Interactive mining processing method and system based on dynamic images

Abstract

The embodiments of the present invention relate to the field of intelligent transportation technology, and more specifically, to an interactive mining processing method and system based on dynamic images. The embodiments of the present invention determine the influence weight by combining the spatiotemporal state information of the road network area of ​​the target traffic road network server and the image description vector of the dynamic traffic monitoring image, and can comprehensively and multi-levelly evaluate the importance of the dynamic traffic monitoring image. The target flow trend characterization vector based on the target cross-trajectory relationship network is obtained from the traffic road network database, and the correlation characteristics of the pre-mined trajectory relationship network and the flow trend characterization vector are utilized to make the analysis of traffic conditions more reliable. The target weighted weight is obtained by combining the two and used for traffic event mining, which improves the accuracy of traffic event mining. This helps to grasp the traffic conditions more accurately, warn of traffic events in advance, optimize traffic management, reduce traffic congestion, improve transportation efficiency and enhance traffic safety.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of intelligent transportation technology, and in particular, to a method and system for interactive mining processing based on dynamic images. Background Art

[0002] In the field of modern transportation, traffic congestion, frequent accidents and other problems are becoming increasingly serious, posing a huge challenge to efficient traffic management. Traditional traffic management methods have obvious deficiencies in analyzing traffic conditions and predicting traffic events. On the one hand, it is difficult to fully evaluate the importance of traffic monitoring information and accurately judge the changing trend of traffic conditions by relying solely on single-dimensional data. For example, it is impossible to deeply explore the complex relationships behind traffic behavior based only on the spatiotemporal state information of the region or the content of a single image. On the other hand, there is a lack of effective methods for integrating multiple traffic data for traffic event mining, resulting in untimely and inaccurate warnings of traffic events. Summary of the invention

[0003] In order to at least overcome the above-mentioned deficiencies in the prior art, one of the objectives of the embodiments of the present invention is to provide a method and system for interactive mining processing based on dynamic images.

[0004] An embodiment of the present invention provides an interactive mining processing method based on dynamic images, the method is applied to an interactive mining processing system, and the method includes:

[0005] In response to a road network data mining task of a target traffic road network server, determining the influence weight of each dynamic traffic monitoring image at at least two attention levels according to the spatiotemporal state information of the road network area of ​​the target traffic road network server and an image description vector of at least one dynamic traffic monitoring image;

[0006] Obtaining a target flow trend representation vector corresponding to the target cross-trajectory relationship network from a traffic network database through the target cross-trajectory relationship network of each dynamic traffic monitoring image; wherein the traffic network database includes correlation features between at least one trajectory relationship network and a flow trend representation vector corresponding to each trajectory relationship network, and the flow trend representation vector corresponding to the trajectory relationship network is obtained by mining the traffic behavior data corresponding to the trajectory relationship network in advance;

[0007] Obtaining target weightings of each dynamic traffic monitoring image according to the target traffic trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels;

[0008] Traffic event mining is performed on each dynamic traffic monitoring image according to the target weighted weight of each dynamic traffic monitoring image.

[0009] Optionally, obtaining target weightings of each dynamic traffic monitoring image based on the target traffic trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels includes:

[0010] For each dynamic traffic monitoring image, according to the influence weights of each attention level corresponding to the dynamic traffic monitoring image and the eigenvalues ​​corresponding to the influence weights of each attention level, each eigenvalue in the original feature matrix is ​​updated to obtain a target feature matrix;

[0011] According to the target flow trend characterization vector and the target feature matrix, cross-weighting of the influence weights of each attention level is implemented to obtain the target weighted weight of the dynamic traffic monitoring image.

[0012] Optionally, the road network data mining task belongs to a target traffic operation and maintenance project among at least two currently activated traffic operation and maintenance projects;

[0013] The method of obtaining the original feature matrix includes:

[0014] According to the target traffic operation and maintenance project to which the road network data mining task belongs, the original feature matrix corresponding to the target traffic operation and maintenance project is obtained from a mapping list between each traffic operation and maintenance project and the original feature matrix;

[0015] Among them, each traffic operation and maintenance project corresponds to its own original feature matrix, and the original feature matrix includes each eigenvalue in the cross-track relationship network corresponding to the target traffic operation and maintenance project and the default parameters of the eigenvalues.

[0016] Optionally, the target cross-trajectory relationship network represents a cross-weighted rule between at least two preset feature values; after determining the influence weight of each dynamic traffic monitoring image at at least two attention levels based on the spatiotemporal state information of the road network area of ​​the target traffic road network server and the image description vector of at least one dynamic traffic monitoring image, the method further includes:

[0017] For each dynamic traffic monitoring image, obtaining a target intersection trajectory relationship network corresponding to the dynamic traffic monitoring image;

[0018] According to the mapping list between the influence weights of each attention level of the dynamic traffic monitoring image and each eigenvalue in the target cross-trajectory relationship network, the target cross-trajectory relationship network corresponding to the dynamic traffic monitoring image, as well as each eigenvalue and the influence weight of each attention level are integrated, and each eigenvalue is associated with an influence weight corresponding to an attention level.

[0019] Optionally, the step of obtaining a target flow trend representation vector corresponding to the target intersection trajectory relationship network from a traffic road network database through the target intersection trajectory relationship network of each dynamic traffic monitoring image includes:

[0020] According to the target cross-trajectory relationship network, a flow trend representation vector corresponding to the target cross-trajectory relationship network is roamed in the traffic network database;

[0021] If the traffic network database includes a target flow rate trend representation vector corresponding to the target intersection trajectory relationship network, obtaining the target flow rate trend representation vector included in the traffic network database;

[0022] If the traffic network database does not include a flow trend representation vector corresponding to the target cross-trajectory relationship network, the traffic behavior data corresponding to the target cross-trajectory relationship network is mined as the target flow trend representation vector, and the correlation feature between the target cross-trajectory relationship network and the target flow trend representation vector is cached in the traffic network database.

[0023] Optionally, the target traffic trend characterization vector is generated by:

[0024] Processing the target intersection track relationship network through a spatiotemporal graph neural network to obtain traffic behavior data corresponding to the target intersection track relationship network;

[0025] Data mining is performed on the traffic behavior data to obtain a behavior element chain corresponding to the traffic behavior data, and a deductive description between each flow node and each characteristic value is analyzed based on the behavior element chain corresponding to the traffic behavior data to generate a target flow trend representation vector corresponding to the target cross-trajectory relationship network.

[0026] Optionally, caching the correlation feature between the target cross-trajectory relationship network and the target traffic trend representation vector into the traffic network database includes:

[0027] If the total number of flow trend representation vectors corresponding to the traffic network database is greater than a first set value, obtaining the noise coefficient of each flow trend representation vector in the traffic network database within a target period, cleaning the flow trend representation vectors whose noise coefficients meet the target requirements, and caching the correlation characteristics between the target flow trend representation vector and the target cross-trajectory relationship network in the traffic network database;

[0028] If the total number of flow trend representation vectors corresponding to the traffic network database is not greater than a first set value, the correlation feature between the target flow trend representation vector and the target intersection trajectory relationship network is cached in the traffic network database.

[0029] Optionally, the method further includes:

[0030] If the total number of flow trend characterization vectors corresponding to the traffic network database is greater than a first set value, determining step information for cleaning the flow trend characterization vectors in the traffic network database within a preset period;

[0031] If the step size information meets the optimization requirements, a flow trend update report is issued.

[0032] Optionally, the road network data mining task belongs to a target traffic operation and maintenance project among at least two currently activated traffic operation and maintenance projects; the traffic network database also includes a current congestion trend of each flow trend characterization vector corresponding to the target traffic operation and maintenance project, and the current congestion trend is used to indicate whether it is tracked by any task in the target traffic operation and maintenance project;

[0033] After obtaining the target flow trend representation vector corresponding to the target cross-trajectory relationship network from the traffic road network database through the target cross-trajectory relationship network of each dynamic traffic monitoring image, the method further includes: annotating the current congestion trend of the target flow trend representation vector as a tracking label;

[0034] After obtaining the target weighted weights of each dynamic traffic monitoring image based on the target flow trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels, the method also includes: annotating the current congestion trend of the target flow trend characterization vector as a label to be tracked.

[0035] An embodiment of the present invention also provides an interactive mining processing system, including a processor and a memory and a bus connected to the processor; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the above-mentioned interactive mining processing method based on dynamic images.

[0036] The embodiment of the present invention further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned interactive mining processing method based on dynamic images is implemented.

[0037] The embodiment of the present invention determines the influence weight by combining the spatiotemporal state information of the road network area of ​​the target traffic road network server and the image description vector of the dynamic traffic monitoring image, and can comprehensively and multi-levelly evaluate the importance of the dynamic traffic monitoring image. The target flow trend characterization vector based on the target cross-trajectory relationship network is obtained from the traffic road network database, and the correlation characteristics of the pre-mined trajectory relationship network and the flow trend characterization vector are utilized to make the analysis of traffic conditions more reliable. The target weighted weight is obtained by combining the two and used for traffic event mining, which improves the accuracy of traffic event mining. This helps to grasp the traffic conditions more accurately, warn of traffic events in advance, optimize traffic management, reduce traffic congestion, improve transportation efficiency and enhance traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 A flowchart of an interactive mining method based on dynamic images provided by an embodiment of the present invention.

[0040] Figure 2 A block diagram of an interactive mining processing system provided by an embodiment of the present invention.

[0041] icon:

[0042] 100 - interactive mining processing system; 101 - processor; 102 - memory; 103 - bus. DETAILED DESCRIPTION

[0043] The exemplary embodiments disclosed in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the embodiments of the present invention are shown in the accompanying drawings, it should be understood that the embodiments of the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the embodiments of the present invention and to fully convey the scope of the embodiments of the present invention to those skilled in the art.

[0044] In order to better understand the above technical scheme, the technical scheme of the embodiment of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiment of the present invention and the specific features in the embodiments are detailed descriptions of the technical scheme of the embodiment of the present invention, rather than limitations on the technical scheme of the embodiment of the present invention. In the absence of conflict, the embodiment of the present invention and the technical features in the embodiments can be combined with each other.

[0045] Figure 1 This is a flowchart of an interactive mining method based on dynamic images provided according to an embodiment of the present invention, which is applied to an interactive mining system and includes steps 110 to 130.

[0046] Step 110, in response to the road network data mining task of the target traffic road network server, determine the influence weight of each dynamic traffic monitoring image at at least two attention levels based on the spatiotemporal state information of the road network area of ​​the target traffic road network server and the image description vector of at least one dynamic traffic monitoring image.

[0047] (1) Road network data mining task of target traffic network server

[0048] In the fields of transportation, smart cities, and the Internet of Things, the target traffic network server is responsible for mining the road network data. This task aims to extract valuable information from massive amounts of traffic-related data in order to achieve goals such as effective management, optimization, and monitoring of traffic conditions. For example, in the traffic management of a large city, the target traffic network server may need to mine the causes of traffic congestion, predict changes in traffic flow, etc.

[0049] The spatiotemporal status information of the road network area is a key element. It contains the status of a specific road network area at different times and spaces, such as the traffic flow, vehicle speed and other status information of a certain area in the morning rush hour (time dimension) in the busy area of ​​the city center (spatial dimension). For example, during the morning rush hour (7-9 o'clock) on weekdays in a certain downtown commercial district, the average speed is 20-30 kilometers per hour, and the traffic flow reaches 500-800 vehicles per hour. These are the specific numerical manifestations of the spatiotemporal status information of the area.

[0050] The image description vector of a dynamic traffic monitoring image is a quantitative representation of the content of the dynamic traffic monitoring image. For example, the image description vector can represent the number of vehicles in the image, the type of vehicle (such as cars, trucks, etc.), the direction of vehicle travel, and other information. For example, in a dynamic traffic monitoring image, the image description vector obtained through image analysis contains information such as the number of vehicles is 50, of which cars account for 80%, trucks account for 20%, and most vehicles are traveling eastward.

[0051] (2) Determine the influence weight of each dynamic traffic monitoring image at at least two attention levels

[0052] Attention levels can be divided into different dimensions, such as macro and micro. At the macro level, attention may focus on the overall traffic flow trend of the entire traffic area, the smoothness of major roads, etc. At the micro level, attention may focus on the behavior of individual vehicles (such as illegal lane changes, sudden braking, etc.), local traffic conditions at intersections, etc.

[0053] Taking the macro level as an example, if a dynamic traffic monitoring image covers a major transportation hub area, and the image description vector shows that the traffic flow in the area is large and the vehicle speed is slow, and at the same time the spatiotemporal status information of the road network area of ​​the target traffic road network server shows that the area is usually a high-incidence area of ​​traffic congestion during this period, then the impact weight of this image at the macro level may be relatively high, such as set to 0.8.

[0054] At the micro level, if a dynamic traffic monitoring image shows that a vehicle has violated the lane change behavior (obtained through image description vector analysis), and the spatiotemporal state information of the road network area in the area shows that the area is strengthening traffic order management recently, then the impact weight of this image at the micro level may be 0.6. This determination of impact weights at different levels helps to more comprehensively and meticulously consider the importance of dynamic traffic monitoring images in the entire traffic data mining.

[0055] Step 120 , obtaining a target flow trend representation vector corresponding to the target intersection trajectory relationship network from a traffic road network database through the target intersection trajectory relationship network of each dynamic traffic monitoring image.

[0056] The traffic network database includes correlation features between at least one trajectory relationship network and a flow trend representation vector corresponding to each trajectory relationship network. The flow trend representation vector corresponding to the trajectory relationship network is obtained by mining the traffic behavior data corresponding to the trajectory relationship network in advance.

[0057] For example, each dynamic traffic monitoring image has its corresponding target intersection trajectory network. This network describes the intersection of the trajectories of vehicles or other traffic participants in the traffic scene. For example, in a dynamic traffic monitoring image of an intersection, the target intersection trajectory network may show how vehicle trajectories in different directions intersect, merge, or diverge. For example, at an intersection where four main roads intersect, the vehicle trajectories from east to west intersect with the vehicle trajectories from south to north at the center of the intersection, and a certain proportion of vehicles turn to other directions at the intersection. This is what the target intersection trajectory network describes.

[0058] In addition, the traffic network database contains rich information, among which the correlation characteristics between at least one trajectory relationship network and the flow trend representation vector corresponding to each trajectory relationship network are important components. The flow trend representation vector is obtained by mining the traffic behavior data corresponding to the trajectory relationship network in advance. For example, for the trajectory relationship network of a specific roundabout, after long-term data mining, it is found that when the proportion of vehicles in certain specific trajectories reaches a certain value (such as 30% of the total number of vehicles entering the inner ring from the outer ring), the flow trend representation vector may indicate that the traffic flow begins to increase slowly and the average vehicle speed begins to decrease.

[0059] When the target intersection trajectory relationship network of each dynamic traffic monitoring image is obtained, the corresponding target flow trend representation vector can be obtained from the traffic network database. For example, a dynamic traffic monitoring image shows a complex multi-intersection trajectory relationship network. According to this relationship network, the corresponding flow trend representation vector can be obtained by searching in the traffic network database. This vector may indicate that the traffic flow in the area will increase by 20% and the vehicle speed will decrease by 10% in the next period of time.

[0060] Step 130: obtain the target weighting weight of each dynamic traffic monitoring image based on the target flow trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels; and perform traffic event mining on each dynamic traffic monitoring image based on the target weighting weight of each dynamic traffic monitoring image.

[0061] It can be understood that the target weighted weight of each dynamic traffic monitoring image can be calculated based on the target traffic trend characterization vector obtained in step 120 and the influence weights of each dynamic traffic monitoring image at multiple attention levels determined in step 110. For example, there are dynamic traffic monitoring images A, B, and C. The influence weight of image A at the macro level is 0.7, and the influence weight at the micro level is 0.5. The corresponding target traffic trend characterization vector shows that the traffic flow in the area will change greatly (weight is 0.6); the influence weight of image B at the macro level is 0.5, and the influence weight at the micro level is 0.4. The corresponding target traffic trend characterization vector shows that the traffic flow changes less (weight is 0.3); the influence weight of image C at the macro level is 0.6, and the influence weight at the micro level is 0.3. The corresponding target traffic trend characterization vector shows that the traffic flow is stable (weight is 0.2).

[0062] Calculate the target weighting of image A: (0.7×0.6+0.5×0.6)=0.72; the target weighting of image B: (0.5×0.3+0.4×0.3)=0.27; the target weighting of image C: (0.6×0.2+0.3×0.2)=0.18.

[0063] According to the calculated target weighted weights of each dynamic traffic monitoring image, traffic event mining can be performed on these images. Images with higher weights will be given more attention and resources in traffic event mining. For example, for image A, since its target weighted weight is higher at 0.72, it will be given priority in traffic event mining. If the image description vector of image A shows that there is an abnormal gathering of vehicles (which may indicate a traffic accident or temporary road control), combined with its higher weight, the system will be more inclined to judge this situation as an important traffic event for further analysis, such as determining the specific location of the accident, the scope of impact, etc.

[0064] For image C, its target weight is 0.18, which is relatively low. Even though the image description vector shows some small changes in the direction of vehicle travel, it may be considered as normal traffic fluctuations and will not be the focus of attention in traffic event mining unless there is more evidence that these small changes may cause larger traffic problems.

[0065] Through the implementation of the above steps 110 to 130, the interactive mining processing system can effectively utilize the relevant information of the target traffic network server, the various information of the dynamic traffic monitoring image, and the information in the traffic network database to achieve accurate mining of traffic events, which is of great significance for applications such as IoT management and vehicle monitoring in the fields of transportation, smart cities, and IoT. For example, in IoT management, the operating status of IoT devices (such as smart traffic lights, etc.) can be better managed through the results of traffic event mining; in vehicle monitoring, abnormal traffic events related to vehicles can be discovered in a timely manner and corresponding measures can be taken to improve traffic safety and efficiency. In 5G vehicle-mounted terminal equipment applications, the mined traffic information can also be fed back to the vehicle-mounted terminal in a timely manner to provide more accurate road condition information for vehicle-mounted navigation, etc., to assist the driver in making better driving decisions.

[0066] This helps build a more intelligent and efficient traffic management and monitoring system to meet the needs of modern transportation and smart city development. It takes into account multiple factors, avoids the one-sidedness that may be caused by single factor analysis, and improves the accuracy and reliability of traffic event mining through the fusion and weighted processing of multi-faceted information.

[0067] The following describes the technical solutions described in steps 110 to 130 through three specific application scenario examples.

[0068] 1. NB-IOT management application scenarios

[0069] In a city's smart traffic management system, NB-IOT (Narrowband Internet of Things) technology is used to manage traffic equipment and collect data. The target traffic network server undertakes the task of road network data mining, aiming to optimize the traffic conditions of the entire city and efficiently manage numerous IoT devices.

[0070] The target traffic network server has detailed spatiotemporal status information of the road network area. For example, in the road network of a certain commercial district, during the morning rush hour (7-9 o'clock) on weekdays, the traffic flow is extremely large and the average speed is slow, while during the noon period (12-14 o'clock), the traffic flow decreases and the speed increases. At the same time, many sensors are deployed in the city, which will collect dynamic traffic monitoring images. The image description vector contains information such as the number of vehicles in the image, the type of vehicle, and the speed range of the vehicle. For example, a dynamic traffic monitoring image shows that a large number of small cars are gathered on a certain road section with extremely low speeds. Based on the spatiotemporal status information of these road network areas and the image description vectors of the dynamic traffic monitoring images, the influence weight of each dynamic traffic monitoring image at different attention levels can be determined. At the macro level, the weight of images in areas with large traffic flow is higher; at the micro level, if special vehicle types or abnormal speeds appear in the image, the weight will be adjusted accordingly.

[0071] Each dynamic traffic monitoring image has a corresponding target intersection trajectory network, which describes the intersection of vehicle trajectories on the road. Through these target intersection trajectory networks, the corresponding target flow trend representation vectors can be obtained from the traffic network database. The traffic network database stores the correlation characteristics between each trajectory network and the flow trend representation vector, which is obtained by mining the traffic behavior data corresponding to the trajectory network. For example, for a complex intersection trajectory network, the corresponding flow trend representation vector in the database may show that the traffic flow will increase rapidly during a specific time period (such as 5-7 pm on weekdays), and the average waiting time of vehicles will increase.

[0072] According to the target flow trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels, the target weighted weight of each dynamic traffic monitoring image is calculated. For example, an image in a traffic congestion area and related to an upcoming traffic peak may get a higher target weighted weight. Finally, traffic events are mined for each dynamic traffic monitoring image based on these target weighted weights. In the NB-IOT management scenario, if it is found that a certain section of road is about to be severely congested, the duration of the smart traffic lights on the section can be adjusted in time through NB-IOT technology, or nearby vehicles can be notified to slow down to avoid traffic paralysis.

[0073] 2. Vehicle Monitoring Application Scenarios

[0074] In the vehicle monitoring system, the road network data mining task of the target traffic network server focuses on all-round monitoring of vehicle driving conditions. The spatiotemporal state information of the road network area includes information such as traffic restrictions, average vehicle speed, and traffic volume of different sections at different times. For example, near the school area, the speed limit during the school hours (7:30-8:30, 16:30-17:30) is 30 kilometers per hour, and the traffic volume is large and complex.

[0075] Dynamic traffic monitoring images come from surveillance cameras installed on the road and some on-board cameras. The image description vector contains information such as the appearance features of the vehicle (color, model, etc.), the status of the occupants (whether they are wearing seat belts, etc.), and the vehicle's driving trajectory. For example, a dynamic traffic monitoring image shows a red car without seat belts and frequent lane changes. Based on this information, the influence weight of each dynamic traffic monitoring image at different attention levels is determined. At the macro level, images located in special areas such as school areas have higher weights; at the micro level, the weight of images with violations of vehicle behavior will also increase.

[0076] The target intersection trajectory network of each dynamic traffic monitoring image can reflect the intersection of vehicle driving trajectories. The target flow trend representation vector is obtained from the traffic network database through these target intersection trajectory networks. The correlation features in the traffic network database help to accurately obtain the flow trend representation vector related to the current vehicle driving trajectory. For example, for a curve trajectory network where traffic accidents often occur, its corresponding flow trend representation vector may show that excessive speed at the curve is the main cause of the accident, and the accident rate is higher in certain weather (such as rainy days).

[0077] The target weighted weight is obtained based on the target flow trend representation vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels. A dynamic traffic monitoring image corresponding to a vehicle that is in a high-accident section and has violated regulations will have a higher target weighted weight. Traffic events are mined for each dynamic traffic monitoring image based on these target weighted weights. In the vehicle monitoring scenario, if a vehicle is found to have dangerous driving behavior and is in a dangerous section, the traffic management department or the owner of the relevant vehicle can be notified in time to ensure road traffic safety.

[0078] 3. Application scenarios of in-vehicle navigation

[0079] In the vehicle navigation system, the road network data mining task of the target traffic road network server is to provide vehicles with more accurate navigation suggestions. The spatiotemporal state information of the road network area includes real-time traffic conditions, road construction conditions, and road grades of different roads. For example, due to road construction, a main road will be operated as a single lane in the next week (8:00-18:00), resulting in slow speed and heavy traffic.

[0080] Dynamic traffic monitoring images can come from data shared by on-board cameras and road monitoring systems. The image description vector contains information such as the visual conditions of the road (such as whether there is fog that affects the line of sight), traffic signs on the road, etc. For example, a dynamic traffic monitoring image shows that the traffic sign at a certain intersection is blocked. Based on this information, the influence weight of each dynamic traffic monitoring image at different attention levels is determined. At the macro level, images related to main roads have a higher weight; at the micro level, the weight of images involving abnormal traffic signs will also be taken seriously.

[0081] The target intersection trajectory network of each dynamic traffic monitoring image can reflect the intersection of vehicle trajectories. Through these target intersection trajectory networks, the target flow trend representation vector is obtained from the traffic network database. For example, for a frequently congested roundabout trajectory network, the flow trend representation vector in the database shows that vehicles travel slowly in the roundabout during the evening rush hour (17:00-19:00) on weekdays, and the average time for each vehicle to pass the roundabout increases.

[0082] The target weighting weight is obtained based on the target flow trend representation vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels. For example, an image showing road construction ahead and in a traffic congestion period will have a higher target weighting weight. Traffic events are mined for each dynamic traffic monitoring image based on these target weighting weights. In the in-vehicle navigation scenario, if the road ahead is congested or dangerous, the in-vehicle navigation system can re-plan the route for the driver in time to avoid the vehicle getting stuck in congestion or dangerous sections.

[0083] In some optional embodiments, the target weighted weight of each dynamic traffic monitoring image is obtained based on the target flow trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels, including: for each dynamic traffic monitoring image, based on the influence weights of each attention level corresponding to the dynamic traffic monitoring image and the eigenvalues ​​corresponding to the influence weights of each attention level, each eigenvalue in the original feature matrix is ​​updated to obtain a target feature matrix; based on the target flow trend characterization vector and the target feature matrix, the influence weights of each attention level are cross-weighted to obtain the target weighted weight of the dynamic traffic monitoring image.

[0084] Furthermore, the road network data mining task belongs to a target traffic operation and maintenance project among at least two currently activated traffic operation and maintenance projects; the method for obtaining the original feature matrix includes: according to the target traffic operation and maintenance project to which the road network data mining task belongs, obtaining the original feature matrix corresponding to the target traffic operation and maintenance project from a mapping list between each traffic operation and maintenance project and the original feature matrix; wherein each traffic operation and maintenance project corresponds to its own original feature matrix, and the original feature matrix includes each eigenvalue in the cross-track relationship network corresponding to the target traffic operation and maintenance project and the default parameters of the eigenvalues.

[0085] In the above technical solution, the two important factors of the target flow trend representation vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels are first clarified. They are the key basis for obtaining the target weighted weights of each dynamic traffic monitoring image.

[0086] For each dynamic traffic monitoring image, in the process of updating each eigenvalue in the original feature matrix according to the influence weights of each attention level and the eigenvalues ​​corresponding to the influence weights of each attention level to obtain the target feature matrix, many considerations are involved. The acquisition of the original feature matrix is ​​not arbitrary, but is determined according to the target traffic operation and maintenance project to which the road network data mining task belongs. Each traffic operation and maintenance project corresponds to its own original feature matrix, which contains each eigenvalue in the cross-track relationship network corresponding to the target traffic operation and maintenance project and the default parameters of the eigenvalues. For example, there are traffic operation and maintenance projects A and B. Traffic operation and maintenance project A mainly targets the traffic conditions in the central area of ​​the city. Its original feature matrix may contain eigenvalues ​​related to the intersection trajectories of roads in the area, such as the inflow and outflow ratio of vehicles at a specific intersection (for example, the default parameter of the inflow and outflow ratio is 1:1), the average speed of vehicles in lanes in different directions (for example, the default parameter of the average speed of the main road is 50 kilometers per hour), etc.; if traffic operation and maintenance project B is aimed at the traffic conditions in the suburbs of the city, the eigenvalues ​​in its original feature matrix will be related to the characteristics of suburban roads, such as the vehicle capacity corresponding to the road width (for example, the default vehicle capacity of a certain section of road is 20 vehicles per lane per kilometer), etc.

[0087] After obtaining the original feature matrix, take a dynamic traffic monitoring image as an example. For example, this image corresponds to two attention levels: macro and micro. At the macro level, the impact weight is 0.6, and the corresponding eigenvalue is the overall traffic flow in the area (for example, the current monitored traffic value is 800 vehicles per hour, and its default parameter in the original feature matrix is ​​1000 vehicles per hour); at the micro level, the impact weight is 0.4, and the corresponding eigenvalue is the proportion of a specific type of vehicle (such as large trucks) (for example, the current monitored large trucks account for 10%, and its default parameter in the original feature matrix is ​​8%). Then, when updating these eigenvalues ​​in the original feature matrix, the macro level adjusts the eigenvalues ​​related to traffic flow according to the impact weight and the relationship between the current value and the default parameter. Similarly, at the micro level, the eigenvalues ​​related to the proportion of large trucks are adjusted, and finally the target feature matrix is ​​obtained.

[0088] Next, according to the target flow trend representation vector and the target feature matrix, the cross-weighting of the influence weights of each attention level is implemented to obtain the target weighted weight of the dynamic traffic monitoring image. Continuing with the above example, for example, the target flow trend representation vector shows that the traffic flow in the area has an increasing trend in the future (for example, an increase of 20%). This trend will affect the cross-weighting of the influence weights of different attention levels. In the cross-weighting process, the traffic flow characteristics at the macro level are directly related to the traffic flow trend in the target flow trend representation vector. Its weight will be recalculated according to the specific value in the target flow trend representation vector (an increase of 20%) and the adjusted traffic flow characteristic value in the target feature matrix (for example, adjusted to 850 vehicles per hour) during cross-weighting, and the same is true for the micro level. Through a complex calculation process, the target weighted weight of the dynamic traffic monitoring image is finally obtained.

[0089] In this way, first of all, by determining the original feature matrix according to different traffic operation and maintenance projects, the basic eigenvalues ​​and default parameters can be accurately set according to different regions or types of traffic conditions, which lays the foundation for subsequent accurate eigenvalue updates and weight calculations. For example, the traffic conditions in the city center and suburbs are very different, and setting different original feature matrices can better meet actual needs. Secondly, the comprehensive consideration of the influence weights and eigenvalues ​​of each attention level enables the importance of dynamic traffic monitoring images to be fully and meticulously reflected when obtaining the target weighted weight. For example, when considering the overall macro traffic flow and the proportion of micro specific vehicles at the same time, combined with the target traffic trend representation vector, the weight of the image in the entire traffic data processing can be more accurately judged, so as to more reasonably allocate resources in subsequent operations such as traffic event mining, improve the accuracy and efficiency of traffic management and monitoring, and help build a more intelligent and efficient traffic management system.

[0090] In some alternative technical solutions, the target cross-trajectory relationship network represents a cross-weighted rule between at least two preset eigenvalues; after determining the influence weight of each dynamic traffic monitoring image at at least two attention levels based on the spatiotemporal state information of the road network area of ​​the target traffic road network server and the image description vector of at least one dynamic traffic monitoring image, the method further includes: for each dynamic traffic monitoring image, obtaining the target cross-trajectory relationship network corresponding to the dynamic traffic monitoring image; based on a mapping list between the influence weights of each attention level of the dynamic traffic monitoring image and each eigenvalue in the target cross-trajectory relationship network, integrating the target cross-trajectory relationship network corresponding to the dynamic traffic monitoring image, and each eigenvalue and the influence weight of each attention level, and each eigenvalue is associated with an influence weight corresponding to an attention level.

[0091] In the above technical solution, the target cross-trajectory relationship network represents the cross-weighted rule between at least two preset feature values, which plays a key role in connection and calculation basis in the whole technical solution.

[0092] After determining the influence weight of each dynamic traffic monitoring image at at least two attention levels based on the spatiotemporal state information of the road network area of ​​the target traffic road network server and the image description vector of at least one dynamic traffic monitoring image, a series of operations are performed for each dynamic traffic monitoring image. First, the target intersection trajectory relationship network corresponding to the dynamic traffic monitoring image is obtained. For example, in an urban traffic network, the target intersection trajectory relationship network corresponding to a certain dynamic traffic monitoring image may cover the vehicle flow direction characteristic values ​​and vehicle speed characteristic values ​​at the intersection. For example, the vehicle flow direction characteristic values ​​have the flow ratios of the four directions of east to west, west to east, south to north, and north to south, which are 30%, 25%, 20%, and 25% respectively, and the vehicle speed characteristic value is 30 kilometers per hour on average.

[0093] Then, the integration operation is performed based on the mapping list between the influence weights of each attention level of the dynamic traffic monitoring image and each eigenvalue in the target cross-trajectory relationship network. For example, the dynamic traffic monitoring image has two attention levels, macro and micro, with the influence weight of the macro level being 0.7 and the influence weight of the micro level being 0.3. In the mapping list, the vehicle flow eigenvalue is associated with the influence weight of the macro level, and the vehicle speed eigenvalue is associated with the influence weight of the micro level. Then, the target cross-trajectory relationship network corresponding to the dynamic traffic monitoring image and each eigenvalue are integrated with the influence weight of each attention level, so that each eigenvalue is associated with an influence weight corresponding to an attention level. Taking the vehicle flow eigenvalue as an example, since it is associated with the influence weight of 0.7 at the macro level, in subsequent calculations or analyses, the weight of 0.7 will be used to measure the importance of the vehicle flow eigenvalue in the overall traffic condition assessment; for the vehicle speed eigenvalue, since it is associated with the influence weight of 0.3 at the micro level, the weight of 0.3 will be considered when analyzing specific vehicle behaviors or local traffic efficiency.

[0094] It can be seen that by clarifying the cross-weighting rules between the eigenvalues ​​in the target cross-trajectory relationship network, an orderly framework is provided for the integration and analysis of traffic data. After determining the influence weights of the dynamic traffic monitoring image at different attention levels, the eigenvalues ​​in the target cross-trajectory relationship network are further associated and integrated with the influence weights, so that each eigenvalue can be reasonably weighted at the corresponding attention level. This helps to more accurately assess traffic conditions. For example, in the above example, the direction of traffic flow and the driving status of vehicles can be accurately analyzed based on the weights of different eigenvalues. At the same time, it also improves the accuracy of traffic data mining, and these weighted eigenvalues ​​can be more effectively used for judgment in operations such as traffic event mining, thereby optimizing traffic management and improving the operating efficiency of the entire traffic system.

[0095] In the next step, the target intersection trajectory relationship network of each dynamic traffic monitoring image is used to obtain the target flow trend representation vector corresponding to the target intersection trajectory relationship network from the traffic network database, including: according to the target intersection trajectory relationship network, the flow trend representation vector corresponding to the target intersection trajectory relationship network is wandered in the traffic network database; if the traffic network database includes the target flow trend representation vector corresponding to the target intersection trajectory relationship network, the target flow trend representation vector included in the traffic network database is obtained; if the traffic network database does not include the flow trend representation vector corresponding to the target intersection trajectory relationship network, the traffic behavior data corresponding to the target intersection trajectory relationship network is mined as the target flow trend representation vector, and the correlation feature between the target intersection trajectory relationship network and the target flow trend representation vector is cached in the traffic network database.

[0096] Furthermore, the method for generating the target traffic trend representation vector includes: processing the target cross-trajectory relationship network through a spatiotemporal graph neural network to obtain traffic behavior data corresponding to the target cross-trajectory relationship network; performing data mining on the traffic behavior data to obtain a behavior element chain corresponding to the traffic behavior data, and analyzing the deductive description between each traffic node and each characteristic value based on the behavior element chain corresponding to the traffic behavior data to generate a target traffic trend representation vector corresponding to the target cross-trajectory relationship network.

[0097] In addition, the caching of the correlation characteristics between the target cross-trajectory relationship network and the target flow trend characterization vector in the traffic network database includes: if the total number of flow trend characterization vectors corresponding to the traffic network database is greater than a first set value, obtaining the noise coefficient of each flow trend characterization vector in the traffic network database within a target period, cleaning the flow trend characterization vectors whose noise coefficients meet the target requirements, and caching the correlation characteristics between the target flow trend characterization vector and the target cross-trajectory relationship network in the traffic network database; if the total number of flow trend characterization vectors corresponding to the traffic network database is not greater than the first set value, caching the correlation characteristics between the target flow trend characterization vector and the target cross-trajectory relationship network in the traffic network database.

[0098] It can be understood that the process of obtaining the target flow trend representation vector corresponding to the target cross-trajectory relationship network from the traffic road network database through the target cross-trajectory relationship network of each dynamic traffic monitoring image includes multiple links.

[0099] First, based on the target intersection trajectory relationship network, the flow trend representation vector corresponding to the target intersection trajectory relationship network is roamed in the traffic network database. For example, in the traffic network database of an urban traffic network, the target intersection trajectory relationship network covers information such as the intersection of vehicle trajectories at a specific intersection, and the database stores a large number of flow trend representation vectors corresponding to different intersections and sections. Taking the target intersection trajectory relationship network of a certain cross intersection as an example, the system will search for the flow trend representation vector that matches it in the database.

[0100] If the traffic network database includes a target flow trend representation vector corresponding to the target intersection trajectory relationship network, for example, the flow trend representation vector of the cross intersection is found, then the target flow trend representation vector included in the traffic network database is obtained. For example, this target flow trend representation vector includes the traffic flow change values ​​of the intersection at different time periods, such as the traffic flow during the morning rush hour (7-9 o'clock) increases from 500 vehicles per hour to 700 vehicles per hour, and the average speed decreases from 40 kilometers per hour to 30 kilometers per hour.

[0101] If the traffic network database does not include the flow trend representation vector corresponding to the target intersection trajectory network, then the traffic behavior data corresponding to the target intersection trajectory network is mined as the target flow trend representation vector. Taking the target intersection trajectory network of a newly planned complex intersection as an example, its traffic behavior data may include the turning ratio of vehicles in different directions, the acceleration and deceleration frequency of vehicles, etc. The target intersection trajectory network is processed by the spatiotemporal graph neural network to obtain the traffic behavior data corresponding to the target intersection trajectory network. For example, the spatiotemporal graph neural network analyzes the state changes of the vehicle trajectory of the intersection in different time slices (such as every 5 minutes as a time slice) to obtain the dynamic behavior data of the vehicle.

[0102] Then, data mining is performed on the traffic behavior data to obtain the behavior element chain corresponding to the traffic behavior data. For example, the vehicle turning ratio, acceleration and deceleration frequency and other elements in the traffic behavior data constitute the behavior element chain. According to the behavior element chain corresponding to the traffic behavior data, the description between each flow node and each characteristic value is analyzed to generate the target flow trend representation vector corresponding to the target cross-trajectory relationship network. For example, based on the relationship between the vehicle turning ratio and the flow node (such as the flow entering and leaving the intersection), as well as the characteristic values ​​(such as the width of different lanes, the duration of traffic lights, etc.), a target flow trend representation vector containing future traffic flow change trends (such as the traffic flow in the next hour may increase by 10%), vehicle speed change trends (such as the average vehicle speed may decrease by 5 kilometers per hour), etc. is generated.

[0103] In the process of caching the correlation features between the target cross-trajectory relationship network and the target flow trend characterization vector into the traffic network database, if the total number of flow trend characterization vectors corresponding to the traffic network database is greater than the first set value, for example, the first set value is 1000. Obtain the noise coefficient of each flow trend characterization vector in the traffic network database within the target period (such as the past day). For example, a flow trend characterization vector has a high noise coefficient due to abnormal data in some time periods caused by a failure of the data acquisition equipment. Clean the flow trend characterization vectors whose noise coefficients meet the target requirements (such as the noise coefficient is greater than 0.5), and then cache the correlation features between the target flow trend characterization vector and the target cross-trajectory relationship network into the traffic network database. If the total number of flow trend characterization vectors corresponding to the traffic network database is not greater than the first set value, the correlation features between the target flow trend characterization vector and the target cross-trajectory relationship network are directly cached into the traffic network database.

[0104] With this design, by accurately searching or mining the target flow trend representation vector in the traffic network database, it is possible to fully utilize the existing information in the database or timely supplement new traffic condition related information. The target cross-trajectory relationship network is processed using a spatiotemporal graph neural network to obtain traffic behavior data and mine the target flow trend representation vector, which improves the accuracy of traffic condition prediction. When caching correlation features, different operations are performed according to the total number of flow trend representation vectors, which can not only ensure the effective use of the database and avoid excessive noise data interference, but also timely supplement new correlation features, which is conducive to the effective management of traffic data and accurate analysis and prediction of traffic conditions.

[0105] On the basis of "if the total number of flow trend characterization vectors corresponding to the traffic network database is greater than the first set value, the noise coefficient of each flow trend characterization vector in the traffic network database within the target period is obtained, the flow trend characterization vectors whose noise coefficients meet the target requirements are cleaned, and the correlation characteristics between the target flow trend characterization vector and the target cross-trajectory relationship network are cached in the traffic network database; if the total number of flow trend characterization vectors corresponding to the traffic network database is not greater than the first set value, the correlation characteristics between the target flow trend characterization vector and the target cross-trajectory relationship network are cached in the traffic network database", the method also includes: if the total number of flow trend characterization vectors corresponding to the traffic network database is greater than the first set value, determining the step information for cleaning the flow trend characterization vectors in the traffic network database within a preset period; if the step information meets the optimization requirements, issuing a flow trend update report.

[0106] It can be understood that there is further operation logic based on the comparison relationship between the total number of flow trend characterization vectors corresponding to the traffic network database and the first set value.

[0107] When the total number of flow trend characterization vectors corresponding to the traffic network database is greater than the first set value, in addition to obtaining the noise coefficient of each flow trend characterization vector in the traffic network database within the target period and cleaning the flow trend characterization vectors whose noise coefficient meets the target requirements, and caching the correlation characteristics between the target flow trend characterization vector and the target cross-trajectory relationship network in the traffic network database, it is also necessary to determine the step information for cleaning the flow trend characterization vectors in the traffic network database within a preset period.

[0108] For example, the first set value is 500, if the total number of flow trend characterization vectors in the traffic network database is 600. In the target period (such as the past week), each flow trend characterization vector has a corresponding noise coefficient. If a flow trend characterization vector has abnormal fluctuations in some data due to the instability of the data acquisition device or the interference in the data transmission process, its noise coefficient is high. According to certain rules, the step information of the cleaning process is determined. This step information may be related to factors such as the update frequency of the database, the amount of data, and the importance of the data. For example, the step information is that every 10 flow trend characterization vectors are grouped for cleaning. If this step information meets the optimization requirements, a flow trend update report will be issued. The optimization requirements here may involve multiple considerations such as whether the step information can ensure the cleaning efficiency and whether it can minimize the impact on normal data. For example, if the cleaning method corresponding to the step information can complete the processing of all flow trend characterization vectors that need to be cleaned in a short time (such as 1 hour), and the proportion of normal data mistakenly deleted during the processing does not exceed 5%, then this step information is considered to meet the optimization requirements.

[0109] When the total number of flow trend representation vectors corresponding to the traffic network database is not greater than the first set value, the correlation characteristics between the target flow trend representation vector and the target cross-trajectory relationship network are directly cached in the traffic network database. This is a relatively simple operation method, which avoids unnecessary complex processing when the data volume is small.

[0110] Therefore, when the total number of flow trend characterization vectors is large, determining the cleaning step information and issuing the flow trend update report based on whether it meets the optimization requirements will help improve the accuracy and effectiveness of the traffic network database. Through reasonable step information, noisy data can be efficiently cleaned while ensuring the quality of data processing. Issuing a flow trend update report can promptly notify the relevant system or module database of the update status for subsequent traffic analysis and decision-making. When the total number of flow trend characterization vectors is small, directly caching the correlation features simplifies the operation process, improves the efficiency of data caching, and is conducive to the efficiency and accuracy of overall traffic data management.

[0111] In some optional design ideas, the road network data mining task belongs to a target traffic operation and maintenance project among at least two currently activated traffic operation and maintenance projects; the traffic network database also includes the current congestion trend of each flow trend characterization vector corresponding to the target traffic operation and maintenance project, and the current congestion trend is used to characterize whether it is tracked by any task in the target traffic operation and maintenance project; after obtaining the target flow trend characterization vector corresponding to the target cross-trajectory relationship network from the traffic network database through the target cross-trajectory relationship network of each dynamic traffic monitoring image, the method also includes: annotating the current congestion trend of the target flow trend characterization vector as a tracking label; after obtaining the target weighted weight of each dynamic traffic monitoring image based on the target flow trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels, the method also includes: annotating the current congestion trend of the target flow trend characterization vector as a label to be tracked.

[0112] In the field of transportation, the road network data mining task belongs to the target transportation operation and maintenance project among at least two currently activated transportation operation and maintenance projects. The transportation road network database contains the current congestion trend of each flow trend representation vector corresponding to the target transportation operation and maintenance project, and this current congestion trend is used to indicate whether it is tracked by any task in the target transportation operation and maintenance project.

[0113] After obtaining the target flow trend representation vector corresponding to the target cross-trajectory relationship network from the traffic network database through the target cross-trajectory relationship network of each dynamic traffic monitoring image, the current congestion trend of the target flow trend representation vector will be annotated as a tracking label. For example, in a city's traffic operation and maintenance project, the traffic network database stores flow trend representation vectors of different regions. If the current congestion trend corresponding to the flow trend representation vector of a certain region shows that the region often experiences congestion during the morning rush hour (7-9 o'clock), and this congestion is being tracked by a task to improve the morning rush hour congestion, then the current congestion trend of this flow trend representation vector will be annotated as a tracking label. For example, the traffic flow in the area reaches 800 vehicles per hour during the morning rush hour, and the average speed is 20-30 kilometers per hour. This data indicates that the congestion is relatively serious.

[0114] After obtaining the target weighted weights of each dynamic traffic monitoring image based on the target flow trend representation vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels, the current congestion trend of the target flow trend representation vector will be annotated as a label to be tracked. For example, for an area with a new congestion trend, the traffic flow corresponding to its flow trend representation vector suddenly increases in a certain period of time, such as from 400 vehicles per hour to 600 vehicles per hour, and the average speed decreases from 40 km / h to 30 km / h. There was no task for the congestion situation in this area before, and its current congestion trend is annotated as a label to be tracked. This operation shows that although the congestion in this area has not been tracked before, with the changes in traffic conditions and analysis based on dynamic traffic monitoring images, the congestion trend in this area has begun to attract attention and may become the tracking object of tasks in subsequent traffic operation and maintenance projects.

[0115] It can be seen that by annotating the current congestion trend of the target flow trend representation vector with different labels, the traffic congestion situation can be classified and managed more clearly. Annotating with tracking labels can clarify which congestion situations are being processed, which is convenient for tracking and managing tasks in traffic operation and maintenance projects. Annotating with to-be-tracked labels helps to discover new congestion trends in a timely manner, provides a basis for task planning of subsequent traffic operation and maintenance projects, improves the initiative and foresight of traffic management, and helps to optimize the entire traffic operation and maintenance system, so that traffic resources can be more reasonably allocated to deal with different congestion situations.

[0116] In summary, the embodiment of the present invention determines the influence weight by combining the spatiotemporal state information of the road network area of ​​the target traffic network server and the image description vector of the dynamic traffic monitoring image, and can comprehensively and multi-level evaluate the importance of the dynamic traffic monitoring image. The target flow trend characterization vector based on the target cross-trajectory relationship network is obtained from the traffic network database, and the correlation characteristics of the pre-mined trajectory relationship network and the flow trend characterization vector are utilized to make the analysis of traffic conditions more reliable. The target weighted weight is obtained by combining the two and used for traffic event mining, which improves the accuracy of traffic event mining. This helps to grasp the traffic conditions more accurately, warn of traffic events in advance, optimize traffic management, reduce traffic congestion, improve transportation efficiency and enhance traffic safety.

[0117] An embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the interactive mining processing method based on dynamic images is implemented.

[0118] An embodiment of the present invention provides a processor, which is used to run a program, wherein the interactive mining processing method based on dynamic images is executed when the program is running.

[0119] In the embodiment of the present invention, Figure 2 As shown, the interactive mining processing system 100 includes at least one processor 101, and at least one memory 102 and a bus 103 connected to the processor 101; wherein the processor 101 and the memory 102 communicate with each other through the bus 103; the processor 101 is used to call the program instructions in the memory 102 to execute the above-mentioned interactive mining processing method based on dynamic images.

[0120] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, interactive mining processing systems (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0121] In a typical configuration, the interactive mining processing system includes one or more processors (CPU), memory and bus. The interactive mining processing system may also include input / output interface, network interface and the like.

[0122] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0123] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage computer readable storage media or any other non-transmission media that can be used to store information that can be accessed by the interactive mining processing system. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0124] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or computer-readable storage medium that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements inherent to such process, method, commodity or computer-readable storage medium. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the presence of other identical elements in the process, method, commodity or computer-readable storage medium that includes the elements.

[0125] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0126] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. An interactive mining processing method based on dynamic images, characterized in that: The method is applied to an interactive mining processing system, and the method comprises: In response to the road network data mining task of the target traffic road network server, the influence weight of each dynamic traffic monitoring image at at least two attention levels is determined according to the spatiotemporal state information of the road network area of ​​the target traffic road network server and the image description vector of at least one dynamic traffic monitoring image; wherein the attention level includes a macro level and a micro level, the macro level focuses on the overall flow trend of the entire traffic area and the smoothness of the road, and the micro level focuses on the behavior of vehicles and the local traffic conditions at the intersection; Obtaining a target flow trend representation vector corresponding to the target cross-trajectory relationship network from a traffic network database through the target cross-trajectory relationship network of each dynamic traffic monitoring image; wherein the traffic network database includes correlation features between at least one trajectory relationship network and a flow trend representation vector corresponding to each trajectory relationship network, and the flow trend representation vector corresponding to the trajectory relationship network is obtained by mining the traffic behavior data corresponding to the trajectory relationship network in advance; Obtaining target weightings of each dynamic traffic monitoring image according to the target traffic trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels; According to the target weights of each dynamic traffic monitoring image, traffic event mining is performed on each dynamic traffic monitoring image; The target weighting weight of each dynamic traffic monitoring image is obtained according to the target flow trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels, including: For each dynamic traffic monitoring image, according to the influence weights of each attention level corresponding to the dynamic traffic monitoring image and the eigenvalues ​​corresponding to the influence weights of each attention level, each eigenvalue in the original feature matrix is ​​updated to obtain a target feature matrix; According to the target flow trend characterization vector and the target feature matrix, cross-weighting of the influence weights of each attention level is implemented to obtain the target weighted weight of the dynamic traffic monitoring image.

2. The method according to claim 1, characterized in that The road network data mining task belongs to a target traffic operation and maintenance project among at least two currently activated traffic operation and maintenance projects; The method of obtaining the original feature matrix includes: According to the target traffic operation and maintenance project to which the road network data mining task belongs, the original feature matrix corresponding to the target traffic operation and maintenance project is obtained from a mapping list between each traffic operation and maintenance project and the original feature matrix; Among them, each traffic operation and maintenance project corresponds to its own original feature matrix, and the original feature matrix includes each eigenvalue in the cross-track relationship network corresponding to the target traffic operation and maintenance project and the default parameters of the eigenvalues.

3. The method according to any one of claims 1 to 2, characterized in that: The target cross-trajectory relationship network represents a cross-weighting rule between at least two preset feature values; after determining the influence weight of each dynamic traffic monitoring image at at least two attention levels based on the road network area spatiotemporal state information of the target traffic road network server and the image description vector of at least one dynamic traffic monitoring image, the method further includes: For each dynamic traffic monitoring image, obtaining a target intersection trajectory relationship network corresponding to the dynamic traffic monitoring image; According to the mapping list between the influence weights of each attention level of the dynamic traffic monitoring image and each eigenvalue in the target cross-trajectory relationship network, the target cross-trajectory relationship network corresponding to the dynamic traffic monitoring image, as well as each eigenvalue and the influence weight of each attention level are integrated, and each eigenvalue is associated with an influence weight corresponding to an attention level.

4. The method according to claim 1, characterized in that: The target cross-trajectory relationship network of each dynamic traffic monitoring image is used to obtain a target flow trend representation vector corresponding to the target cross-trajectory relationship network from a traffic network database, including: According to the target cross-trajectory relationship network, a flow trend representation vector corresponding to the target cross-trajectory relationship network is roamed in the traffic network database; If the traffic network database includes a target flow rate trend representation vector corresponding to the target intersection trajectory relationship network, obtaining the target flow rate trend representation vector included in the traffic network database; If the traffic network database does not include a flow trend representation vector corresponding to the target cross-trajectory relationship network, the traffic behavior data corresponding to the target cross-trajectory relationship network is mined as the target flow trend representation vector, and the correlation feature between the target cross-trajectory relationship network and the target flow trend representation vector is cached in the traffic network database.

5. The method according to claim 1 or 4, characterized in that: The method for generating the target traffic trend representation vector includes: Processing the target intersection track relationship network through a spatiotemporal graph neural network to obtain traffic behavior data corresponding to the target intersection track relationship network; Data mining is performed on the traffic behavior data to obtain a behavior element chain corresponding to the traffic behavior data, and a deductive description between each flow node and each characteristic value is analyzed based on the behavior element chain corresponding to the traffic behavior data to generate a target flow trend representation vector corresponding to the target cross-trajectory relationship network.

6. The method according to claim 4, characterized in that The step of caching the correlation features between the target cross-trajectory relationship network and the target traffic trend representation vector into the traffic network database includes: If the total number of flow trend representation vectors corresponding to the traffic network database is greater than a first set value, obtaining the noise coefficient of each flow trend representation vector in the traffic network database within a target period, cleaning the flow trend representation vectors whose noise coefficients meet the target requirements, and caching the correlation characteristics between the target flow trend representation vector and the target cross-trajectory relationship network in the traffic network database; If the total number of flow trend representation vectors corresponding to the traffic network database is not greater than a first set value, caching the correlation feature between the target flow trend representation vector and the target intersection trajectory relationship network in the traffic network database; The method further comprises: If the total number of flow trend characterization vectors corresponding to the traffic network database is greater than a first set value, determining step information for cleaning the flow trend characterization vectors in the traffic network database within a preset period; If the step size information meets the optimization requirements, a flow trend update report is issued.

7. The method according to claim 1, characterized in that The road network data mining task belongs to a target traffic operation and maintenance project among at least two currently activated traffic operation and maintenance projects; the traffic network database also includes the current congestion trend of each flow trend characterization vector corresponding to the target traffic operation and maintenance project, and the current congestion trend is used to indicate whether it is tracked by any task in the target traffic operation and maintenance project; After obtaining the target flow trend representation vector corresponding to the target cross-trajectory relationship network from the traffic road network database through the target cross-trajectory relationship network of each dynamic traffic monitoring image, the method further includes: annotating the current congestion trend of the target flow trend representation vector as a tracking label; After obtaining the target weighted weights of each dynamic traffic monitoring image based on the target flow trend characterization vector and the influence weights of each dynamic traffic monitoring image at multiple attention levels, the method also includes: annotating the current congestion trend of the target flow trend characterization vector as a label to be tracked.

8. An interactive mining processing system, characterized in that: It comprises a processor, a memory and a bus connected to the processor; wherein the processor and the memory communicate with each other via the bus; the processor is used to call program instructions in the memory to execute the interactive mining processing method based on dynamic images as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the interactive mining processing method based on dynamic images described in any one of claims 1 to 7 is implemented.

Images