations include:

[0047] By multiplying the weight matrix output from the activation operation with the channels of the feature map using weighted multiplication, the information representation of important features is highlighted, thereby improving the feature extraction capability of the detection model. The specific expression is as follows:

[0048] X c =F scale (u c ,s c ) = s c u c .

[0049] Another aspect of the present invention provides an event allocation system based on dynamically changing islands, the system comprising:

[0050] The acquisition module is used to acquire urban road information. According to the intelligent transportation network, urban roads are divided into multiple areas. Each area corresponds to a dynamic island. Each dynamic island includes the area information of the corresponding area. The area information includes road information, traffic facility information and vehicle information. The vehicle information includes vehicle location and speed. The traffic facility information includes traffic lights, road sensors and monitoring cameras.

[0051] The identification module is used to identify vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results, cluster vehicles according to the vehicle distribution information map to obtain several vehicle distribution maps, filter several vehicle distribution maps to obtain a first distribution sub-map and a second distribution sub-map with the highest density in the same main road based on vehicle density, and identify the first distribution sub-map and the second distribution sub-map to obtain the sub-map distance and the lane ratio of the number of lanes occupied by each distribution sub-map to the total number of lanes of the main road.

[0052] The judgment module is used to determine whether the sub-map distance is greater than the distance threshold and whether the lane ratio is less than the ratio threshold, respectively.

[0053] The first execution module is used to analyze the distribution subgraph according to the de-aggregation model to obtain event analysis results if the subgraph distance is less than the distance threshold or the lane ratio is greater than the ratio threshold, and push the event analysis results to the traffic management department to avoid traffic congestion in the area. The event analysis results include accident scenes and traffic light anomalies.

[0054] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the event allocation method based on dynamically changing islands as described above.

[0055] In another aspect, the present invention provides a data processing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the event allocation method based on dynamically changing islands as described above. Attached Figure Description

[0056] Figure 1 This is a flowchart of the event allocation method based on dynamically changing the island according to the present invention;

[0057] Figure 2 Here is a structural diagram of the GSConv module;

[0058] Figure 3 Here is a diagram of the SE attention module structure;

[0059] Figure 4 This is a structural diagram of the D-Bottleneck module;

[0060] Figure 5 This is a structural diagram of the D-C3 module;

[0061] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0062] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0063] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used in this application includes any and all combinations of one or more of the associated listed items.

[0064] To facilitate understanding of the present invention, several embodiments are given below. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the present invention more thorough and complete.

[0065] Example 1

[0066] Please see Figure 1 The figure shows an event allocation method based on dynamically changing islands in the first embodiment of the present invention, the method including steps S101 to S106:

[0067] S101. Obtain urban road information, divide urban roads into multiple areas according to the intelligent transportation network, each area corresponds to a dynamic island, and each dynamic island includes the area information of the corresponding area.

[0068] In this embodiment, each dynamic island includes multiple main roads for vehicle traffic. Furthermore, in order to improve the accuracy and timeliness of the event allocation system, the regional information includes road information, traffic facility information, and vehicle information. The vehicle information includes vehicle location and speed; the traffic facility information includes traffic lights, road sensors, and surveillance cameras.

[0069] S102. Identify vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results.

[0070] Regarding vehicle information identification, in practical monitoring scenarios, research on multi-target vehicle detection algorithms often encounters problems such as large detection model size, susceptibility to occlusion interference, and missed detection of small-scale vehicles. Therefore, this technical solution adopts a detection algorithm based on deep learning methods, using the YOLOv5s algorithm as a baseline for optimization and improvement. Specifically, three measures are taken to improve YOLOv5s: replacing the GSConv module, introducing the SE attention module, and designing the D-C3 module by drawing inspiration from the dense connection concept of DenseNet.

[0071] YOLOv5s mainly consists of four parts: Input, Backbone, Neck, and Prediction. At the Input end, YOLOv5s preprocesses the dataset using Mosaic data augmentation, adaptive anchor box calculation, and adaptive image scaling to improve the robustness and generalization of the network model. The Backbone mainly includes the CBS, C3, and SPPF modules, which are the main components for feature extraction of image targets. The Neck uses a Feature Pyramid Network (FPN) composed of CBS and C3, and a Path Aggregation Network (PAN). Combining FPN and PAN forms PANet to fuse shallow localization information with deep semantic information, enhancing feature extraction. The Prediction end uses the CIOU bounding box loss function and non-maximum suppression to predict the output.

[0072] Furthermore, in this embodiment, the C3 module in the YOLOv5s model is replaced with the GSConv module, an SE attention mechanism is introduced at the lateral connection of the neck network in the YOLOv5s model, and a D-C3 module is designed in the backbone network to obtain an improved YOLOv5s detection algorithm; the improved YOLOv5s detection algorithm is used to identify vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results.

[0073] First, the C3 module in the YOLOv5s model is replaced with the GSConv module. In this embodiment, only the neck network is lightened by replacing the C3 module in the YOLOv5s model with the GSConv module, making the model lightweight and balancing detection accuracy. Specifically, the GSConv convolutional module mainly consists of standard convolution, depthwise separable convolution, a Concat module, and a channel shuffle module. The input features first enter the standard convolution and are split into two paths along the channel dimension. One path of features is fed into the depthwise separable convolution, and then the Concat module fuses the features from the two paths. Finally, the channel shuffle module enhances feature interaction. The GSConv convolutional module uses the Shuffle operation to fuse the feature information extracted by the standard convolution into the features after the depthwise separable convolution. This combines the advantages of both standard convolution and depthwise separable convolution, giving the GSConv module a certain advantage among lightweight convolutional modules. The GSConv module structure is as follows: Figure 2 As shown.

[0074] In convolutional neural networks, increasing the number of convolutional layers is often used to improve the detection performance of the model. However, this also leads to a significant increase in the computational cost and number of parameters, resulting in low processing efficiency. In real-world traffic monitoring scenarios, vehicle detection models are typically deployed on embedded devices, which have limited memory and computational power. Therefore, this application introduces the GSConv module to compress the model, ensuring that the model complexity and inference speed meet deployment requirements.

[0075] Secondly, an SE attention mechanism is introduced at the lateral connections of the neck network in the YOLOv5s model. In human vision, to make full use of limited visual information, the human eye often selectively focuses on more important information in an image while ignoring other useless information. The attention mechanism in convolutional neural networks is also proposed based on this characteristic of human vision. The attention mechanism enables the model to learn a set of weight parameters and allocate the importance of each feature of the image according to the weight parameters, making the model pay more attention to important feature information, suppressing the interference of irrelevant feature information, and improving the detection performance and generalization ability of the model.

[0076] Currently, deep learning convolutional neural networks (CNNs) need to consider both channel and spatial information when processing images. However, compared to spatial information, the network correlation in the channel dimension is more important. Therefore, this technical solution introduces a channel-based attention mechanism (SE) module. The SE attention mechanism learns the relationships between feature map channels and assigns different weight values ​​to each channel, enabling the detection model to focus on the most important channel information in the feature map and reduce interference from useless channel features. The SE attention mechanism mainly consists of three operations: Squeeze, Excitation, and Scale, as shown in the specific structure below. Figure 3 As shown.

[0077] Specifically, the Squeeze operation uses global average pooling to transform the two-dimensional features of each channel into real numbers. By compressing the features in the spatial dimension, each real number is given a global receptive field, thus achieving the fusion of global contextual information. The specific expression is as follows:

[0078]

[0079] In the formula, W, H, and C represent the width, height, and channel of the feature map, respectively; u represents the feature map of a channel.

[0080] The excitation operation uses two fully connected layers to determine the correlation between different channels, thereby dynamically distributing corresponding weights to each channel according to its importance. The specific expression is as follows:

[0081] S = F ex (z,W)=σ(W2δ(W1z));

[0082] In the formula, z is the output feature vector after the activation operation; W1 and W2 are two fully connected layers; σ is the ReLU activation function; and δ is the Sigmoid activation function.

[0083] The Scale operation multiplies the weight matrix output from the Excitation operation with the channels of the feature map using a multiplicative weighting method. This highlights the information of important features and improves the feature extraction capability of the detection model. The specific expression is as follows:

[0084] X c =F scale (u c ,s c ) = s c u c .

[0085] In video images captured on actual traffic roads, vehicles often occlude each other when densely distributed, and other objects such as trees can also obscure vehicles. This can lead to confusion between the feature information of the vehicle target and the feature information of the occluding objects. This complex feature information can interfere with the network model's ability to quickly and accurately identify vehicle targets, resulting in false positives and false negatives in vehicle detection. Adding an SE attention mechanism to the model network can make the model pay more attention to the feature channel information of the vehicle itself and suppress the interference of other unimportant channel feature information. Based on this, this application introduces an SE attention mechanism at the lateral connection of the neck network in the YOLOv5s model to enhance the model's ability to identify occluded vehicles.

[0086] Furthermore, a D-C3 module is designed in the backbone network to obtain an improved YOLOv5s detection algorithm. In traffic monitoring videos, the size of vehicle targets in a fixed camera view varies greatly, and distant vehicle targets appear smaller, causing the detection model to easily lose feature information and resulting in missed detections. This problem mainly arises because the shallow information of vehicle targets in the network is not fully utilized. As the network depth increases, the feature information acquired by the shallow network gradually weakens or is lost, leading to the model's inability to identify targets. To enhance the utilization of shallow feature information, this application integrates the idea of ​​DenseNet dense connections into the YOLOv5s network, designing the D-Bottleneck module and the D-C3 module, with the specific structure as follows: Figure 4 and Figure 5 As shown, the D-Bottleneck module structure is similar to the dense block structure in the DenseNet network, containing four convolutional layers and establishing dense connections between multiple layers. Unlike the dense block structure, the D-Bottleneck module designed in this application uses an Add operation to achieve feature fusion between layers. Although Add and Concat are implemented differently, both can integrate information from multiple feature maps, achieving the same effect of multi-layer interconnected feature fusion, while requiring fewer parameters and less computation.

[0087] To maximize feature reuse, theoretically all four C3 modules in the backbone network should be replaced with D-C3 modules. However, the D-C3 modules employ a dense connection method with multiple branches, resulting in higher computational complexity compared to the C3 modules. Replacing all of them would significantly increase the model's computational load and make it prone to overfitting. Therefore, this application chooses to replace only the first C3 module to improve detection capabilities. After the D-C3 module extracts shallow feature information, it achieves feature reuse through dense connections. The extracted shallow feature information is richer and can be better fed into the deep network, allowing the model to fully utilize shallow semantic features and improve the detection capability for small-scale vehicles.

[0088] S103. Cluster the vehicles according to the vehicle distribution information map to obtain several vehicle distribution maps.

[0089] As a specific example, step S103 specifically includes steps S1031-S1033:

[0090] S1031. Identify the slow-moving area of ​​the vehicle based on the vehicle speed. The slow-moving area is the area where the vehicle speed is lower than the preset speed. Determine the current area range based on the slow-moving area and determine the number of lanes in the current area range as the number of lanes occupied by slow-moving vehicles.

[0091] As a concrete example, traffic congestion is often related to reduced vehicle speed. Understandably, when speeds fall below a certain minimum, congestion is more likely to occur. For instance, on highways, the minimum speed limit for passenger cars is 80 km / h. When speeds fall below 80 km / h, congestion is highly probable when traffic volume reaches a certain level. Similarly, in urban roads, due to high traffic volume, monitoring systems can determine the minimum speed to avoid congestion by monitoring lanes. This minimum speed is set as the system's preset speed. When speeds fall below this minimum, a slow-moving zone will appear. Furthermore, since the number of passable lanes is also a crucial factor in alleviating traffic congestion, when a slow-moving zone appears, it's necessary to further monitor and obtain the number of lanes currently occupied by slow-moving vehicles. By comparing this number with the total number of lanes on the main road, the number of passable lanes can be determined.

[0092] S1032. Determine whether the number of lanes occupied by slow-moving vehicles is greater than the preset number of lanes based on lane information.

[0093] In practice, on urban arterial roads, there are generally no fewer than two lanes, meaning the total number of lanes on the arterial road is greater than or equal to two. Therefore, when slow-moving vehicles occupy one lane, the impact on the overall traffic capacity of the road is within an acceptable range, because under certain circumstances, another lane can provide normal traffic capacity. However, when slow-moving vehicles occupy two or more lanes, the impact on the overall traffic capacity of the arterial road is often significant. Therefore, in this embodiment, the preset number of lanes is one.

[0094] If so, proceed to step S1033;

[0095] If not, return to step S102 to continue monitoring the road condition;

[0096] S1033. Construct a vehicle distribution map based on the current vehicle distribution within the area.

[0097] S104. Filter several vehicle distribution maps to obtain the first and second distribution sub-maps with the highest density on the same main road based on vehicle density. Identify the first and second distribution sub-maps to obtain the sub-map distance and the lane ratio of the number of lanes in each distribution sub-map to the total number of lanes on the main road.

[0098] Based on actual statistics, when two or more traffic jams occur on the same main road, traffic congestion will occur. Therefore, in this embodiment, the vehicle density in the vehicle distribution map is identified based on vehicle density and vehicle distribution map to predict areas where traffic jams may occur. Specifically: the vehicle location is identified based on the vehicle distribution map, and the distance between adjacent vehicles is obtained based on the vehicle location to divide and classify vehicles on the same main road according to the vehicle distance, thereby obtaining multiple distribution sub-maps; the vehicle density in each distribution sub-map is identified, and the first and second distribution sub-maps with the highest density on the same main road are obtained based on the vehicle density.

[0099] In essence, the system determines whether or not traffic congestion has occurred or is about to occur by identifying the distances between adjacent vehicles. Specifically: The system obtains the current vehicle's position and uses this position as the origin; it then identifies adjacent vehicles and obtains the distances between the current vehicle and its adjacent vehicles; it checks if the distances are less than a distance threshold; if the distances are not less than the threshold, the current vehicle is considered a boundary vehicle in the vehicle distribution map, and the identification of adjacent vehicles is stopped; if the distances are less than the threshold, adjacent vehicles with distances less than the threshold are classified as secondary vehicles, and these secondary vehicles are grouped with the current vehicle into the same distribution submap. The origin of the secondary vehicles is updated, and the system returns to the previous steps of identifying adjacent vehicles and obtaining the distances between the current vehicle and its adjacent vehicles, until the distances are no less than the distance threshold. As a concrete example, the distance threshold can be the width of two independent lanes, and the distribution submap is obtained by identifying and dividing the lanes based on vehicle distances. However, in practice, when the distance to a vehicle is less than the distance threshold, it is still possible to identify a vehicle in the oncoming lane due to the close proximity. Therefore, a secondary judgment is needed to exclude oncoming vehicles from the identified secondary vehicles, because under normal circumstances, oncoming vehicles will not affect the driving of vehicles in the current lane (i.e., the forward lane). Specifically:

[0100] Obtain the relative positional relationship between the secondary vehicle and the current vehicle; determine whether the secondary vehicle and the current vehicle are in the same lane based on the relative positional relationship; if they are not in the same lane, determine whether there is a physical barrier between the secondary vehicle and the current vehicle, including fences; if there is a physical barrier, the current vehicle is a boundary vehicle of the vehicle distribution map, discard the currently located secondary vehicle, and exit the identification of adjacent vehicles.

[0101] S105. Determine whether the sub-image distance is greater than the distance threshold and whether the lane ratio is less than the ratio threshold.

[0102] As a specific example, in practice, it is necessary to distinguish between daily hours and peak hours. During peak hours, due to the large traffic volume, even a brief stop or pause can cause severe traffic congestion. Therefore, special handling is required during peak hours. Specifically: obtain the current time; determine whether it is a peak hour based on the current time, which includes morning and evening peak hours; if the current time is a peak hour, there is no need to check whether the subgraph distance is greater than the distance threshold or whether the lane ratio is less than the ratio threshold. Instead, directly execute the step of analyzing the distribution subgraph based on the de-aggregation model to obtain the event analysis results, so as to avoid timely handling of dynamic events and prevent traffic congestion.

[0103] S106. If the subgraph distance is less than the distance threshold or the lane ratio is greater than the ratio threshold, the distribution subgraph is analyzed according to the de-aggregation model to obtain the event analysis results, and the event analysis results are pushed to the traffic management department to avoid traffic congestion in the area. The event analysis results include accident scenes and traffic light anomalies.

[0104] In summary, the event allocation method based on dynamically changing islands in the above embodiments of the present invention obtains vehicle information within a region by dividing the dynamic islands, thereby acquiring a vehicle distribution information map. Each region corresponds to a dynamic island, and each region includes multiple main roads. Based on the vehicle distribution information map, vehicles are clustered to obtain several vehicle distribution maps, which are then filtered to obtain the first and second distribution sub-maps with the highest density on the same main road. Based on the sub-map distance and lane ratio, the specific areas requiring adjustment are further located, and the event analysis results are pushed to the traffic management department to avoid traffic congestion in the area. This enables dynamic detection and analysis, improving the accuracy and efficiency of event detection and allocation technology.

[0105] Example 2

[0106] A second embodiment of the present invention provides an event allocation system based on dynamically changing islands, comprising:

[0107] The acquisition module is used to acquire urban road information. According to the intelligent transportation network, urban roads are divided into multiple areas. Each area corresponds to a dynamic island. Each dynamic island includes the area information of the corresponding area. The area information includes road information, traffic facility information and vehicle information. The vehicle information includes vehicle location and speed. The traffic facility information includes traffic lights, road sensors and monitoring cameras.

[0108] The identification module is used to identify vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results, cluster vehicles according to the vehicle distribution information map to obtain several vehicle distribution maps, filter several vehicle distribution maps to obtain a first distribution sub-map and a second distribution sub-map with the highest density in the same main road based on vehicle density, and identify the first distribution sub-map and the second distribution sub-map to obtain the sub-map distance and the lane ratio of the number of lanes occupied by each distribution sub-map to the total number of lanes of the main road.

[0109] The judgment module is used to determine whether the sub-map distance is greater than the distance threshold and whether the lane ratio is less than the ratio threshold, respectively.

[0110] The first execution module is used to analyze the distribution subgraph according to the de-aggregation model to obtain event analysis results if the subgraph distance is less than the distance threshold or the lane ratio is greater than the ratio threshold, and push the event analysis results to the traffic management department to avoid traffic congestion in the area. The event analysis results include accident scenes and traffic light anomalies.

[0111] In summary, the event allocation system based on dynamically changing islands in the above embodiments of the present invention obtains vehicle information within a region by dividing the dynamic islands, thereby acquiring a vehicle distribution information map. Each region corresponds to a dynamic island, and each region includes multiple main roads. Based on the vehicle distribution information map, vehicles are clustered to obtain several vehicle distribution maps, which are then filtered to obtain the first and second distribution sub-maps with the highest density on the same main road. Based on the sub-map distance and lane ratio, the specific areas requiring intervention are further located, and the event analysis results are pushed to the traffic management department to avoid traffic congestion in the area. This enables dynamic detection and analysis, improving the accuracy and efficiency of event detection and allocation technology.

[0112] Furthermore, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the methods described above.

[0113] Furthermore, embodiments of the present invention also propose a data processing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the methods described above.

[0114] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0115] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0116] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0117] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0118] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An event allocation method based on dynamically changing islands, characterized in that, include: The system acquires urban road information and divides urban roads into multiple regions based on the intelligent transportation network. Each region corresponds to a dynamic island, and each dynamic island includes the region information of the corresponding region. The region information includes road information, traffic facility information, and vehicle information. The vehicle information includes vehicle location and speed. The traffic facility information includes traffic lights, road sensors, and surveillance cameras. Identify vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results. Cluster vehicles according to the vehicle distribution information map to obtain several vehicle distribution maps. Filter several vehicle distribution maps to obtain the first and second distribution sub-maps with the highest density on the same main road based on vehicle density. Identify the first and second distribution sub-maps to obtain the sub-map distance and the lane ratio of the number of lanes occupied by each distribution sub-map to the total number of lanes on the main road. Determine whether the sub-image distance is greater than the distance threshold and whether the lane proportion is less than the proportion threshold, respectively. If the subgraph distance is less than the distance threshold or the lane ratio is greater than the ratio threshold, the distribution subgraph is analyzed according to the de-aggregation model to obtain the event analysis results, and the event analysis results are pushed to the traffic management department to avoid traffic congestion in the area. The event analysis results include accident scenes and traffic light anomalies.

2. The event allocation method based on dynamically changing islands according to claim 1, characterized in that, Before the steps of determining whether the subgraph distance is greater than the distance threshold and whether the lane proportion is less than the proportion threshold, the following steps are also included: Get the current time; Determine whether it is a peak period based on the current time. Peak periods include the morning peak and the evening peak. If so, then proceed with the step of analyzing the distribution subgraph based on the de-aggregation model to obtain the event analysis results.

3. The event allocation method based on dynamically changing islands according to claim 1, characterized in that, The steps of clustering vehicles based on the vehicle distribution information map to obtain several vehicle distribution maps include: The vehicle speed is used to identify the slow driving area, which is the area where the vehicle speed is lower than the preset speed. The current area range is determined based on the vehicle speed slow driving area, and the number of lanes in the current area range is determined as the number of lanes occupied by slow driving. Based on lane information, determine whether the number of lanes occupied by the slow-moving vehicle is greater than the preset number of lanes; If so, a vehicle distribution map will be constructed based on the current vehicle distribution within the area.

4. The event allocation method based on dynamically changing islands according to claim 3, characterized in that, The steps for filtering several vehicle distribution maps to obtain the first and second sub-maps of the top two vehicle density along the same main road include: The vehicle locations are identified based on the vehicle distribution map, and the distances between adjacent vehicles are obtained based on the vehicle locations. Based on the vehicle distances, vehicles on the same main road are divided and classified to obtain multiple distribution sub-maps. Identify the vehicle density in each distribution submap to obtain the first and second distribution submaps with the highest density on the same main road based on vehicle density.

5. The event allocation method based on dynamically changing islands according to claim 4, characterized in that, The steps of obtaining the vehicle distance between adjacent vehicles based on the vehicle location, and then dividing and classifying vehicles on the same main road based on the vehicle distance to obtain multiple distribution sub-maps include: Obtain the current vehicle's location, and using this vehicle's location as the origin, identify adjacent vehicles and obtain the vehicle distance between the current vehicle and adjacent vehicles; Determine whether the distance to the vehicle is less than a distance threshold; If not, the current vehicle is a boundary vehicle on the vehicle distribution map, and the process of identifying adjacent vehicles is stopped. If so, then the adjacent vehicles whose distance is less than the distance threshold are classified as secondary vehicles, the secondary vehicles are classified with the current vehicle in the same distribution subgraph, the secondary vehicles are updated to the origin, and the process returns to the step of identifying adjacent vehicles and obtaining the vehicle distance between the current vehicle and the adjacent vehicles, until the vehicle distance is not less than the distance threshold.

6. The event allocation method based on dynamically changing islands according to claim 5, characterized in that, The steps for classifying adjacent vehicles whose distance is less than a distance threshold as level 2 vehicles include: Obtain the relative positional relationship between the secondary vehicle and the current vehicle; Based on the relative positional relationship, determine whether the secondary vehicle and the current vehicle are in the same lane; If they are not in the same lane, then determine whether there is a physical barrier between the secondary vehicle and the current vehicle, the physical barrier including a fence; If a physical barrier exists, the current vehicle is a boundary vehicle on the vehicle distribution map. The currently located secondary vehicle is discarded, and the identification of adjacent vehicles is stopped.

7. The event allocation method based on dynamically changing islands according to claim 1, characterized in that, The steps for identifying vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results include: The improved YOLOv5s detection algorithm is obtained by replacing the C3 module in the YOLOv5s model with the GSConv module, introducing the SE attention mechanism at the lateral connection of the neck network in the YOLOv5s model, and designing the D-C3 module in the backbone network. The improved YOLOv5s detection algorithm is used to identify vehicle information in the current dynamic island, and a vehicle distribution information map is obtained based on the identification results.

8. The event allocation method based on dynamically changing islands according to claim 7, characterized in that, SE attention mechanisms include squeezing operations, incentive operations, and weighted operations. The extrusion operation includes: Global average pooling is used to transform the two-dimensional features of each channel into real numbers. By compressing the features in the spatial dimension, each real number is given a global receptive field, thereby achieving the fusion of global contextual information. The specific expression is as follows: In the formula, W, H, and C represent the width, height, and channel of the feature map, respectively; u represents the feature map of a channel. Incentive operations include: Two fully connected layers are used to determine the correlation between different channels, thereby dynamically distributing corresponding weights to each channel according to its importance. The specific expression is as follows: s=F ex (z,W)=σ(W2δ(W1z)); In the formula, z is the output feature vector after the activation operation; W1 and W2 are two fully connected layers; σ is the ReLU activation function; and δ is the Sigmoid activation function. Weighted operations include: By multiplying the weight matrix output from the activation operation with the channels of the feature map using weighted multiplication, the information representation of important features is highlighted, thereby improving the feature extraction capability of the detection model. The specific expression is as follows: X c =F scale (u c ,s c )=s c u c 。 9. An event allocation system based on dynamically changing islands, characterized in that, The system includes: The acquisition module is used to acquire urban road information. According to the intelligent transportation network, urban roads are divided into multiple areas. Each area corresponds to a dynamic island. Each dynamic island includes the area information of the corresponding area. The area information includes road information, traffic facility information and vehicle information. The vehicle information includes vehicle location and speed. The traffic facility information includes traffic lights, road sensors and monitoring cameras. The identification module is used to identify vehicle information in the current dynamic island to obtain a vehicle distribution information map based on the identification results, cluster vehicles according to the vehicle distribution information map to obtain several vehicle distribution maps, filter several vehicle distribution maps to obtain a first distribution sub-map and a second distribution sub-map with the highest density in the same main road based on vehicle density, and identify the first distribution sub-map and the second distribution sub-map to obtain the sub-map distance and the lane ratio of the number of lanes occupied by each distribution sub-map to the total number of lanes of the main road. The judgment module is used to determine whether the sub-map distance is greater than the distance threshold and whether the lane ratio is less than the ratio threshold, respectively. The first execution module is used to analyze the distribution subgraph according to the de-aggregation model to obtain event analysis results if the subgraph distance is less than the distance threshold or the lane ratio is greater than the ratio threshold, and push the event analysis results to the traffic management department to avoid traffic congestion in the area. The event analysis results include accident scenes and traffic light anomalies.

Images