Dynamic traffic identification system based on artificial intelligence

Abstract

The invention belongs to the technical field of intelligent traffic, and provides a traffic dynamic identification system, in order to solve the problem of low traffic dynamic identification accuracy in the traditional technology, the system adopts a first determination module to respond to a traffic dynamic identification instruction, determine a target road section and determine the road condition type of the target road section; a second determination module is used for determining a road condition information acquisition mode according to the road condition type; an information acquisition module is used for acquiring road condition information data based on a road condition information acquisition mode; a target identification module is used for identifying target objects of corresponding types according to the road condition information data and based on a plurality of preset single-class target identification models; the traffic dynamic presentation module is used for presenting all the target objects on the virtualized roads corresponding to the target road sections, the traffic dynamic states corresponding to the target road sections are obtained, and the accuracy of traffic dynamic state recognition is improved from the two aspects of road condition information data collection and target object recognition.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to an artificial intelligence-based traffic dynamics recognition system. Background Art

[0002] The Intelligent Transportation System (ITS) is an advanced traffic management system that comprehensively uses modern information technologies (such as artificial intelligence, the Internet of Things, big data, and 5G) to optimize traffic management, improve travel efficiency, enhance road safety, and reduce environmental pollution. Traffic dynamic identification is a core component of the Intelligent Transportation System (ITS), aiming to monitor and analyze dynamic information such as traffic flow, vehicle behavior, and pedestrian activities in real time to enable safer and more efficient traffic management.

[0003] In traditional technologies, traffic dynamics are generally identified by collecting relevant information through information collection devices such as cameras, sensors or radars, and analyzing the relevant information to identify traffic dynamics and assist in traffic management.

[0004] However, the inventors realized that in traditional technologies, when using various modal information including but not limited to the above-described ones to perform traffic dynamics recognition, since the analysis is generally performed through several pre-set types of information, it is still impossible to accurately perform traffic dynamics recognition relative to complex road conditions, thereby reducing the effectiveness of traffic dynamics recognition for traffic management.

[0005] Therefore, how to improve the accuracy of traffic dynamic identification has become an urgent problem to be solved in the field of intelligent transportation. Summary of the Invention

[0006] The technical problem solved by the present invention is to solve the problem of low accuracy in traffic dynamic recognition in traditional technologies.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: an artificial intelligence-based traffic dynamics recognition system, comprising: a first determination module, for responding to traffic dynamics recognition instructions, determining a target road section, and determining the road condition type corresponding to the target road section; a second determination module, for determining a road condition information collection method based on the road condition type; an information collection module, for acquiring road condition information data based on the road condition information collection method; a target recognition module, for identifying target objects of corresponding types based on the road condition information data and based on several preset single-class target recognition models; a traffic dynamics presentation module, for presenting all the target objects on a virtualized road corresponding to the target road section, to obtain the traffic dynamics corresponding to the target road section.

[0008] As a preferred solution of the artificial intelligence-based traffic dynamics recognition system described in the present invention, the traffic dynamics recognition system also includes a third determination module, which includes: a fourteenth determination submodule, used to determine the historical traffic information data corresponding to the target road section; a first identification submodule, used to identify the historical traffic information data based on a preset object recognition model to obtain a number of historical target objects; a clustering submodule, used to cluster all the historical target objects based on a preset object clustering model to obtain a number of cluster clusters; a fifteenth determination submodule, used to treat each of the cluster clusters as a corresponding single type of target object to obtain a number of single-class target objects.

[0009] Beneficial effects of the present invention: The artificial intelligence-based traffic dynamics recognition system provided by the present invention, when performing traffic dynamics recognition, not only determines the road condition information collection method according to the road condition type of the target road section, but also adopts the corresponding preset single-class target recognition model for recognition for each type of recognition target object. Since the road condition information collection method is determined according to the road condition type, the effectiveness of road condition information data collection under the road condition type can be improved, thereby achieving accurate recognition of the target object, and adopting the corresponding single-class target recognition model for recognition of each type of target object. Since each single-class target recognition model only needs to focus on identifying target objects of the same type, it can not only reduce the complexity of the single-class target recognition model, but also improve the recognition accuracy of the single-class target recognition model, thereby improving the accuracy of traffic dynamics recognition from the two perspectives of road condition information data collection and target object recognition, achieving accurate recognition of traffic dynamics, and then assisting intelligent transportation, which can improve the safety, efficiency and effectiveness of intelligent transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic block diagram of an artificial intelligence-based traffic dynamics recognition system provided by an embodiment of the present invention; Figure 2 A schematic diagram of the overall concept of an artificial intelligence-based traffic dynamics recognition system provided by an embodiment of the present invention; Figure 3 Another schematic block diagram of an artificial intelligence-based traffic dynamics recognition system provided by an embodiment of the present invention; Figure 4 This is a first sub-schematic block diagram of the artificial intelligence-based traffic dynamics recognition system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0012] An embodiment of the present invention provides an artificial intelligence-based traffic dynamics recognition system, which can be applied to devices including but not limited to desktop computers, servers, and cloud platforms, and can be used when performing traffic dynamics recognition in fields including but not limited to intelligent transportation systems (such as intelligent traffic management, vehicle-road collaboration, or autonomous driving).

[0013] In response to the technical problem of low accuracy in traffic dynamics recognition in traditional technologies, the inventors have proposed an artificial intelligence-based traffic dynamics recognition system according to an embodiment of the present invention. The core concept of the embodiment of the present invention is as follows: when performing traffic dynamics recognition, the road condition information collection method is determined according to the road condition type of the target road section to improve the effectiveness of road condition information collection. For each type of identification target, a corresponding target recognition model is used for recognition to improve the focus, specificity, and accuracy of the target recognition model, and to make the target recognition model lightweight. All target objects are then visualized on a virtual road corresponding to the target road section to obtain the traffic dynamics corresponding to the target road section. Since the effectiveness of information data collection in the road environment can be improved, and each target recognition model only needs to focus on identifying target objects of the same type, the recognition accuracy of the target recognition model can be improved, thereby improving the accuracy of traffic dynamics recognition from both the perspectives of information data collection and target recognition. The improved accuracy of traffic dynamics recognition can further assist intelligent transportation, including but not limited to traffic management, and can improve the safety, efficiency, and effectiveness of intelligent transportation.

[0014] Example 1, please refer to Figure 1 and Figure 2 , Figure 1 A schematic block diagram of an artificial intelligence-based traffic dynamics recognition system provided by an embodiment of the present invention is provided. Figure 2 This is a schematic diagram of the overall concept of the traffic dynamic recognition system based on artificial intelligence provided by the embodiment of the present invention. Figure 1 As shown, in this embodiment, the traffic dynamics recognition system 100 includes a first determination module 101, a second determination module 102, an information collection module 103, a target recognition module 104 and a traffic dynamics presentation module 105. The above functional modules are described in detail as follows: The first determining module 101 is configured to respond to a traffic dynamics recognition instruction, determine a target road section, and determine a road condition type corresponding to the target road section.

[0015] Explanatoryally, starting traffic dynamics recognition will generate a corresponding traffic dynamics recognition instruction. In response to the traffic dynamics recognition instruction, a target road section is determined. The target road section indicates the road section for which traffic dynamics recognition is to be performed. Then, based on the target road section, the road condition type corresponding to the target road section is determined. The road condition type indicates the road condition type corresponding to the road condition characteristics of the target road section formed by various characteristic factors including but not limited to environmental characteristics, road characteristics, pedestrian flow, and vehicle flow. The road condition type can be determined based on relevant factors including but not limited to the road type corresponding to the road attributes, the corresponding weather conditions, and the current time. The road condition types include but are not limited to "city sunny type", "city night type", and "daytime city rainy type". Among them, "city sunny type" indicates that the target road section is a city road and the road condition is during the day, which can be simply referred to as "city sunny", "city night type" indicates that the target road section is a city road and the road condition is at night, which can be simply referred to as "city night", and "daytime city rainy type" indicates that the target road section is a city road and the road condition is at night, which can be simply referred to as "daytime city rainy". Other similar descriptions in the embodiments of the present invention are similar.

[0016] Road attributes represent road properties corresponding to, but not limited to, their function and level, administrative authority, form, and structure. For example, based on function and level, roads can be categorized into, but not limited to, highways, urban roads, and dedicated roads. Highways can be further categorized into, but not limited to, expressways, first-class highways, second-class highways, third-class highways, and fourth-class highways. Urban roads can be further categorized into, but not limited to, expressways, main roads, secondary roads, and branch roads. Dedicated roads can be further categorized into, but not limited to, bus lanes, bicycle lanes, and pedestrian streets. Based on administrative authority, roads can be categorized into, but not limited to, national highways, provincial highways, county roads, and township roads. Based on form and structure, roads can be categorized into, but not limited to, flat roads and three-dimensional roads. Three-dimensional roads can be further categorized into, but not limited to, elevated roads, tunnels, and overpasses. Different road properties and road types affect traffic dynamics such as vehicle and pedestrian flows, and thus the identification of these dynamics. Consequently, road type influences the collection of road condition data. Different road condition data collection devices are suitable for collecting information on vehicle and pedestrian flows and their dynamics corresponding to different road types.

[0017] Weather conditions include but are not limited to sunny days, rainy days, rainy and foggy days, frost, ice and snow days, and strong winds. Weather conditions can be obtained from but are not limited to meteorological service agencies. As meteorological services become more and more sophisticated and accurate, weather conditions can be accurate to the weather conditions within a certain range of the target road section and its surroundings. Weather conditions will affect the collection of road condition information data. Different road condition information data collection equipment is adapted to different weather conditions.

[0018] The current time indicates the time relevant to traffic dynamic identification. The current time includes but is not limited to daytime and nighttime. Daytime can be further divided into early morning, dusk and other daylight periods. The current time is mainly used to determine the brightness of the target road section. The brightness affects the collection of road condition information data. Different road condition information data collection equipment adapts to different lighting conditions.

[0019] In summary, according to the above concept and description, when performing traffic dynamics recognition, the target road section is first determined, and then the road condition type corresponding to the target road section is determined.

[0020] The second determining module 102 is configured to determine a road condition information collection method according to the road condition type.

[0021] Explanatory note: Different road condition types require corresponding road condition information collection methods. These methods indicate the types or equipment used to collect road condition data. This means that different road condition types generally require different road condition information collection methods. This means that there's a correlation and compatibility between road condition types and road condition information collection methods. For example, for road condition type A, method A1 is used; for road condition type B, method B1 is used. A and B, as well as A1 and B1, are generally different. Therefore, the corresponding road condition information collection method is determined based on the road condition type corresponding to the target road segment.

[0022] For example, when the road condition type of the target road is "urban sunny day", the road condition information collection method is determined to be "high-definition camera + millimeter-wave radar + geomagnetic coil", which means that a combination of different devices corresponding to high-definition cameras, millimeter-wave radars, and geomagnetic coils is used to collect road condition information data. The combination of the above different devices is suitable for collecting corresponding information data under the road condition of "urban sunny day", which can enable the collected road condition information data to effectively identify traffic dynamics; when the road condition type of the target road is "urban night", the road condition information collection method is determined to be "infrared thermal imaging camera + low-light starlight camera + millimeter-wave radar", which means that an infrared thermal imaging camera, a low-light starlight camera, and a millimeter-wave radar are used. Road condition data is collected using a combination of different devices. This combination is suitable for collecting data in "urban nighttime" conditions and enables the collected data to effectively identify traffic dynamics. For the target road's road condition type, "millimeter-wave radar + infrared camera + LiDAR system" is selected as the road condition collection method. This indicates that a combination of millimeter-wave radar, infrared camera, and LiDAR system is used to collect road condition data. This combination is suitable for collecting data in "urban daytime rainy" conditions and enables the collected data to effectively identify traffic dynamics. LiDAR (Light Detection and Ranging) is a system that combines laser, Global Positioning System (GPS), and Inertial Navigation System (INS) technologies. It should be noted that the road condition information collection methods for different road condition types in the above examples may utilize more, fewer, or other types of interchangeable devices.

[0023] Therefore, for the same target road section, different corresponding road condition information collection methods are adopted under different road condition types, so that the road condition information collection method changes accordingly with the change of road condition type. The road condition information collection method can be adapted to the road condition characteristics corresponding to the road condition type, so that the collection of road condition information data under the road condition type is adapted to the road conditions, thereby improving the effectiveness of the road condition information data, and then improving the accuracy of traffic dynamics recognition, and realizing accurate recognition of traffic dynamics.

[0024] It should be noted that although the above example involves a seemingly large number of device types, firstly, with the development of technology and products, the prices of various different products are likely to fall. Secondly, with the development of integration technology, the above different devices can be integrated into the same product to achieve corresponding functions. Therefore, the above example has practical implementation. For example, in traditional technology, if five types of road condition information collection devices are deployed, these five types of road condition information collection devices generally work all the time regardless of the road condition type to collect corresponding road condition information data. However, in the embodiment of the present invention, if five types of road condition information collection devices are also deployed, only three types of road condition information collection devices may be required for one road condition type, and only two or four types of road condition information collection devices may be required for another road condition type. In a horizontal comparison, not only are fewer road condition information collection devices enabled, reducing the loss of corresponding road condition information collection devices and the loss of collection resources, but the amount of corresponding road condition information data is also smaller. Moreover, due to the less invalid data, the noise and interference of the valid road condition information data are reduced, which can more effectively identify traffic dynamics and reduce the use and loss of road condition information data processing resources.

[0025] The information collection module 103 is used to obtain road condition information data based on the road condition information collection method.

[0026] Explanatoryally, after determining the road condition information collection method corresponding to the target road section, based on the road condition information collection method, determine and enable the corresponding road condition information collection equipment to collect road condition information data and obtain road condition information data. Therefore, for the same target road section, under different road condition types, based on the appropriate road condition information collection method and the corresponding road condition information collection equipment, road condition information data is collected to obtain road condition information data, which can improve the effectiveness of the road condition information data under the road condition type, and thus realize accurate identification of the target object and accurate identification of traffic dynamics.

[0027] The target recognition module 104 is configured to recognize target objects of corresponding types based on the traffic information data and a plurality of preset single-class target recognition models.

[0028] Explanatoryally, several single-class target recognition models are pre-set, that is, several preset single-class target recognition models. The preset single-class target recognition model represents a recognition model used to identify a certain type / category of target objects. The types of target objects include depending on the content that needs to be identified in traffic dynamics. The content that needs to be identified in traffic dynamics includes but is not limited to the following: ① dynamic behavior of traffic participants; ② macro characteristics of traffic flow; ③ traffic events and abnormal conditions; ④ traffic signals and control conditions; ⑤ road infrastructure conditions; ⑥ external environmental impact.

[0029] The dynamic behaviors of traffic participants include but are not limited to vehicle motion status, speed, acceleration, driving direction, lane keeping / lane changing behavior, priority passage requirements of special vehicles (such as ambulances and police cars), pedestrian crossing trajectories, bicycle / electric vehicle driving paths and violations (driving against traffic, running red lights).

[0030] The macro characteristics of traffic flow include but are not limited to flow rate, density, and speed. Density reflects the degree of congestion, and speed is used to evaluate road traffic efficiency.

[0031] Traffic incidents and abnormal conditions include but are not limited to sudden accidents, collisions, breakdowns, vehicle failures, running red lights, illegal parking, occupying emergency lanes, speeding, road obstacles, construction areas, and the impact of severe weather (rain, snow, fog) on traffic.

[0032] Traffic signal and control status includes but is not limited to real-time traffic light timing, adjustments to adaptive signal control systems, bus priority signals, and emergency vehicle signal priority (such as extended green lights for fire trucks).

[0033] The status of road infrastructure includes but is not limited to road surface conditions (such as collapsed, damaged, slippery, icy, and snowy), traffic signs, and the visibility and integrity of markings.

[0034] External environmental influences include but are not limited to the impact of weather (rain, snow, fog, strong light) on visibility and vehicle control, or sudden changes in traffic flow caused by holidays and large-scale events.

[0035] Therefore, the target object representation includes but is not limited to the traffic dynamics corresponding to the roads, people, vehicles and the relationships between them described above that need to be identified. Therefore, the preset single-class target recognition model includes but is not limited to the recognition model for identifying the dynamic behavior of traffic participants, the macro characteristics of traffic flow, traffic events and abnormal conditions, traffic signals and control conditions, road infrastructure conditions, external environmental influences or the more specific dimensions of the above descriptions as single-class targets. The more specific dimensions of the above descriptions as single-class targets include but are not limited to the recognition model for pedestrians as single-class targets and the recognition model for identifying vehicles as single-class targets. Among them, the recognition model for identifying vehicles as single-class targets can be further divided into single-class target recognition models corresponding to vehicles of the respective types, including but not limited to bicycles, electric bicycles, motorcycles, cars, buses, and trucks. The preset single-class target recognition model can also be but is not limited to a recognition model for identifying target objects by identifying single-modal road condition information data corresponding to image types or audio types. Each preset single-class target recognition model can be modeled according to the target object to be identified using but is not limited to the corresponding machine learning model or deep learning model, which will not be repeated here.

[0036] According to the above concept and setting, the target recognition module 104 is used to identify the corresponding type of target objects based on the road condition information data and a number of preset single-class target recognition models, that is, each preset single-class target recognition model only identifies the target object of that type from the road condition information data, and for all the preset single-class target recognition models, the corresponding type of target objects are identified respectively, thereby obtaining the corresponding target objects. Among them, for the preset single-class target recognition model, since each single-class target recognition model only needs to focus on identifying the target objects of the same type, the complexity of the single-class target recognition model can be reduced, that is, the lightweight of the single-class target recognition model can be achieved. Therefore, in one example, the advantages of edge computing can be fully utilized to put different single-class target recognition models and their related operations into edge devices for operation, and the single-class target recognition models and their related operations can be distributedly deployed. It can not only fully utilize the computing power of edge devices to improve computing efficiency, but also reduce the large amount of computing power required for centralized operation of large models and reduce the computing resources required for centralization.

[0037] Exemplarily, based on road condition information data, a preset pedestrian recognition model that identifies pedestrians as a single type of target will correspondingly identify pedestrian target objects corresponding to pedestrians; similarly, based on road condition information data, a preset electric bicycle recognition model that identifies electric bicycles as a single type of target will correspondingly identify electric bicycle target objects corresponding to electric bicycles, a preset car recognition model that identifies cars as a single type of target will correspondingly identify car target objects corresponding to cars, a preset bus recognition model that identifies buses as a single type of target will correspondingly identify bus target objects corresponding to buses, and a preset truck recognition model that identifies trucks as a single type of target will correspondingly identify truck target objects corresponding to trucks. Thus, based on several preset single type of target recognition models, target objects of corresponding types are identified respectively, and pedestrian target objects, electric bicycle target objects, car target objects, bus target objects, and truck target objects are used as traffic subjects in the traffic dynamics of the identified target road section, thereby obtaining several corresponding target objects. Since each single type of target recognition model only needs to focus on identifying target objects of the same type, it can not only reduce the complexity of the single type of target recognition model, but also improve the recognition accuracy of the single type of target recognition model.

[0038] The traffic dynamics presentation module 105 is configured to present all the target objects on the virtualized road corresponding to the target road section, and obtain the traffic dynamics corresponding to the target road section.

[0039] Explanatory note: After identifying all target objects corresponding to a target road segment, a traffic dynamics presentation module is used to present all target objects on the virtual road corresponding to the target road segment. This is typically done using virtualized identifiers for the target objects, such as pedestrians, cars, buses, and trucks. Because the traffic conditions on the target road segment are dynamically changing in real time, the target objects presented on the virtual road corresponding to the target road segment also dynamically change in real time, thereby obtaining the traffic dynamics corresponding to the target road segment. For example, based on a preset visualization method, including but not limited to digital twin technology and simulation technology, through bidirectional data exchange between physical entities and virtual models, the target objects are presented on the virtual road corresponding to the target road segment, obtaining the traffic dynamics corresponding to the target road segment. This virtualizes the actual traffic conditions on the target road segment into traffic dynamics, thereby achieving traffic dynamics identification for the target road segment.

[0040] In the embodiment of the present invention, a first determination module is used to respond to a traffic dynamics identification instruction, determine a target road section, and determine the road condition type corresponding to the target road section; a second determination module is then used to determine a road condition information collection method based on the road condition type; an information collection module is then used to obtain road condition information data based on the road condition information collection method; a target recognition module is then used to identify target objects of corresponding types based on the road condition information data and a number of preset single-class target recognition models; and a traffic dynamics presentation module is finally used to present all target objects on a virtual road corresponding to the target road section, thereby obtaining the traffic dynamics corresponding to the target road section. Determining the road condition information collection method based on the road condition type can improve the effectiveness of the road condition information data under the road condition type, thereby achieving accurate identification of the target object, and using the corresponding single-class target recognition model to identify each type of target object. Since each single-class target recognition model only needs to focus on identifying target objects of the same type, it can not only reduce the complexity of the single-class target recognition model, but also improve the recognition accuracy of the single-class target recognition model, thereby improving the accuracy of traffic dynamic recognition from the two perspectives of road condition information data collection and target object recognition, achieving accurate identification of traffic dynamics, and then assisting intelligent transportation, which can improve the safety, efficiency and effectiveness of intelligent transportation.

[0041] In one embodiment, the first determining module 101 includes: a first acquiring submodule, a first determining submodule; The first acquisition submodule is configured to acquire at least one of the following: Used to obtain the road type corresponding to the target road section; Used to obtain the weather conditions corresponding to the target road section; Used to obtain the current time; The first determining submodule is configured to determine the road condition type corresponding to the target road section according to the road type, the weather condition and / or the current time.

[0042] Explanatoryally, the first determination module 101 includes: a first acquisition submodule and a first determination submodule; The first acquisition submodule is configured to acquire at least one of the following: Used to obtain the road type corresponding to the target road segment. The road type is used to classify roads. The road type can be classified using, but not limited to, the aforementioned road attributes. The road can also be classified using, but not limited to, road condition characteristics corresponding to pedestrian and vehicle flows. For example, roads can be divided into, but not limited to, sections and intersections, or into congested sections and unobstructed sections. It should be noted that as long as the roads can be classified according to certain common characteristics or rules, the classification of road types is not limited in this embodiment of the present invention. Used to obtain the weather conditions corresponding to the target road section. The weather conditions are as described above and will not be repeated here; Used to obtain the current time. The current time is as described above and will not be repeated here; The first determination submodule is used to determine the road condition type corresponding to the target section based on the road type, the weather condition and / or the current time. For example, the road condition type corresponding to the target section is determined based on the road type and weather condition; the road condition type corresponding to the target section is determined based on the road type and current time; the road condition type corresponding to the target section is determined based on the weather condition and current time; the road condition type corresponding to the target section is determined based on the road type, weather condition and current time.

[0043] Further, the first determining submodule includes at least one of the following: a second determining submodule, a third determining submodule, a fourth determining submodule, a fifth determining submodule, a sixth determining submodule, and a seventh determining submodule; The second determining submodule is configured to determine that the road condition type corresponding to the target road section is "urban sunny" when the road type is "urban road" and the weather condition is "sunny"; A third determining submodule is configured to determine that the road condition type corresponding to the target road section is "city night" when the road type is "city road" and the current time is "night"; A fourth determining submodule is configured to determine that the road condition type corresponding to the target road section is "daytime city rainy day" when the road type is "city road", the weather condition is "rainy day", and the current time is "daytime"; A fifth determining submodule is configured to determine that the road condition type corresponding to the target road section is "highway and sunny" when the road type is "highway" and the weather condition is "sunny"; A sixth determining submodule is configured to determine, when the road type is "highway" and the current time is "nighttime", that the road condition type corresponding to the target road section is "highway at night"; The seventh determination submodule is used to determine that the road condition type corresponding to the target section is "daytime high-speed rainy day" when the road type is "highway", the weather condition is "rainy day" and the current time is "daytime".

[0044] Specifically, the second determining submodule is configured to determine that the road condition type corresponding to the target road section is "urban road" and the weather condition is "sunny", indicating that the road condition of the target road section is urban road and the weather condition is sunny; The third determination submodule is configured to determine that the road condition type corresponding to the target road section is "city night" when the road type is "city road" and the current time is "night", indicating that the road condition of the target road section is a city road and the current time is night; The fourth determination submodule is used to determine that the road condition type corresponding to the target road section is "daytime city rainy day" when the road type is "city road", the weather condition is "rainy day", and the current time is "daytime", indicating that the road condition of the target road section is city road, the weather condition is rainy day, and the current time is daytime; a fifth determining submodule, configured to determine, when the road type is "highway" and the weather condition is "sunny", that the road condition type corresponding to the target road section is "highway sunny", indicating that the road condition of the target road section is a highway and the weather condition is sunny; a sixth determining submodule, configured to, when the road type is "highway" and the current time is "night", determine that the road condition type corresponding to the target road section is "highway nighttime", indicating that the road condition of the target road section is a highway and the current time is nighttime; The seventh determination submodule is used to determine that the road condition type corresponding to the target section is "daytime highway rainy day" when the road type is "highway", the weather condition is "rainy day", and the current time is "daytime", indicating that the road condition of the target section is highway, the weather condition is rainy day, and the current time is daytime.

[0045] Therefore, according to the road type, weather condition and current time, different combinations of the road type, weather condition and current time are made to name the road condition type corresponding to the target section to indicate the road condition of the target section at that time.

[0046] It should be noted that: First, in the implementation of the technical solution of the present invention, the road conditions described above are included but not limited to the types described above. For example, according to different actual situations, the road conditions may also include but not limited to the following types: When the road type is "city road", the weather condition is "rain and fog", and the current time is "daytime", determining that the road condition type corresponding to the target road section is "daytime city rain and fog" indicates that the road condition of the target road section is city road, the weather condition is rain and fog, and the current time is daytime; When the road type is "city road" and the weather condition is "wind and snow", the road condition type corresponding to the target road section is determined to be "city wind and snow", indicating that the road condition of the target road section is city road and the weather condition is wind and snow; When the road type is "expressway", the weather condition is "rain and fog", and the current time is "daytime", the road condition type corresponding to the target section is determined to be "daytime expressway rain and fog", indicating that the road condition of the target section is expressway, the weather condition is rain and fog, and the current time is daytime.

[0047] Since road types, weather conditions, and current time include many different situations, different combinations of specific road types, weather conditions, and current time can be used to combine different road condition types to characterize the road conditions corresponding to the target section at that time. It is impossible to list them all here. The above only uses examples of some road condition types to illustrate the concept of the technical solution of the present invention, and it is not a complete enumeration. The road condition type can be flexibly set according to needs to characterize the road conditions corresponding to the target section. It needs to be determined according to specific needs when implementing the technical solution.

[0048] Second, the above-mentioned road condition types include but are not limited to the names of the road condition types corresponding to "city sunny day", "city night", "daytime city rainy day", and "highway sunny day". They are merely identifiers of the corresponding road condition types, which are used to distinguish different road condition types and are not used to limit different road condition types. Other corresponding identifiers can be used to distinguish different road condition types, and will not change the essential content of different road condition types. Therefore, including but not limited to the above-mentioned road condition type names are only used to understand and distinguish the corresponding characteristics of the road condition types, and are not used to limit the road condition types. In other examples, the road condition types may also be called other names, but the meaning and content of the road condition types represented are not substantially different, and are all within the scope of the technical solution of the present invention.

[0049] The embodiment of the present invention obtains at least one of the following: the road type, weather conditions and / or current time corresponding to the target road section, and determines the road condition type corresponding to the target road section based on the road type, weather conditions and / or current time, so as to reflect the current characteristics of the target road section from different dimensions, and determines the road condition type based on the above-mentioned corresponding characteristics, and then determines the road condition information collection method, which can make the road condition information collection method adapt to the road condition characteristics corresponding to the road condition type, so that the collection of road condition information data under the road condition type adapts to the road conditions, thereby improving the validity of the road condition information data, and then improving the accuracy of traffic dynamic identification, and realizing accurate identification of traffic dynamics.

[0050] In one embodiment, the second determining module 102 includes at least one of the following: an eighth determining submodule, a ninth determining submodule, and a tenth determining submodule; The eighth determining submodule is configured to determine, when the road condition type is "urban sunny day", that the road condition information collection method is "high-definition camera + millimeter-wave radar + geomagnetic coil"; a ninth determining submodule, configured to, when the road condition type is "city night", determine that the road condition information collection method is "infrared thermal imaging camera + low-light starlight-level camera + millimeter-wave radar"; The tenth determining submodule is used to determine that the road condition information collection method is "millimeter wave radar + infrared camera + LiDAR system" when the road condition type is "daytime city rainy day".

[0051] Explanatoryally, the second determining module 102 includes at least one of the following: The eighth determination submodule is configured to, when the road condition type is "urban, sunny," determine the road condition information collection method to be "high-definition camera + millimeter-wave radar + geomagnetic coil." Specifically, when the target road condition is urban and the weather is sunny, road condition information data is collected using road condition information collection equipment corresponding to a high-definition camera, millimeter-wave radar, and geomagnetic coil. Since the target road condition is urban and the weather is sunny, such conditions generally exhibit traffic characteristics corresponding to ample lighting, clear vision, heavy traffic volume, and a mix of pedestrians and non-motorized vehicles. In this case, a high-definition camera is used as the primary road condition information data collection device to identify vehicles, pedestrians, traffic signs, and violations (such as running a red light). Millimeter-wave radar is used to assist with speed measurement and distance monitoring to compensate for camera blind spots. A geomagnetic coil is used to collect traffic flow statistics at fixed points. The data collected by these road condition information data collection devices is capable of effectively identifying traffic dynamics and is therefore compatible with the road condition type corresponding to the target road condition being urban and the weather being sunny, enabling accurate identification of traffic dynamics.

[0052] The ninth determination submodule is used to determine that the road condition information collection method is "infrared thermal imaging camera + low-light starlight camera + millimeter-wave radar" when the road condition type is "city night". That is, when the road condition of the target road is a city road and the current time is night, the road condition information collection equipment corresponding to the infrared thermal imaging camera, low-light starlight camera, and millimeter-wave radar is used to collect road condition information data. Since the road condition of the target road is a city road and the current time is night, this situation generally has insufficient light, low signal-to-noise ratio of traditional cameras, poor visibility of pedestrians or non-motor vehicles, and interference from glare from car lights. Corresponding traffic status characteristics, in this case, an infrared thermal imaging camera is used as the main road condition information data collection device, which detects pedestrians and vehicles through thermal radiation (not affected by visible light), and a low-light starlight-level camera is used to assist in identifying license plates and traffic signs (needs to be combined with a fill light), and a millimeter-wave radar is used to detect vehicle speed and distance to make up for the blind spots of the visual sensor. The data collected by the above-mentioned road condition information data collection equipment can effectively identify traffic dynamics, and is therefore adapted to the road condition type corresponding to the target road being an urban road and the current time being night, and can achieve accurate identification of traffic dynamics.

[0053] The tenth determination submodule is used to determine the road condition information collection method as "millimeter wave radar + infrared camera + LiDAR system" when the road condition type is "daytime city rainy day", that is, when the road condition of the target road is a city road and the current time is daytime and the weather condition is rainy, the road condition information collection equipment corresponding to the millimeter wave radar, infrared camera, and LiDAR system is used to collect road condition information data. Since the road condition of the target road is a city road and the current time is daytime and the weather condition is rainy, this situation generally has low visibility, the camera is easily interfered by raindrops, the road surface is slippery, and it is necessary to monitor dangerous behaviors such as sudden braking. Corresponding traffic status characteristics, in this case, millimeter-wave radar is used as the main road condition information data collection equipment, which is not affected by rain and fog, and can stably detect vehicle speed and distance. The infrared camera can penetrate rain and fog to assist in identifying pedestrian and vehicle outlines, and the LiDAR system is used for high-precision 3D modeling (such as key intersections), but a rain cover is required. The data collected by the above-mentioned road condition information data collection equipment can effectively identify traffic dynamics, and is therefore adapted to the road condition type corresponding to the target road being an urban road, the current time being daytime, and the weather condition being rainy, and can achieve accurate identification of traffic dynamics.

[0054] Therefore, when conducting traffic dynamics identification on urban roads, the road condition type of the target section is first determined, and based on the road condition type, the corresponding road condition information collection method is determined, and the road condition information data collected by different road condition information data collection devices are combined, so as to utilize the primary and secondary and auxiliary of several types of road condition information data under the road condition type, and through the complementarity of different road condition information data, improve the accuracy of traffic dynamics identification, and then assist traffic management, which can improve the safety, efficiency and effectiveness of traffic management.

[0055] It should be noted that with technological advancements, the costs of the various road condition information data collection devices described above are likely to decrease and integration is possible, potentially meeting deployment requirements. Therefore, the technical solutions of the embodiments of the present invention are practical. Furthermore, the examples described above primarily illustrate the technical solution concept and core ideas of the present invention and are not intended to be exhaustive. During implementation, road condition types can be flexibly configured as needed, and more, fewer, or other alternative road condition information data collection devices can be employed.

[0056] The embodiment of the present invention collects road condition information data of corresponding urban roads by adopting corresponding road condition information collection methods under different road condition types of urban roads, that is, reasonably configuring the combination of road condition information collection methods and equipment, and then performs traffic dynamics identification based on the corresponding road condition information data. Since each of the above-mentioned road condition information collection methods is suitable for the road environment characteristics corresponding to the corresponding road condition type of urban roads, the collected road condition information data can accurately and clearly express and reflect the traffic dynamics of the corresponding urban roads. Therefore, the road condition information data under this road condition type has high validity, can realize accurate identification of target objects on urban roads, and can ensure the accuracy, real-time and robustness of traffic dynamics identification in various scenarios of urban roads, thereby improving the accuracy, safety and reliability of traffic dynamics identification on urban roads, to assist the intelligent transportation of urban roads, and can improve the safety, efficiency and effectiveness of intelligent transportation on urban roads.

[0057] In one embodiment, the second determining module 102 includes at least one of the following: an eleventh determining submodule, a twelfth determining submodule, and a thirteenth determining submodule; The eleventh determining submodule is configured to determine, when the road condition type is "high-speed sunny day", that the road condition information collection method is "LiDAR system + high-definition camera + microwave radar"; A twelfth determining submodule is configured to determine, when the road condition type is "high-speed nighttime", that the road condition information collection method is "long-range infrared laser radar + microwave radar array + high-sensitivity panoramic camera"; The thirteenth determination submodule is used to determine that the road condition information collection method is "millimeter wave radar + infrared thermal imaging + meteorological sensor" when the road condition type is "daytime high speed rainy day".

[0058] Explanatoryally, based on the same technical concept and core idea as the above-described traffic dynamics identification for urban roads, when performing traffic dynamics identification for expressways, the second determination module 102 determines the road condition information collection method based on the road condition type, including: The eleventh determination submodule is used to determine the road condition information collection method as "LiDAR system + high-definition camera + microwave radar" when the road condition type is "high-speed sunny day". That is, when the road condition of the target road is a highway and the weather condition is sunny, the road condition information collection equipment corresponding to the LiDAR system, high-definition camera, and microwave radar is used to collect road condition information data. Since the road condition of the target road is a highway and the weather condition is sunny, this situation generally has the traffic state characteristics of high speed, long-distance monitoring, and sudden accidents (such as tire blowouts) that require rapid response. In this situation, Under such circumstances, the LiDAR system is used as the main road condition information data collection equipment, which can realize high-precision detection of the vehicle's 3D position at a long distance (200m+), and use high-definition cameras to assist in license plate recognition and event recording (such as illegal parking), and use microwave radar to monitor the vehicle speed of the entire road section. Moreover, its cost is lower than that of LiDAR and it has more cost advantages. The data collected by the above-mentioned road condition information data collection equipment can effectively identify traffic dynamics, and is therefore adapted to the road condition type corresponding to the target road being a highway and the weather condition being sunny, and can achieve accurate identification of traffic dynamics.

[0059] The twelfth determination submodule is used to determine the road condition information collection method as "long-range infrared laser radar + microwave radar array + high-sensitivity panoramic camera" when the road condition type is "high-speed nighttime". That is, when the target road is a highway and it is nighttime, road condition data is collected using road condition information collection equipment corresponding to long-range infrared lidar, microwave radar array, and high-sensitivity panoramic camera. Since the target road is a highway and it is nighttime, this situation generally exhibits traffic state characteristics corresponding to high speeds, the need for ultra-long-distance monitoring (≥200 meters), limited vision due to obstruction by large trucks, and fatigue driving, which correspond to high-risk behavior detection. In this case, a long-range infrared lidar is used as the primary road condition data collection equipment, capable of performing 3D point cloud detection (anti-glare) within a range of 200-300 meters. A microwave radar array is used for multi-target speed and ranging measurement across the entire road section. Its cost is lower than that of lidar, making it more cost-effective. In addition, a high-sensitivity panoramic camera is used to assist in event recording (such as illegal parking) and license plate recognition. The data collected by these road condition data collection equipment can effectively identify traffic dynamics, and is therefore suitable for the road condition type corresponding to the target road being a highway and the current time being nighttime, and can achieve accurate identification of traffic dynamics.

[0060] The thirteenth determination submodule is used to determine that the road condition information collection method is "millimeter wave radar + infrared thermal imaging + meteorological sensor" when the road condition type is "daytime high speed rainy day", that is, when the road condition of the target road is a highway and the current time is daytime and the weather condition is rainy day, the road condition information collection equipment corresponding to the millimeter wave radar, infrared thermal imaging, and meteorological sensor is used to collect road condition information data. Since the road condition of the target road is a highway and the current time is daytime and the weather condition is rainy day, this situation generally has the traffic state corresponding to high-risk events such as rain and fog affecting optical equipment and needing to be vigilant against hydroplaning, rear-end collisions, etc. Features: In this case, millimeter-wave radar is used as the main road condition information data collection equipment to work around the clock, detect vehicle speed, vehicle distance and abnormal movement (such as sudden lane changes), and use infrared thermal imaging to assist in identifying faulty vehicles (engine overheating) or stranded pedestrians, and meteorological sensors are used to monitor rainfall intensity and road slipperiness in real time to link speed limit prompts. The data collected by the above-mentioned road condition information data collection equipment can effectively identify traffic dynamics, and is therefore adapted to the road condition type corresponding to the target road being a highway, the current time being daytime, and the weather condition being rainy, and can achieve accurate identification of traffic dynamics.

[0061] Therefore, when performing traffic dynamics identification on a highway, the road condition type of the target section is first determined, and based on the road condition type, the corresponding road condition information collection method is determined, and the road condition information data collected by different road condition information data collection devices are combined, thereby utilizing the primary and secondary and auxiliary of several types of road condition information data under the road condition type, and through the complementarity of different road condition information data, the accuracy of traffic dynamics identification is improved, thereby assisting traffic management, which can improve the safety, efficiency and effectiveness of traffic management.

[0062] It should be noted that with technological advancements, the costs of the various road condition information data collection devices described above are likely to decrease and integration is possible, potentially meeting deployment requirements. Therefore, the technical solutions of the embodiments of the present invention are practical. Furthermore, the examples described above primarily illustrate the technical solution concept and core ideas of the present invention and are not intended to be exhaustive. During implementation, road condition types can be flexibly configured as needed, and more, fewer, or other alternative road condition information data collection devices can be employed.

[0063] In an embodiment of the present invention, corresponding road condition information collection methods are adopted under different road condition types of the expressway, that is, a combination of road condition information collection methods and equipment is reasonably configured to collect road condition information data of the corresponding expressway, and then traffic dynamics identification is performed based on the corresponding road condition information data. Since each of the above-mentioned road condition information collection methods is suitable for the road environment characteristics corresponding to the corresponding road condition type of the expressway, the collected road condition information data can accurately and clearly express and reflect the traffic dynamics of the corresponding expressway. Therefore, the road condition information data under this road condition type has high validity, can realize accurate identification of target objects on the expressway, can ensure the accuracy, real-time and robustness of traffic dynamics identification in various scenarios on the expressway, and thus improve the accuracy, safety and reliability of traffic dynamics identification on the expressway, to assist the intelligent transportation of the expressway, and can improve the safety, efficiency and effectiveness of the intelligent transportation of the expressway.

[0064] In one embodiment, see Figure 3 , Figure 3 Another schematic block diagram of the traffic dynamics recognition system based on artificial intelligence provided by an embodiment of the present invention. Figure 3 As shown, in this embodiment, the traffic dynamics recognition system 100 further includes: A third determining module 106 is configured to determine a number of single-type target objects corresponding to the target road segment; The fourth determining module 107 is configured to determine a corresponding preset single-class target recognition model according to each type of the single-class target object, and obtain a plurality of preset single-class target recognition models.

[0065] Explanatory note: the determination of several preset single-class target recognition models can be manually configured based on experience, that is, the relevant personnel configure the corresponding several preset single-class target recognition models based on the target road section. This is limited by the subjective understanding of the relevant personnel. Alternatively, the traffic dynamic recognition system can automatically configure them based on the corresponding traffic dynamic characteristics of the target road section. The specific implementation can refer to the following process: A single-class target recognition model set is pre-set, and the single-class target recognition model set includes several initial single-class target recognition models. The initial single-class target recognition model is set for each category of single-class target objects. The single-class target object refers to the relevant description of the target object when describing the preset single-class target recognition model as mentioned above. The single-class target object represents a target object of a single type / category, that is, the target object belongs to the same type / category, which is obtained by distinguishing the types of the above-mentioned target objects. Then, according to the corresponding traffic dynamic characteristics of the target section, several categories of single-class target objects corresponding to the target section are determined, and then Different target road sections correspond to different categories of single-class target objects. Then, according to each category of single-class target object, the corresponding initial single-class target recognition model is determined from the single-class target recognition model set as the corresponding preset single-class target recognition model. The preset single-class target recognition model is as described above and will not be repeated here. For the several categories of single-class target objects corresponding to the target road section, several preset single-class target recognition models are obtained, that is, several preset single-class target recognition models corresponding to the target road section, that is, the target road section is instantly matched with several preset single-class target recognition models based on the preset single-class target recognition model set.

[0066] In an embodiment of the present invention, a number of single-class target objects corresponding to a target road section are determined, and a corresponding preset single-class target recognition model is determined based on each type of single-class target object to obtain a number of preset single-class target recognition models, so as to realize the classification of target objects into the same or similar types according to dimensional features including but not limited to attributes, shapes, dynamics, etc., and configure a corresponding single-class target recognition model for each type of single-class target object. Since each single-class target recognition model only needs to focus on identifying target objects of the same type, it can not only reduce the complexity of the single-class target recognition model, but also improve the recognition accuracy of the single-class target recognition model. Furthermore, combined with the collection of road condition information data, it can improve the accuracy of traffic dynamic recognition from the two perspectives of road condition information data collection and target object recognition, realize accurate recognition of traffic dynamics, and then assist intelligent transportation, which can improve the safety, efficiency and effectiveness of intelligent transportation.

[0067] In one embodiment, see Figure 4 , Figure 4 This is the first schematic block diagram of the traffic dynamics recognition system based on artificial intelligence provided by the embodiment of the present invention. Figure 4 As shown, in this embodiment, the third determination module 106 includes: a fourteenth determination submodule 401, a first identification submodule 402, a clustering submodule 403, and a fifteenth determination submodule 404; The fourteenth determining submodule 401 is configured to determine the historical traffic information data corresponding to the target road segment; A first identification submodule 402 is configured to identify the historical traffic information data based on a preset object recognition model to obtain a plurality of historical target objects; The clustering submodule 403 is configured to cluster all the historical target objects based on a preset object clustering model to obtain a plurality of clusters; The fifteenth determining submodule 404 is configured to take each of the clusters as a target object of a corresponding single type, and obtain a plurality of single-type target objects.

[0068] Explanatoryally, an object recognition model is pre-set, that is, a preset object recognition model. The preset object recognition model represents a model for identifying historical target objects of the target road section. The historical target objects refer to the relevant content of the target object description when the single-class target recognition model is preset in the above description. For different historical target objects, a comprehensive multi-modal, multi-recognition task model can be set for recognition, or different single-class historical object recognition models can be set according to the above description. There is no limitation here, as long as the historical target objects can be accurately identified.

[0069] The object clustering model is pre-set, i.e., the preset object clustering model. The preset object clustering model represents a model for clustering and identifying a number of target objects related to a target road section. The preset object clustering model includes but is not limited to the K-Means algorithm, Gaussian mixture model (GMM), and BIRCH (Balanced Iterative Reduction and Clustering).

[0070] According to the above conception and arrangement, the fourteenth determination submodule 401 is used to determine the historical traffic information data corresponding to the target road section. The historical traffic information data represents the past traffic information data of the target road section. The historical traffic information data includes but is not limited to the information data involved in the relevant description corresponding to the above-mentioned content that needs to be identified regarding traffic dynamics. The historical traffic information data may also include but is not limited to information data of different modes corresponding to audio, image, and infrared. For example, the sounds of different vehicles are different. Therefore, different vehicles can also be identified from the sound perspective, and so on.

[0071] The first identification submodule 402 is used to identify the historical traffic information data based on a preset object recognition model to obtain a number of historical target objects. The historical target objects include but are not limited to the following contents and their corresponding specific contents: ① historical dynamic behaviors of historical traffic participants; ② historical macro characteristics of historical traffic flows; ③ historical traffic events and historical abnormal conditions; ④ historical traffic signals and historical control conditions; ⑤ historical road infrastructure conditions; ⑥ historical external environmental influences, thereby identifying different historical target objects contained in the historical traffic information data.

[0072] The clustering submodule 403 is used to cluster all historical target objects based on the preset object clustering model to obtain a number of cluster clusters, thereby classifying the different historical target objects contained in the historical traffic information data to obtain a single type of historical target objects. Through clustering, historical target objects with similar characteristics are gathered together, and then the same preset single-class target recognition model is used to recognize a single type of target object, which can improve the focus and accuracy of the preset single-class target recognition model in identifying the target object.

[0073] The fifteenth determination submodule 404 is used to treat each cluster as a corresponding single type of target object to obtain a number of single-class target objects, thereby obtaining a number of single-class target objects corresponding to the target road section, and based on the single-class target objects, determine the corresponding preset single-class target recognition model to obtain a number of preset single-class target recognition models corresponding to the target road section.

[0074] Therefore, the historical traffic information data corresponds to and adapts to the target road section, and the identification of the historical target objects and single-category target objects based on this also corresponds to and adapts to the target road section. The several preset single-category target recognition models corresponding to the target road section also correspond to and adapt to the target road section, and have the characteristics of personalization and customization, thereby improving the accuracy and comprehensiveness of the several preset single-category target recognition models corresponding to the target road section, and thus improving the accuracy of traffic dynamic identification of the target road section.

[0075] Furthermore, the first identification submodule includes at least one of the following: a second identification submodule, a third identification submodule, a fourth identification submodule, and a first cross-modal identification submodule; The second recognition submodule is configured to perform visual modality recognition on the historical traffic information data based on a preset visual modality recognition model to obtain an image historical target object; The third recognition submodule is configured to perform audio modality recognition on the historical traffic information data based on a preset audio modality recognition model to obtain an audio history target object; The fourth identification submodule is configured to perform sensor modality identification on the historical traffic information data based on a preset sensor modality identification model to obtain a sensor modality historical target object; The first cross-modal recognition submodule is configured to perform cross-modal recognition on the historical traffic information data based on a preset multi-modal recognition model to obtain a multi-modal historical target object; In addition, the first identification submodule further includes: a first union submodule, configured to union the image history target object, the audio history target object, the sensor modality history target object, and the multimodal history target object to obtain a plurality of history target objects.

[0076] Specifically, a visual modality recognition model is pre-set, that is, a preset visual modality recognition model. The preset visual modality recognition model represents a model for recognizing historical traffic information data of visual modalities including but not limited to images, videos, visible light, and non-visible light. The preset visual modality recognition model may include but is not limited to an image recognition model. The preset visual modality recognition model includes but is not limited to a convolutional neural network (CNN) (such as a Transformer-Based visual model) and a YOLOv8 model. The above are just examples. During implementation, the corresponding specific model can be selected as needed.

[0077] The pre-set audio modal recognition model is a preset audio modal recognition model. The preset audio modal recognition model represents a model for recognizing historical traffic information data of audio modalities corresponding to, but not limited to, human voices, car sounds, and truck sounds. The preset audio modal recognition model includes, but is not limited to, a Gaussian mixture model (GMM) and a Transformer model. The above are only examples. During implementation, the corresponding specific model can be selected as needed to identify different historical target objects based on sound recognition.

[0078] A sensor modal recognition model is pre-set, i.e., a preset sensor modal recognition model. The preset sensor modal recognition model represents a model for recognizing historical traffic information data corresponding to sensor modalities including but not limited to radar, LiDAR, and infrared point clouds. Sensor data, i.e., physical perception modalities, is raw environmental signals directly collected by physical devices (such as radar, temperature sensors, and GPS) that reflect the physical properties of the target (such as shape, distance, and temperature). Examples include 3D point clouds (distance and reflectivity) corresponding to LiDAR systems, speed, orientation, and distance corresponding to millimeter-wave radars, acceleration and angular velocity corresponding to inertial sensors (IMUs), GPS latitude and longitude corresponding to spatial coordinates, GIS map data, vehicle movement trajectories (time + position series) corresponding to spatiotemporal trajectories, temperature change curves corresponding to time series, and historical traffic flow data. Pre-set audio modal recognition models include but are not limited to the YOLO model, Faster R-CNN model, DETR model, and convolutional neural networks (CNNs). The above are only examples, and specific models can be selected as needed during implementation.

[0079] A multimodal recognition model is pre-set, i.e., a preset multimodal recognition model. The preset multimodal recognition model represents a model for cross-modal recognition of multimodal historical traffic information data corresponding to text, images, audio, and video. More complex tasks are achieved through cross-modal fusion of historical traffic information. The preset multimodal recognition models include but are not limited to the CLIP model (i.e., Contrastive Language–Image Pretraining), the Flamingo model, and the Kosmos-1 model. The above are just examples. During implementation, the corresponding specific model can be selected as needed.

[0080] According to the above-mentioned conception and setting, the second identification submodule is used to perform visual modality recognition on the historical traffic information data based on a preset visual modality recognition model. For example, it identifies the images or videos in the historical traffic information data, identifies the corresponding target objects, and obtains image historical target objects. The image historical target objects represent the historical target objects identified based on the visual modality. Since the preset visual modality recognition model only recognizes the historical traffic information data of the corresponding visual modality, it will automatically ignore or filter the historical traffic information data of other modalities, that is, it will not recognize the historical traffic information data of other modalities.

[0081] The third identification submodule is used to perform audio modality recognition on the historical traffic information data based on a preset audio modality recognition model. For example, it recognizes the human voice or different sounds of different vehicles in the historical traffic information data, identifies the corresponding target object, and obtains the audio historical target object. The audio historical target object represents the historical target object identified based on the audio modality. Since the preset audio modality recognition model only recognizes the historical traffic information data of the corresponding audio modality, it will automatically ignore or filter the historical traffic information data of other modalities, that is, it will not recognize the historical traffic information data of other modalities.

[0082] The fourth identification submodule is used to perform sensor modality recognition on the historical traffic information data based on a preset sensor modality recognition model. For example, the historical traffic information data including but not limited to radar, temperature sensor, humidity sensor, geomagnetic coil, GPS, and infrared point cloud are identified to identify the corresponding target object and obtain the sensor modality historical target object. The sensor modality historical target object represents the historical target object identified based on the sensor modality. Since the preset sensor modality recognition model only recognizes the historical traffic information data of the corresponding sensor modality, it will automatically ignore or filter the historical traffic information data of other modalities, that is, it will not recognize the historical traffic information data of other modalities.

[0083] The first cross-modal recognition submodule is used to perform cross-modal recognition on historical traffic information data based on a preset multimodal recognition model. That is, the multiple modal types of data (such as text, images, audio, video, etc.) contained in the historical traffic information data are processed and understood simultaneously, and more complex tasks are achieved through cross-modal information fusion to obtain multimodal historical target objects. The multimodal historical target objects represent historical target objects identified based on information data of multiple modal types.

[0084] Then, the first identification submodule also includes: a first union submodule, which is used to take the union of image historical target objects, audio historical target objects, sensor modality historical target objects, and multimodal historical target objects, and deduplicate them to obtain several historical target objects, that is, several historical target objects corresponding to the target section, that is, the target objects involved in the target section in the past traffic conditions, to realize the identification of the historical target objects of the target section from different dimensions, and to be able to identify the historical target objects corresponding to the target section as comprehensively as possible, and then cluster all historical target objects to obtain several single-category target objects corresponding to the target section, so as to further improve the accuracy of traffic dynamic identification.

[0085] Furthermore, the historical traffic information data includes at least one of the following: historical traffic visual information data, historical traffic audio information data, and historical traffic sensor information data; the first recognition submodule includes at least one of the following: a fifth recognition submodule, a sixth recognition submodule, a seventh recognition submodule, and a second cross-modal recognition submodule; The fifth recognition submodule is configured to recognize the historical traffic visual information data based on a preset visual modality recognition model to obtain an image historical target object; The sixth identification submodule is configured to identify the historical traffic audio information data based on a preset audio modality recognition model to obtain an audio history target object; The seventh identification submodule is configured to identify the historical traffic sensor information data based on a preset sensor modality identification model to obtain a sensor modality historical target object; The second cross-modal recognition submodule is configured to perform cross-modal recognition on the historical traffic information data based on a preset multimodal recognition model to obtain a multimodal historical target object; In addition, the first identification submodule further includes: a second union submodule, which is used to take the union of the image historical target object, the audio historical target object, the sensor modality historical target object, and the multimodal historical target object to obtain a plurality of historical target objects.

[0086] Specifically, the preset visual modal recognition model, the preset audio modal recognition model, the preset sensor modal recognition model, and the preset multimodal recognition model refer to the description of the above-mentioned corresponding embodiments, which are consistent with the above description. It should be noted that the only difference between the embodiment of the present invention and the above-mentioned related embodiments is that the corresponding information data processed by the preset visual modal recognition model, the preset audio modal recognition model, and the preset sensor modal recognition model are different, and everything else is the same. For relevant descriptions, please refer to the relevant descriptions of the above-mentioned embodiments, which will not be repeated here.

[0087] According to the above conception and arrangement, the fifth recognition submodule is used to recognize the corresponding historical traffic visual information data based on a preset visual modality recognition model to obtain an image historical target object.

[0088] The sixth identification submodule is used to identify the corresponding historical traffic audio information data based on a preset audio modality recognition model to obtain an audio history target object.

[0089] The seventh identification submodule is used to identify the corresponding historical traffic sensor information data based on a preset sensor modality identification model to obtain a sensor modality historical target object.

[0090] A second cross-modal recognition submodule is used to perform cross-modal recognition on the historical traffic information data based on a preset multi-modal recognition model to obtain a multi-modal historical target object; In addition, the first recognition submodule also includes: a second union submodule, which is used to take the union of image historical target objects, audio historical target objects, sensor modality historical target objects, and multimodal historical target objects, and perform deduplication to obtain a number of historical target objects, namely, a number of historical target objects corresponding to the target road section, namely, the target objects involved in the target road section in the past traffic status, to realize the recognition of the historical target objects of the target road section from different dimensions, to be able to identify the historical target objects corresponding to the target road section as comprehensively as possible, and then cluster all historical target objects to obtain a number of single-category target objects corresponding to the target road section, so as to further improve the accuracy of traffic dynamic recognition, since the historical traffic information data is divided into historical traffic visual Information data, historical traffic audio information data, historical traffic sensor information data, and the above-mentioned preset visual modal recognition model, preset audio modal recognition model, and preset sensor modal recognition model are used to process the above-mentioned corresponding categories of information data respectively, so as to be different from the above-mentioned corresponding recognition models in the previous embodiment that process the comprehensive and complex same historical traffic information data, reduce the amount of information data processed by the preset visual modal recognition model, preset audio modal recognition model, and preset sensor modal recognition model, improve the corresponding information data processing efficiency, and, since the interference and noise of other different categories of information data are reduced, the recognition accuracy of the corresponding historical target object can be further improved, thereby further achieving the improvement of the accuracy of traffic dynamic recognition.

[0091] In an embodiment of the present invention, historical traffic information data corresponding to a target road section is determined, and the historical traffic information data is identified based on a preset object recognition model to obtain a number of historical target objects. Then, based on a preset object clustering model, all historical target objects are clustered to obtain a number of cluster clusters. Each cluster cluster is then used as a target object of a corresponding single type to obtain a number of single-class target objects. This achieves automatic classification based on dimensional features of different target objects, including but not limited to attributes, shape, dynamics, etc., and through clustering to perform the same or similar classification. Therefore, the single-class target objects of each category have high similarity and are clearly distinguishable from other single-class target objects. Then, the same single-class target recognition model is used for target object recognition. Since each single-class target recognition model only needs to focus on identifying target objects of the same type, it can not only reduce the complexity of the single-class target recognition model, but also improve the recognition accuracy of the single-class target recognition model. Then, combined with the collection of road condition information data, the accuracy of traffic dynamic recognition is improved from the two perspectives of road condition information data collection and target object recognition, and accurate recognition of traffic dynamics is achieved, thereby assisting intelligent transportation and improving the safety, efficiency and effectiveness of intelligent transportation.

[0092] In one embodiment, the traffic dynamics presentation module 105 is specifically used to present all the target objects on the virtualized road corresponding to the target road section based on a preset traffic dynamics digital twin method, and obtain the traffic dynamics corresponding to the target road section.

[0093] Explanatory speaking, Digital Twin is a virtual model that is constructed in virtual space through digital means to mirror the road traffic status in real time, and uses data interaction and simulation analysis to present the operation of physical entities. The combination of traffic dynamics and digital twins drives the virtual model through relevant road condition information data of real-time traffic dynamics to achieve holographic perception, accurate prediction and intelligent presentation of the traffic system. According to the above description, the traffic dynamics digital twin mode is pre-set, that is, the traffic dynamics digital twin mode is preset. The preset traffic dynamics digital twin mode represents a traffic dynamics presentation mode that uses digital twins to conduct "virtual and real symbiosis" of traffic dynamics.

[0094] According to the above conception and setting, the traffic dynamics presentation module 105 is specifically used to present all target objects on the virtual road corresponding to the target section based on the preset traffic dynamics digital twin method, obtain the traffic dynamics corresponding to the target section, and realize the dynamic digital mirroring of the physical world of traffic dynamics.

[0095] The embodiment of the present invention presents all target objects on the virtual road corresponding to the target section based on a preset traffic dynamics digital twin method, obtains the traffic dynamics corresponding to the target section, and optimizes the presentation of the traffic dynamics corresponding to the target section by using data interaction and simulation analysis. It can achieve holographic perception, real-time, realistic, intuitive and accurate presentation of traffic dynamics, thereby assisting intelligent transportation and improving the safety, efficiency, effectiveness and intelligent optimization of intelligent transportation.

[0096] It should be noted that the artificial intelligence-based traffic dynamic identification system described in the above embodiments can recombine the technical features contained in different embodiments as needed to obtain a combined implementation plan, but all of them are within the scope of protection required by the present invention.

[0097] Each module in the aforementioned AI-based traffic dynamics recognition system can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0098] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium may be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0099] The software tools, components, and models not produced by our company that appear in the embodiments of the present invention are for illustrative purposes only and do not represent actual use.

[0100] The relevant data collection in the embodiments of the present invention complies with the requirements of relevant laws and regulations, such as China's Personal Information Protection Law, GDPR (EU General Data Protection Regulation) or information security standards of other countries and regions.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A traffic dynamic recognition system based on artificial intelligence, characterized in that: include: A first determination module is configured to respond to a traffic dynamics recognition instruction, determine a target road section, and determine a road condition type corresponding to the target road section; A second determining module is used to determine a road condition information collection method according to the road condition type; An information collection module, configured to obtain road condition information data based on the road condition information collection method; A target recognition module, configured to identify target objects of corresponding types based on the traffic information data and a plurality of preset single-class target recognition models; The traffic dynamics presentation module is used to present all the target objects on the virtualized road corresponding to the target road section to obtain the traffic dynamics corresponding to the target road section.

2. The traffic dynamic recognition system based on artificial intelligence according to claim 1, characterized in that: The first determination module includes: a first acquisition submodule and a first determination submodule; The first acquisition submodule is configured to acquire at least one of the following: Used to obtain the road type corresponding to the target road section; Used to obtain the weather conditions corresponding to the target road section; Used to obtain the current time; The first determining submodule is configured to determine the road condition type corresponding to the target road section according to the road type, the weather condition and / or the current time.

3. The traffic dynamic recognition system based on artificial intelligence as claimed in claim 2, characterized in that: The first determining submodule includes at least one of the following: The second determining submodule is configured to determine that the road condition type corresponding to the target road section is "urban sunny" when the road type is "urban road" and the weather condition is "sunny"; A third determining submodule is configured to determine that the road condition type corresponding to the target road section is "city night" when the road type is "city road" and the current time is "night"; A fourth determining submodule is configured to determine that the road condition type corresponding to the target road section is "daytime city rainy day" when the road type is "city road", the weather condition is "rainy day", and the current time is "daytime"; A fifth determining submodule is configured to determine, when the road type is "highway" and the weather condition is "sunny", that the road condition type corresponding to the target road section is "highway sunny"; A sixth determining submodule is configured to determine, when the road type is "highway" and the current time is "nighttime", that the road condition type corresponding to the target road section is "highway nighttime"; The seventh determination submodule is used to determine that the road condition type corresponding to the target section is "daytime high-speed rainy day" when the road type is "highway", the weather condition is "rainy day", and the current time is "daytime".

4. The traffic dynamics recognition system based on artificial intelligence as claimed in claim 3, characterized in that: The second determining module includes at least one of the following: an eighth determining submodule, configured to determine, when the road condition type is "city sunny", that the road condition information collection method is "high-definition camera + millimeter-wave radar + geomagnetic coil"; a ninth determining submodule, configured to, when the road condition type is "city night", determine that the road condition information collection method is "infrared thermal imaging camera + low-light starlight-level camera + millimeter-wave radar"; The tenth determination submodule is used to determine that the road condition information collection method is "millimeter wave radar + infrared camera + LiDAR system" when the road condition type is "daytime city rainy day".

5. The traffic dynamic recognition system based on artificial intelligence as claimed in claim 3, characterized in that: The second determining module includes at least one of the following: an eleventh determining submodule, configured to determine, when the road condition type is "high-speed sunny day," that the road condition information collection method is "LiDAR system + high-definition camera + microwave radar"; A twelfth determining submodule is configured to determine, when the road condition type is "high-speed nighttime", that the road condition information collection method is "long-range infrared lidar + microwave radar array + high-sensitivity panoramic camera"; The thirteenth determination submodule is used to determine that the road condition information collection method is "millimeter wave radar + infrared thermal imaging + meteorological sensor" when the road condition type is "daytime high speed rainy day".

6. The traffic dynamics recognition system based on artificial intelligence according to any one of claims 1 to 5, characterized in that: The traffic dynamic identification system also includes: A third determination module is used to determine a number of single-type target objects corresponding to the target road section; The fourth determining module is used to determine a corresponding preset single-class target recognition model according to each type of the single-class target object, and obtain a plurality of preset single-class target recognition models.

7. The traffic dynamics recognition system based on artificial intelligence according to claim 6, characterized in that: The third determining module includes: A fourteenth determining submodule is used to determine the historical traffic information data corresponding to the target road section; A first identification submodule is configured to identify the historical traffic information data based on a preset object recognition model to obtain a plurality of historical target objects; A clustering submodule, configured to cluster all the historical target objects based on a preset object clustering model to obtain a plurality of clusters; The fifteenth determining submodule is configured to take each of the clusters as a target object of a corresponding single type, and obtain a plurality of single-type target objects.

8. The traffic dynamics recognition system based on artificial intelligence according to claim 7, characterized in that: The first identification submodule includes at least one of the following: A second recognition submodule is configured to perform visual modality recognition on the historical traffic information data based on a preset visual modality recognition model to obtain an image historical target object; a third identification submodule, configured to perform audio modality recognition on the historical traffic information data based on a preset audio modality recognition model to obtain an audio history target object; a fourth identification submodule, configured to perform sensor modality identification on the historical traffic information data based on a preset sensor modality identification model to obtain a sensor modality historical target object; a first cross-modal recognition submodule, configured to perform cross-modal recognition on the historical traffic information data based on a preset multi-modal recognition model to obtain a multi-modal historical target object; In addition, the first identification submodule further includes: a first union submodule, configured to union the image history target object, the audio history target object, the sensor modality history target object, and the multimodal history target object to obtain a plurality of history target objects.

9. The traffic dynamics recognition system based on artificial intelligence according to claim 7, characterized in that: The historical traffic information data includes at least one of the following: historical traffic visual information data, historical traffic audio information data, and historical traffic sensor information data; the first identification submodule includes at least one of the following: a fifth recognition submodule, configured to recognize the historical traffic visual information data based on a preset visual modality recognition model to obtain an image historical target object; a sixth identification submodule, configured to identify the historical traffic audio information data based on a preset audio modality recognition model to obtain an audio history target object; a seventh identification submodule, configured to identify the historical traffic sensor information data based on a preset sensor modality identification model to obtain a sensor modality historical target object; a second cross-modal recognition submodule, configured to perform cross-modal recognition on the historical traffic information data based on a preset multi-modal recognition model to obtain a multi-modal historical target object; In addition, the first identification submodule further includes: a second union submodule, which is used to take the union of the image historical target object, the audio historical target object, the sensor modality historical target object, and the multimodal historical target object to obtain a plurality of historical target objects.

10. The traffic dynamics recognition system based on artificial intelligence according to any one of claims 1 to 5, characterized in that: The traffic dynamics presentation module is specifically used to present all the target objects on the virtualized road corresponding to the target section based on a preset traffic dynamics digital twin method, and obtain the traffic dynamics corresponding to the target section.

Images