CN117456737B

CN117456737B - Intelligent traffic big data processing method and system based on 3D visual intelligence

Info

Publication number: CN117456737B
Application number: CN202311784772.4A
Authority: CN
Inventors: 李华栈; 苏鑫煌; 蔡升沿
Original assignee: Guangdong Bangsheng Beidou Technology Co ltd
Current assignee: Guangdong Bangsheng Beidou Technology Co ltd
Priority date: 2023-12-24
Filing date: 2023-12-24
Publication date: 2024-03-26
Anticipated expiration: 2043-12-24
Also published as: CN117456737A

Abstract

The invention relates to the technical field of image processing and intelligent traffic, in particular to a method and a system for processing intelligent traffic big data based on 3D visual intelligence. The embodiment of the invention realizes the integration of the space-time characteristic connection of the intelligent traffic participation objects in the judging process of the traffic state prediction risk, so that the accuracy and the rationality of the traffic state prediction risk mining can be improved.

Description

Intelligent traffic big data processing method and system based on 3D visual intelligence

Technical Field

The invention relates to the technical field of image processing and intelligent traffic, in particular to an intelligent traffic big data processing method and system based on 3D visual intelligence.

Background

Intelligent traffic big data processing in modern traffic systems is a complex and important task. With the development of 3D vision technology and intelligent traffic systems, accurate and efficient processing of traffic image information to achieve safe and efficient traffic management is an urgent problem.

In the conventional traffic state prediction and risk discrimination processes, analysis and prediction are mainly performed through the space-time characteristics of single traffic participation objects (such as vehicles, pedestrians and the like). However, this approach often ignores interactions between multiple intelligent traffic participants with spatiotemporal linkages, which may lead to poor accuracy and rationality of predictions. For example, if two vehicles travel on the same road segment, their motion states may be affected by each other, which is not fully considered in the conventional predictive model.

Therefore, a new technology is urgently needed to realize more accurate and reasonable traffic state prediction and risk discrimination.

Disclosure of Invention

In order to improve the technical problems in the related art, the invention provides a method and a system for processing intelligent traffic big data based on 3D visual intelligence.

In a first aspect, an embodiment of the present invention provides a 3D visual intelligence-based intelligent traffic big data processing method, which is applied to an intelligent traffic big data processing system, and the method includes:

acquiring 3D visual traffic image information to be subjected to traffic state prediction risk determination, wherein the 3D visual traffic image information comprises a plurality of intelligent traffic participation objects with space-time connection, and the traffic state prediction risk comprises a congestion aggravation state label, a congestion slowing state label or a non-congestion association state label existing between the two intelligent traffic participation objects;

performing space-time state description mining on the intelligent traffic participation object to obtain a first space-time state description vector of the intelligent traffic participation object;

vector integration is carried out on the first time-space state description vectors of a plurality of intelligent traffic participation objects with time-space connection, so as to obtain second time-space state description vectors of the intelligent traffic participation objects;

and carrying out traffic state prediction analysis on the two intelligent traffic participation objects based on the first space-time state description vector and the second space-time state description vector to obtain traffic state prediction risks of the two intelligent traffic participation objects.

Optionally, based on the first space-time state description vector and the second space-time state description vector, performing traffic state prediction analysis on two intelligent traffic participation objects to obtain traffic state prediction risks of the two intelligent traffic participation objects, including:

integrating the first space-time state description vector and the second space-time state description vector of the intelligent traffic participation object to obtain a third space-time state description vector of the intelligent traffic participation object;

performing full connection processing on third space-time state description vectors of two intelligent traffic participation objects according to a preset full connection layer to obtain traffic state discrimination coefficients of the two intelligent traffic participation objects, wherein the traffic state discrimination coefficients comprise the possibility of discriminating traffic state prediction risks of the two intelligent traffic participation objects into a congestion aggravation state label, a congestion slowing state label or a non-congestion association state label;

and determining traffic state prediction risks of the two intelligent traffic participation objects based on the maximum discrimination coefficient of the traffic state discrimination coefficients.

Optionally, when the traffic state prediction risk of the two intelligent traffic participation objects is a non-congestion association state label, the maximum discrimination coefficient of the traffic state discrimination coefficient and the space-time state description commonality score of the two intelligent traffic participation objects have a first quantization relationship;

And when the traffic state prediction risks of the two intelligent traffic participation objects are congestion aggravated state labels or congestion slowing state labels, the maximum discrimination coefficient of the traffic state discrimination coefficients and the space-time state description commonality scores of the two intelligent traffic participation objects have a second quantization relationship.

Optionally, when the traffic state prediction risk of the two intelligent traffic participation objects is a non-congestion association state label, the space-time state description commonality score of the two intelligent traffic participation objects has a second quantization relationship with the path space distribution difference of the two intelligent traffic participation objects and the contribution weight value of the traffic state discrimination coefficient;

and when the traffic state prediction risks of the two intelligent traffic participation objects are congestion aggravation state labels or congestion slowing state labels, the space-time state description commonality scores of the two intelligent traffic participation objects have a first quantization relation with the contribution weight values of the traffic state discrimination coefficients and the path space distribution differences of the two intelligent traffic participation objects.

Optionally, before acquiring the 3D visual traffic image information to be subjected to traffic state prediction risk determination, the method further comprises:

Acquiring a traffic state prediction risk processing network for carrying out traffic state prediction risk determination on 3D visual traffic image information, wherein the traffic state prediction risk processing network comprises a space-time state description mining branch for carrying out space-time state description mining on intelligent traffic participation objects and a traffic state prediction branch for carrying out traffic state prediction analysis on two intelligent traffic participation objects;

acquiring 3D visual traffic image information commissioning examples for commissioning the traffic state prediction risk processing network, the 3D visual traffic image information commissioning examples comprising a number of intelligent traffic participation object commissioning examples with spatiotemporal association, and an a priori traffic state perspective for representing traffic state prediction risk between two intelligent traffic participation object commissioning examples;

acquiring a space-time state description vector debugging example obtained by performing space-time state description mining on the intelligent traffic participation object debugging example through the space-time state description mining branch, and determining a first network debugging cost function of the two intelligent traffic participation object debugging examples based on the space-time state description vector debugging example and the prior traffic state viewpoint;

Acquiring a traffic state prediction thermodynamic diagram obtained by carrying out traffic state prediction analysis on the two intelligent traffic participation object debugging examples based on the space-time state description vector debugging examples through the traffic state prediction branch, and determining a second network debugging cost function of the two intelligent traffic participation object debugging examples based on the traffic state prediction thermodynamic diagram and the prior traffic state viewpoint;

and improving network parameters of the traffic state prediction risk processing network based on the first network debugging cost function and the second network debugging cost function.

Optionally, before acquiring the space-time state description vector debug example obtained by space-time state description mining on the intelligent traffic participation object debug example through the space-time state description mining branch, the method further includes:

and masking the traffic participation units or the traffic road sections in the 3D visual traffic image information debugging example.

Optionally, masking the traffic participation units or the traffic road sections in the 3D visual traffic image information debug example includes:

detecting a target traffic participation unit related to the traffic state prediction risk from the 3D visual traffic image information commissioning example;

Masking the target traffic participant unit in the 3D visual traffic image information commissioning example with a first set probability value;

masking of the intelligent traffic participation object commissioning instance in the 3D visual traffic image information commissioning instance is performed with a second set probability value.

Optionally, determining a first network commissioning cost function of two intelligent traffic participant commissioning examples based on the spatiotemporal state description vector commissioning example and the a priori traffic state perspective comprises:

determining a spatiotemporal state description commonality score for two intelligent traffic participant debugging examples based on the spatiotemporal state description vector debugging examples;

a first network debug cost function of the two intelligent traffic participant debug examples is determined based on the spatiotemporal state descriptive commonality score and the a priori traffic state perspectives.

Optionally:

when the prior traffic state view is a non-congestion association state label, the first network debugging cost function and the space-time state description commonality scores of the two intelligent traffic participation object debugging examples have a first quantization relation;

and when the prior traffic state viewpoint is a congestion aggravated state label or a congestion slowed state label, the first network debugging cost function and the space-time state description commonality scores of the two intelligent traffic participation object debugging examples have a second quantization relation.

Optionally, when the prior traffic state view is a non-congestion association state label, the space-time state description commonality score of the two intelligent traffic participation object debugging examples has a second quantization relationship with a contribution weight value of the first network debugging cost function and a path space distribution difference of the two intelligent traffic participation object debugging examples;

and when the prior traffic state viewpoint is a congestion acceleration state label or a congestion slowing state label, the space-time state description commonality score of the two intelligent traffic participation object debugging examples has a first quantized relationship with the contribution weight value of the first network debugging cost function and the path space distribution difference of the two intelligent traffic participation object debugging examples.

Optionally, the traffic state prediction branch comprises a feature mining node, a feature mapping node and a feature prediction node which are cascaded;

obtaining a traffic state prediction thermodynamic diagram obtained by the traffic state prediction branch through traffic state prediction analysis of the two intelligent traffic participation object debugging examples based on the space-time state description vector debugging examples, wherein the traffic state prediction thermodynamic diagram comprises the following steps:

acquiring initial space-time state description features obtained by feature mining of space-time state description vector debugging examples of the two intelligent traffic participation object debugging examples through the feature mining node;

Acquiring space-time state numerical mapping characteristics obtained by performing interval numerical mapping on the initial space-time state description characteristics through the characteristic mapping nodes;

and acquiring a traffic state prediction thermodynamic diagram obtained by carrying out traffic state discrimination operation on the space-time state numerical value mapping characteristics through the characteristic prediction node.

Optionally, improving the network parameters of the traffic state prediction risk processing network based on the first network debugging cost function and the second network debugging cost function includes:

integrating the first network debugging cost function and the second network debugging cost function according to the set confidence level to obtain a global network debugging cost variable of the traffic state prediction risk processing network;

feeding back the global network debugging cost variable in the traffic state prediction risk processing network to obtain the loss function change of each network parameter;

the network parameter is modified based on the change in the loss function.

In a second aspect, the invention also provides an intelligent traffic big data processing system, which comprises a processor and a memory; the processor is in communication with the memory, and the processor is configured to read and execute a computer program from the memory to implement the method described above.

In a third aspect, the present invention also provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the method described above.

According to the technical scheme provided by the embodiment of the invention, the space-time state description of the intelligent traffic participation objects in the 3D visual traffic image information is mined to obtain the first space-time state description vector, and the first space-time state description vectors of the plurality of intelligent traffic participation objects with space-time connection are integrated to obtain the second space-time state description vector, so that traffic state prediction analysis can be carried out on the two intelligent traffic participation objects according to the first space-time state description vector and the second space-time state description vector to obtain traffic state prediction risks of the two intelligent traffic participation objects. The embodiment of the invention realizes the integration of the space-time characteristic connection of the intelligent traffic participation objects in the judging process of the traffic state prediction risk, so that the accuracy and the rationality of the traffic state prediction risk mining can be improved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

Fig. 1 is a schematic flow chart of a smart traffic big data processing method based on 3D visual intelligence according to an embodiment of the present invention.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the invention. Rather, they are merely examples of apparatus and methods consistent with aspects of the invention.

It should be noted that the terms "first," "second," and the like in the description of the present invention and the above-described drawings are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.

The method embodiment provided by the embodiment of the invention can be executed in an intelligent traffic big data processing system, a computer device or a similar computing device. Taking as an example operation on a smart traffic big data processing system, the smart traffic big data processing system may comprise one or more processors (which may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory for storing data, and optionally the smart traffic big data processing system may further comprise a transmission device for communication functions. It will be appreciated by those skilled in the art that the above-described architecture is merely illustrative and is not intended to limit the architecture of the intelligent traffic big data processing system. For example, the intelligent traffic big data processing system may also include more or fewer components than shown above, or have a different configuration than shown above.

The memory may be used to store a computer program, for example, a software program of application software and a module, for example, a computer program corresponding to a smart traffic big data processing method based on 3D visual intelligence in the embodiment of the present invention, and the processor executes various functional applications and data processing by running the computer program stored in the memory, that is, implements the method. The memory may include high speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid state memory. In some examples, the memory may further include memory remotely located with respect to the processor, the remote memory being connectable to the intelligent transportation big data processing system through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission means is used for receiving or transmitting data via a network. Specific examples of the network described above may include a wireless network provided by a communications provider of the intelligent transportation big data processing system. In one example, the transmission means comprises a network adapter (Network Interface Controller, simply referred to as NIC) that can be connected to other network devices via a base station to communicate with the internet. In one example, the transmission device may be a Radio Frequency (RF) module, which is used to communicate with the internet wirelessly.

Referring to fig. 1, fig. 1 is a flow chart of a method for processing intelligent traffic big data based on 3D visual intelligence according to an embodiment of the present invention, where the method is applied to an intelligent traffic big data processing system, and further includes steps 110-140.

Step 110, obtaining 3D visual traffic image information to be subjected to traffic state prediction risk determination, wherein the 3D visual traffic image information comprises a plurality of intelligent traffic participation objects with space-time connection, and the traffic state prediction risk comprises a congestion aggravation state label, a congestion slowing state label or a non-congestion association state label existing between two intelligent traffic participation objects.

In step 110, the 3D visual traffic image information refers to a traffic scene image captured using a 3D camera or sensor. The images contain information of multiple dimensions, so that the positions and the appearances of intelligent traffic participation objects such as vehicles, pedestrians and the like are recorded, and depth information, namely the distance between an object and a camera, can be provided. For example, a 3D image of an intersection might show several cars waiting for a signal, pedestrians traveling on crosswalks, and bicycles traveling beside a lane.

Several intelligent traffic participants with spatiotemporal association are active units in a traffic system, such as vehicles, pedestrians, bicycles, etc. The spatiotemporal relationship between them means that they are within mutually influential spatial ranges at the same time. For example, one vehicle driving next to another indicates that there is a spatiotemporal relationship between the two; likewise, if a pedestrian is preparing to traverse a road and a vehicle is approaching a crosswalk, then there is also a space-time relationship between the vehicle and the pedestrian.

In certain spatiotemporal contexts, a congestion acceleration status tag is assigned when interaction of two or more intelligent traffic participants may result in a slowing or stopping of traffic flow. For example, if one traffic light fails to cause vehicles on a plurality of roads to pass normally, and traffic is severely jammed, the traffic between the related vehicles is marked with a jam aggravating status tag.

Conversely, if the interaction of two or more intelligent traffic participants can promote smooth traffic, such as by intelligent lights coordinating the smooth passage of vehicles across intersections, a congestion reduction status tag is assigned between the participants.

A non-congestion association status tag may be assigned when the interaction between two intelligent traffic participant objects has no significant effect on traffic flow. For example, a car with a stable speed is driven on a highway and kept at a safe distance from other vehicles, in which case a non-congestion-related state can be considered.

And 120, performing space-time state description mining on the intelligent traffic participation object to obtain a first space-time state description vector of the intelligent traffic participation object.

In step 120, the process of spatiotemporal state description mining involves analyzing the behavioral characteristics of each intelligent traffic participant in time and space. For a vehicle, this may include its speed, acceleration, position coordinates, direction of travel, etc.; for pedestrians, one may be concerned with their speed, movement trajectory, relative distance to surrounding vehicles or other pedestrians, etc.

The first spatio-temporal state description vector is a numerical vector into which the mined spatio-temporal state description is converted, which vector can represent the state characteristics of the participating objects. For example, for a vehicle, its first empty state description vector may be as follows:

[ \mathbf { V } _ { \text { vehicle } = [ x, y, z, v_x, v_y, a_x, a_y, t ] ].

Where (x, y, z) represents the position coordinates of the vehicle in three-dimensional space, (v_x, v_y) represents the velocity components along the x-axis and the y-axis, (a_x, a_y) represents the acceleration components along the x-axis and the y-axis, and (t) represents the current time stamp.

For example, at an intersection, there are two vehicles (vehicle a and vehicle B) approaching the intersection, whose dynamic and static information needs to be converted into vector form. First, information such as real-time position, speed, acceleration and the like of a vehicle is acquired by using a 3D vision technology. For vehicle a, it is assumed that it is currently located on the north side of the intersection, traveling southward at a speed of 5 meters/second, and is decelerating (acceleration is-2 meters/second). For vehicle B, it is assumed to be located on the east side of the intersection, traveling westernly at a speed of 3 m/s, the speed being kept constant (acceleration is 0 m/s). The following vectors may be generated for car a and car B:

[ _mathbf { V } { _text { car A } = [ x { _text { A }, y_ { _text { A }, z_ { _text { A },0, -5,0, -2, t_ { text { A } ];

[ _mathbf { V } { _text { car B } = [ x { _text { B }, y_ { _text { B }, z_ { _text { B }, -3,0, t_ { _text { B } ];

Wherein (x_ { text { A }, y_ { text { A }, z_ { text { A }) and (x_ { text { B }, y_ { text { B }, z_text { B }) represent coordinates of the vehicles A and B in the three-dimensional space, respectively, (t_text { A }) and (t_text { B }) represent respective time stamps.

A detailed numerical characterization may be established for each intelligent traffic participant via step 120, which provides the necessary inputs for the next step 130, namely integrating the vectors to obtain a more comprehensive traffic situation description.

And 130, vector integration is carried out on the first space-time state description vectors of the plurality of intelligent traffic participation objects with space-time connection, so as to obtain a second space-time state description vector of the intelligent traffic participation objects.

In the preceding steps, features have been extracted for each intelligent traffic participant (such as a vehicle, pedestrian, etc.) based on its spatiotemporal activity in the traffic network and a so-called first spatiotemporal state description vector is formed. These vectors may include position, velocity, acceleration, direction, and time characteristics.

Vector integration: in step 130, a first spatio-temporal state description vector of a plurality of intelligent traffic participant objects having spatio-temporal associations is integrated. Suppose an intersection is observed, which involves a plurality of vehicles (a, B, c.) and pedestrians (X, Y, z.). Each participant has its own first empty state description vector, e.g. the vector of vehicle a is [ a position, a speed, a acceleration, a direction, a time ].

Constructing a second space-time state description vector: to fully understand the overall situation in a traffic network, it is necessary to integrate the state description vectors of these individuals. This integration process may take into account interactions and associations between different objects. For example, if vehicle a is very close to vehicle B and their speeds differ widely, this situation may lead to potential risks. Thus, a new vector [ A state, B state, A-B distance, A-B speed difference, time ] can be created, which is a simplified second spatiotemporal state description vector example.

The second spatio-temporal state description vector is not just a mere stacked individual feature, but it is also possible to introduce more advanced features such as group behavior patterns, traffic flow densities, traffic pressure indexes of specific areas, etc., which can be calculated and inferred from the data in the first spatio-temporal state description vector.

Finally, a second spatiotemporal state description vector may be used to better predict traffic conditions, such as whether congestion will occur, which intersections may have traffic accidents, and so on. For example, if a vector shows a sudden increase in vehicle density at a certain intersection, with a significant drop in traffic flow, this may mean that a congestion situation is imminent.

By such vector integration, complex dynamics of the traffic network can be more fully analyzed and understood, and support is provided for further predictions and decisions.

And 140, carrying out traffic state prediction analysis on the two intelligent traffic participation objects based on the first space-time state description vector and the second space-time state description vector to obtain traffic state prediction risks of the two intelligent traffic participation objects.

In step 140, a first spatio-temporal state description vector (a feature description of a single object) and a second spatio-temporal state description vector (an integrated feature description of multiple objects) are utilized, which vectors contain information about the position, velocity, acceleration, etc. of the traffic participant object. By comparing the vector changes at different points in time, patterns and trends in traffic flow can be identified.

Further, traffic state predictive analysis is a process that utilizes historical data, current state, and possibly trends to predict future traffic conditions. It involves machine learning or deep learning algorithms to build predictive models, e.g., a classifier can be trained to determine whether traffic conditions between two objects are evolving toward a congestion exacerbation, congestion relief, or non-congestion association.

At this stage, the system evaluates the traffic state risk between the two intelligent traffic participants. For example, if the trajectories of vehicle a and vehicle B at an intersection have a tendency to meet, and the speeds and accelerations of the two indicate that they may collide at close temporal and spatial points, the system may tag both objects with a high risk "congestion acceleration status tag".

For example, there are two vehicles, vehicle a and vehicle B, approaching the same intersection from north-south and east-west, respectively, without signal lamp control. In step 130, their first and second spatio-temporal state description vectors are obtained. These vectors will now be used to analyze their traffic status and predict risk in step 140.

Wherein the speed of vehicle B remains unchanged although vehicle a is decelerating, and their paths have an intersection at the intersection. Based on this information and the patterns learned from the historical data, the model may predict that if vehicle B is not decelerating, there will be a high risk of collision for both vehicles. Thus, the model assigns a "congestion acceleration status tag" to the interaction between the two vehicles, and perhaps also triggers a warning to slow down the driver of vehicle B or take other risk-avoiding action. Ultimately, the result of step 140 is such a prediction of future traffic state risk, which can be used to take steps in advance to reduce traffic congestion and improve road safety.

Next, steps 110-140 are described by the relevant application scenario.

At a busy urban intersection, a high-precision 3D camera system is installed for capturing traffic in real time. This system is capable of recording the position, speed, acceleration and other relevant parameters of various intelligent traffic participants (e.g., different types of vehicles, pedestrians, bicycles, etc.). At a particular moment, the system captures a set of 3D visual image data showing two vehicles (vehicle a and vehicle B) approaching an intersection and a group of pedestrians (pedestrian group X) ready to cross the zebra crossing.

Based on the obtained data, each intelligent traffic participation object is analyzed in detail, key space-time state information is extracted, and a first space-time state description vector is constructed. For example:

the vector of vehicle a may include its coordinates in three-dimensional space (x_a, y_a, z_a), velocity v_a, acceleration a_a, and current timestamp t_a.

The vector of vehicle B may contain similar information (x_b, y_b, z_b, v_b, a_b, t_b).

The vector of pedestrian group X may then include their average positions (x_x, y_x, z_x), average velocity v_x and current timestamp t_x.

Next, the first empty state description vectors of all intelligent traffic participants are integrated. Since vehicle a, vehicle B and pedestrian group X are all within the same spatiotemporal region, their vectors need to be comprehensively considered to evaluate overall traffic conditions. Thus, by some algorithm (such as using a distance function or neural network model), a second spatio-temporal state description vector may be generated that may contain the distance d_ab between vehicles a and B, the nearest distances d_ax and d_bx between the vehicles and the pedestrian group, and the timestamp t of the current overall scene.

Finally, the traffic state between the two intelligent traffic participants is predicted and analyzed by using the first space-time state description vector and the second space-time state description vector. By means of a pre-trained prediction model, whether the vehicles A and B can cause congestion aggravation, congestion relief or non-congestion association states can be estimated. The model may take into account the following factors: whether the vehicles a and B are decelerating to avoid collision with the pedestrian group X in front, thereby possibly causing congestion of traffic in the rear; whether the pedestrian group X affects the normal running track of the vehicles a and B, thereby changing the traffic flow state.

Through the above analysis, if the model predicts that there is a risk of congestion exacerbation between vehicles a and B, the traffic management system may take corresponding action, such as adjusting the timing of the signal lights, or sending a warning to the vehicle that is about to enter the intersection. Thus, the intelligent transportation system can more intelligently and actively manage and optimize urban traffic through the entire flow of steps 110 to 140.

In other applications, it is desirable to predict and reduce traffic congestion and accident risk by analyzing 3D visual information at an urban intersection. For this purpose, intelligent traffic monitoring systems equipped with 3D cameras are deployed.

The 3D camera captures stereoscopic images of traffic participation objects (e.g., automobiles, pedestrians, bicycles, etc.) in real time and transmits the image information to the central processing unit. These images contain spatial information of the position, speed, direction, etc. of the object.

The captured images are analyzed using computer vision and machine learning algorithms to extract features of each participating object (e.g., vehicle type, speed, acceleration, pedestrian flow properties, etc.). These features are combined into a state description vector that describes the spatio-temporal state of each object.

And integrating vectors of a plurality of traffic participation objects by combining data from different 3D cameras to form a comprehensive second space-time state description vector. This vector reflects the overall traffic conditions of the intersection, including traffic density, flow trends, potential conflict points, etc.

A second spatiotemporal state description vector is input for analysis using a pre-trained predictive model (possibly based on a deep learning neural network). The model evaluates the trend of traffic flow over a period of time in the future, identifies factors that may lead to congestion or accidents, and gives a risk assessment. The prediction results may be used to generate traffic scheduling suggestions, such as traffic light timing adjustments, traffic guidance information publications, and the like.

Real-time traffic monitoring and risk prediction can help to quickly respond to potential problems, and traffic jams and accidents are effectively avoided or reduced. The data-driven decision support enhances the overall operational efficiency and safety of the traffic system. In this way, urban traffic conditions can be more accurately and timely understood and managed.

In detail, applying the above steps 110-140, first, by deeply analyzing the contents in the 3D visual traffic image, key state information of intelligent traffic participation objects (such as vehicles, pedestrians, etc.) can be extracted from complex traffic scenes. Dynamic characteristics (such as position and speed) and static characteristics (such as type and size) of the traffic participation object are converted into quantifiable first time-space state description vectors in combination with time dimension, and structured data is provided for subsequent analysis.

Secondly, the interaction and the relevance among different objects can be captured by integrating the space-time state description vectors of a plurality of participation objects. The generated second space-time state description vector reflects the comprehensive state of the whole traffic environment and provides richer information for understanding the whole traffic flow.

Then, by combining the spatiotemporal data at the individual and population level, predictive analysis can more accurately reflect future likely traffic conditions. The prediction model can adjust the prediction strategy according to the real-time data, and dynamically adapt to the traffic environment with complex changes.

Finally, the degree of risk that may be caused by interactions between traffic participants can be quantitatively assessed, helping to take preventive measures. And the high-risk area or situation is identified through predictive analysis, so that the safety strategy is formulated, and the traffic management is optimized.

In conclusion, the method can improve the accuracy and the rationality of traffic state prediction risk mining, and enhance the effectiveness and the safety of the intelligent traffic system when processing complex traffic scenes.

According to the above steps 110-140, the following benefits can be summarized:

(1) Information extraction and characterization: by deep analysis and mining of 3D visual traffic images, key spatio-temporal state information can be extracted and converted into quantifiable vectors. The representation mode effectively digitizes and standardizes the complex traffic scene and the states of the participated objects;

(2) Data integration and global view: integrating the space-time state description vectors of a plurality of intelligent traffic participation objects can capture the global condition of the traffic environment and the interaction and relevance among the objects. Thus, not only can richer information be obtained, but also the rule of traffic flow can be understood from the macroscopic level;

(3) Accurate prediction and dynamic adaptation: based on the space-time state description vectors of the individuals and the groups, the prediction model can generate accurate traffic state prediction, and can adjust the prediction strategy according to real-time data to flexibly adapt to the changed traffic environment;

(4) Risk assessment and security optimization: the prediction results may help to quantitatively evaluate the degree of traffic risk and identify possible high risk areas or situations. This helps to take corresponding precautions, optimize traffic management policies, thereby improving the security of the overall traffic system;

(5) And the precision and rationality are improved: by integrating the space-time characteristic connection of the intelligent traffic participators, the embodiment of the invention can obviously improve the accuracy and rationality of traffic state prediction risk mining, so that traffic management decisions are more scientific and reasonable, and simultaneously, the effectiveness and safety of the intelligent traffic system in processing complex traffic scenes are improved.

In some possible embodiments, the traffic state prediction analysis is performed on the two intelligent traffic participation objects based on the first space-time state description vector and the second space-time state description vector in step 140 to obtain the traffic state prediction risk of the two intelligent traffic participation objects, including steps 141-143.

And 141, integrating the first space-time state description vector and the second space-time state description vector of the intelligent traffic participation object to obtain a third space-time state description vector of the intelligent traffic participation object.

And 142, performing full connection processing on the third space-time state description vectors of the two intelligent traffic participation objects according to a preset full connection layer to obtain traffic state discrimination coefficients of the two intelligent traffic participation objects, wherein the traffic state discrimination coefficients comprise the possibility of discriminating traffic state prediction risks of the two intelligent traffic participation objects into a congestion aggravation state label, a congestion slowing state label or a non-congestion association state label.

Step 143, determining traffic state prediction risks of the two intelligent traffic participation objects based on the maximum discrimination coefficient of the traffic state discrimination coefficients.

In the above embodiment, the third spatiotemporal state descriptive vector is a vector obtained by integrating the first and second spatiotemporal state descriptive vectors of the intelligent traffic participant. This vector contains more information, providing a more comprehensive view to understand traffic conditions.

In deep learning, the fully connected layer is a neural network layer in which all input nodes are connected to all output nodes. The fully connected layer is able to integrate the input features and generate new feature combinations.

The traffic state discrimination coefficient is an index for discriminating the traffic state prediction risk based on the result after the full connection processing. This coefficient includes the likelihood of congestion aggravated status, congestion slowed status, and non-congestion associated status, among others.

For example, an intersection is being monitored, and first and second spatiotemporal state descriptive vectors of two vehicles (as intelligent traffic participants) have been acquired.

In step 141, a third spatiotemporal state description vector is formed: and integrating the first and second space-time state description vectors of the two vehicles to obtain a third space-time state description vector. This vector captures more comprehensive information such as the position, speed, direction of the vehicle, and the relative distance and speed difference between the two vehicles.

In step 142, a full connection process is performed: and inputting the third space-time state description vector into a preset full connection layer for processing. The fully connected layer generates a traffic condition discrimination coefficient that includes the likelihood of conditions such as congestion aggravation, congestion mitigation, and non-congestion association. For example, if the relative speeds of two vehicles are fast and the distance is very close, the likelihood of congestion increases.

In step 143, a traffic state prediction risk is determined: and determining the traffic state prediction risk of the two vehicles according to the maximum value of the traffic state discrimination coefficient. For example, if the coefficient of the "congestion acceleration" state is the greatest, then it is considered that the traffic condition at this intersection is likely to deteriorate.

Thus, by integrating more information and processing using the fully connected layers, finer and more specific traffic state prediction results can be obtained. According to the real-time traffic state discrimination coefficient, the traffic management strategy can be dynamically adjusted, such as changing the time setting of the signal lamp or issuing real-time traffic information. By the method, possible traffic risks can be predicted more accurately, and measures can be taken in time, so that the overall efficiency and safety of the traffic system are improved.

In some examples, the maximum discrimination coefficient of the traffic state discrimination coefficient has a first quantitative relationship with the spatiotemporal state descriptive commonality score of the two intelligent traffic participation objects when the traffic state prediction risk of the two intelligent traffic participation objects is a non-congestion associated state label. And when the traffic state prediction risks of the two intelligent traffic participation objects are congestion aggravated state labels or congestion slowing state labels, the maximum discrimination coefficient of the traffic state discrimination coefficients and the space-time state description commonality scores of the two intelligent traffic participation objects have a second quantization relationship.

In the above example, the non-congestion associated status tag indicates that the predicted risk of traffic status for two intelligent traffic participant objects (e.g., vehicles or pedestrians) is unlikely to cause traffic congestion. Congestion acceleration status label and congestion slowing status label: these two labels represent the predicted risk of traffic state, respectively, which may lead to an exacerbation or slowing of existing traffic congestion conditions. Maximum discrimination coefficient of traffic state discrimination coefficient: this is a measure reflecting the result of the traffic condition assessment by the traffic condition prediction model. The largest discrimination coefficient represents the one with the largest prediction risk among all discrimination coefficients. Space-time state description commonality score: this is a quantitative indicator reflecting the degree of commonality of two intelligent traffic participants by comparing and integrating their spatiotemporal state descriptive vectors.

For example, at an intersection, there are two cars a and B. Through the previous steps, their spatiotemporal state description vectors have been obtained and their commonality scores calculated.

If the traffic state prediction risk of the two vehicles is marked as a non-congestion associated status tag, the maximum discrimination coefficient of the traffic state discrimination coefficients may be determined by a predetermined first quantitative relationship (possibly a function or model).

If their predicted risk is marked as a congestion acceleration status label or a congestion deceleration status label, then another, different second quantitative relationship is used for calculation.

This approach enables more flexible and accurate traffic state prediction, as it can employ different calculation methods depending on the different predicted risk types. Meanwhile, the space-time commonality among traffic participation objects is considered, and the global property and the rationality of prediction are improved.

Summarizing, the technical scheme leads the traffic state prediction to be better suitable for various complex traffic scenes by introducing different quantization relations. Whether in a non-congestion associated state or in a congestion aggravated or slowed state, more accurate risk assessment results can be obtained, thereby helping to make more efficient decisions.

In other examples, when the traffic state prediction risk of the two intelligent traffic participation objects is a non-congestion associated state label, the spatiotemporal state description commonality score of the two intelligent traffic participation objects has a second quantitative relationship to the contribution weight value of the traffic state discrimination coefficient and the path spatial distribution difference of the two intelligent traffic participation objects. And when the traffic state prediction risks of the two intelligent traffic participation objects are congestion aggravation state labels or congestion slowing state labels, the space-time state description commonality scores of the two intelligent traffic participation objects have a first quantization relation with the contribution weight values of the traffic state discrimination coefficients and the path space distribution differences of the two intelligent traffic participation objects.

For example, a spatiotemporal state descriptive commonality score is an indicator that measures the degree of similarity or difference in spatiotemporal states of two intelligent traffic participants (e.g., two vehicles). The path spatial distribution difference is another index for measuring the difference of the path distribution situation of two intelligent traffic participation objects on the road network space.

First and second quantized relationships: these are two mathematical relationships used to quantitatively describe the correlation between the weight value contributed by the spatiotemporal state description commonality score to the traffic state discrimination coefficient and the path spatial distribution difference. The specific mathematical form may be defined according to the actual application scenario and requirements.

For example, two taxis in a city are being monitored, their spatiotemporal state description vectors have been acquired, and their traffic state prediction risk calculated.

If the traffic state prediction risk is a non-congestion associated state label: in this case, the taxi travel path does not have a significant effect on the urban traffic conditions. At this time, the contribution weight value of the space-time state description commonality score to the traffic state discrimination coefficient can be adjusted according to the path space distribution difference of the taxies through the second quantization relation.

If the traffic state prediction risk is a congestion acceleration state label or a congestion slowing state label: in this case, the driving path of the taxis may have a great influence on the traffic conditions of the city. At this time, the contribution weight value of the space-time state description commonality score to the traffic state discrimination coefficient can be adjusted according to the path space distribution difference of the taxies through the first quantization relation.

In this way, according to different traffic state prediction risks, different quantization relations are adopted to adjust the weight values, so that the model can be better adapted to various traffic environments and conditions. Based on the comprehensive consideration of the path space distribution difference and the space-time state description commonality score, the traffic management system can help to make more optimal and scientific traffic management decisions. The method can improve the accuracy of traffic state prediction risk judgment, thereby improving the overall operation efficiency and the safety of the traffic system.

In another example, taking the first quantized relationship as a positive correlation and the second quantized relationship as a negative correlation as an example, a specific example can be used for further explanation.

Two cars a and B at an intersection are still taken as an example. Their spatiotemporal state description vectors have been acquired by 3D visual images and their commonality scores calculated.

Non-congestion associated status label: the assumption is made that the predictive model determines that the traffic state predictive risk of these two vehicles is marked as a non-congestion associated status tag. Because the first quantitative relationship is a positive correlation, as the spatiotemporal state descriptive commonality score of two vehicles increases (i.e., the traveling states of the two vehicles are more similar), the maximum discrimination coefficient of their traffic state discrimination coefficients also increases. This means that if the driving conditions of two vehicles are more similar, the less likely they are to cause traffic congestion.

Congestion acceleration status tag or congestion mitigation status tag: if the traffic state prediction risk of the two vehicles is marked as a congestion acceleration state label or a congestion deceleration state label, then a second quantitative relationship is used for calculation. Since the second quantitative relationship is a negative correlation, this means that as the spatiotemporal state description commonality score of two vehicles increases, their maximum discrimination coefficient of traffic state discrimination coefficients decreases. In other words, if the driving conditions of two vehicles are more similar, the less likely they are to cause congestion to be exacerbated or slowed down.

Through the above examples, it can be seen that the setting of the quantization relation enables the prediction model to be dynamically adjusted according to actual conditions, so as to better adapt to various complex traffic scenes. The model provides accurate risk assessment, both in non-congested and congested conditions, to help make more efficient decisions.

In some alternative embodiments, the method further comprises steps 210-250 before the 3D visual traffic image information to be subjected to traffic state prediction risk determination is obtained as described in step 110.

Step 210, a traffic state prediction risk processing network for determining traffic state prediction risk of 3D visual traffic image information is obtained, wherein the traffic state prediction risk processing network comprises a space-time state description mining branch for performing space-time state description mining on intelligent traffic participation objects and a traffic state prediction branch for performing traffic state prediction analysis on two intelligent traffic participation objects.

The traffic state prediction risk processing network is used for processing the 3D visual traffic image information and predicting traffic state risks. This network has two branches: the spatiotemporal state describes the mining branch and the traffic state prediction branch. The former is used for acquiring the space-time information of traffic participation objects (such as automobiles, pedestrians and the like), and the latter is used for predicting traffic states based on the space-time information.

For example, the space-time state description mining branch may obtain information such as a position, a speed, a direction, etc. of each vehicle, and the traffic state prediction branch may predict a future traffic state according to the information, for example, whether congestion may occur.

Step 220, obtaining a 3D visual traffic image information commissioning example for commissioning the traffic state prediction risk processing network, the 3D visual traffic image information commissioning example comprising a number of intelligent traffic participation object commissioning examples with spatiotemporal association, and an a priori traffic state perspective for representing traffic state prediction risk between the two intelligent traffic participation object commissioning examples.

This step obtains 3D visual traffic image information for commissioning the predictive risk processing network. The information includes intelligent traffic participation objects with temporal and spatial links, and a priori traffic state views representing the predicted risk of traffic states between these objects. For example, 3D visual traffic images of an intersection over a period of time may be collected and marked as to which conditions traffic congestion may occur.

Step 230, acquiring a space-time state description vector debugging example obtained by performing space-time state description mining on the intelligent traffic participation object debugging example through the space-time state description mining branch, and determining a first network debugging cost function of the two intelligent traffic participation object debugging examples based on the space-time state description vector debugging example and the prior traffic state viewpoint.

And processing the debugging examples through a space-time state description mining branch to obtain space-time state description vectors. Then, based on this vector and the a priori traffic state perspective, a first network debug cost function is determined. For example, the gap between the prediction result and the real situation (a priori traffic state point of view) may be calculated as the first network debug cost function.

Step 240, obtaining a traffic state prediction thermodynamic diagram obtained by performing traffic state prediction analysis on the two intelligent traffic participation object debugging examples based on the space-time state description vector debugging examples through the traffic state prediction branch, and determining a second network debugging cost function of the two intelligent traffic participation object debugging examples based on the traffic state prediction thermodynamic diagram and the prior traffic state viewpoint.

And processing the space-time state description vector through a traffic state prediction branch to obtain a traffic state prediction thermodynamic diagram. Then, based on this thermodynamic diagram and a priori traffic state perspective, a second network debug cost function is determined. For example, a thermodynamic diagram representing the predicted congestion probability may be generated, and then the gap between this predicted result and the real situation (a priori traffic state perspective) is calculated as a second network commissioning cost function.

In other possible examples, the traffic state prediction thermodynamic diagram is a visualization tool for representing a predicted distribution of traffic flow. Such thermodynamic diagrams may help understand traffic congestion conditions and changes in flow.

In a thermodynamic diagram, different colors represent different traffic densities or traffic speeds. For example, red may represent high traffic density or low speed (i.e., congested areas), while green may represent low traffic density or high speed (i.e., clear areas). The shade of the color also reflects the severity of the traffic situation.

For example, an intelligent transportation system is being used to manage traffic for a city. First, real-time traffic data, including vehicle position, speed, direction, etc., needs to be collected by various sensors and cameras. These data may then be input into a traffic state prediction risk processing network to arrive at a traffic state prediction for each intersection or road segment.

Next, a traffic state prediction thermodynamic diagram may be generated. On this thermodynamic diagram, it is clear which intersections or road segments may be congested and which intersections or road segments are relatively clear. For example, if a certain intersection is red in color, it is indicated that the intersection may be congested; if the color of a certain road section is green, the traffic of the road section is clear.

By such traffic state prediction thermodynamic diagrams, measures such as adjusting the time setting of the signal lights or reminding drivers to avoid congested road sections can be taken in advance, thereby effectively improving traffic conditions.

Step 250, improving network parameters of the traffic state prediction risk processing network based on the first network debugging cost function and the second network debugging cost function.

According to the first and second network debugging cost functions, parameters of the traffic state prediction risk processing network can be improved so as to be more accurate in future prediction. For example, an optimization algorithm (e.g., gradient descent) may be used to adjust network parameters based on cost functions to minimize the difference between the predicted result and the real situation.

According to the technical scheme, the 3D visual traffic image information of a plurality of known results is firstly obtained, and the traffic state prediction risk processing network is trained and optimized. After repeated training, the network can more accurately predict future traffic states, thereby helping city managers to take measures in advance and improving traffic conditions. In detail, a risk processing network for traffic state prediction can be effectively constructed and optimized by acquiring 3D visual traffic image information, and combining with a space-time state description of intelligent traffic participants and further analyzing debug examples of the predicted risk processing network. After repeated training and optimization, the network can more accurately and efficiently predict future traffic states, improves the efficiency and accuracy of urban traffic management, and provides powerful technical support for solving urban traffic jam problems.

In yet other alternative embodiments, the method further includes step 310 before obtaining a spatiotemporal state description vector debug example resulting from spatiotemporal state description mining of the intelligent traffic participant debug example by the spatiotemporal state description mining branch described in step 230.

Step 310, masking the traffic participation units or traffic road sections in the 3D visual traffic image information debugging example.

In this aspect, the masking process of step 310 is a technique commonly used for computer vision and image processing. The goal is to mask out certain parts that do not need to be of interest (e.g. background or extraneous objects) in order to focus on the object of real interest.

The following are explanations and specific examples of the respective nouns:

masking processing: prior to acquiring the spatiotemporal state description vector, the traffic participation units or traffic segments in the 3D visual traffic image information debug example are masked.

For example, assume that a 3D visual image of a traffic intersection is being processed. In this image, a lot of information, which is not of interest, such as sidewalks, trees, buildings, etc., may be contained. In this case, attention can be focused on traffic participation units such as vehicles and pedestrians by creating a mask to keep only those traffic participation units.

3D visual traffic image information debug example: this refers to 3D visual traffic image information for commissioning a traffic state prediction risk processing network. It may include 3D visual images at multiple points in time, each image containing a number of traffic participating units or traffic segments.

Traffic participation unit: this refers to the major participants in traffic systems such as cars, trucks, motorcycles, bicycles, pedestrians, and the like.

Traffic segment: this means a part of the road network, which may be an intersection or a section of road between two intersections.

By introducing the masking process, attention can be more effectively focused on critical traffic participation units and traffic segments. Thus, the interference of irrelevant information on the prediction of the model can be reduced, and the prediction accuracy of the model is improved. Meanwhile, by using the time-space state description mining branch and the traffic state prediction branch, a detailed traffic state prediction result can be obtained. After repeated training, the network can more accurately predict future traffic states, thereby helping city managers to take measures in advance and improving traffic conditions.

In some possible embodiments, the step 310 of masking the traffic participation units or the traffic segments in the 3D visual traffic image information debug example includes steps 311-313.

Step 311, detecting a target traffic participation unit related to the traffic state prediction risk from the 3D visual traffic image information debugging example.

Step 312, masking the target traffic participant unit in the 3D visual traffic image information commissioning example with a first set probability value.

Step 313, masking the intelligent traffic participation object commissioning example in the 3D visual traffic image information commissioning example by a second set probability value.

In step 310, the traffic participation units or traffic segments in the 3D visual traffic image information debug example are subjected to a masking process in order to concentrate the points of interest on the targets related to the prediction risk, thereby improving the accuracy and efficiency of the prediction.

A target traffic participant associated with a traffic state prediction risk is detected. In this step, a "target traffic participant" may be understood as an element that may have a significant impact on the traffic conditions, such as an automobile traveling at high speed, or a pedestrian traveling across a road.

For example, all cars and pedestrians can be found from the 3D visual traffic image by image recognition techniques. Then, by analyzing their movement states and positions, those "target traffic participant units" that may cause traffic congestion or accidents are determined.

Masking of the target traffic participation unit is performed in the 3D visual traffic image information commissioning example by the first set probability value. The purpose of this step is to reduce interference of other non-target traffic participant units with the predicted outcome.

For example, a threshold (i.e., a first set probability value) may be set, and a certain traffic participant unit may be considered a target traffic participant unit only if its risk prediction value exceeds the threshold. The portions of the other non-target traffic participant units in the image are then masked, i.e., their pixel values are set to 0, so that only the target traffic participant units are of interest in subsequent predictions.

Masking of the intelligent traffic participation object debugging examples is performed in the 3D visual traffic image information debugging examples by the second set probability value. This step is to further reduce the influence of extraneous information on the outcome of the prediction.

For example, another threshold (i.e., a second set probability value) may be set, which is retained in the image only when the importance of a certain intelligent traffic participant exceeds this threshold. Then, the portions of other low-importance intelligent traffic participants in the image are masked, i.e., their pixel values are set to 0.

The design thought can effectively screen out the target traffic participation units related to the traffic state prediction risk, and reduces the interference of other irrelevant information through mask processing, thereby improving the accuracy and efficiency of traffic state prediction. This approach is particularly useful in complex urban traffic environments because it can help quickly find key factors that can lead to traffic problems and take corresponding action.

In some examples, the first network commissioning cost function described in step 230 that determines two intelligent traffic participant commissioning examples based on the spatiotemporal state description vector commissioning examples and the a priori traffic state perspectives includes steps 231-232.

Step 231, determining a space-time state description commonality score of two intelligent traffic participation object debugging examples based on the space-time state description vector debugging examples.

Step 232, determining a first network debug cost function of the two intelligent traffic participant debug examples based on the spatiotemporal state descriptive commonality score and the prior traffic state perspective.

In this solution, steps 231 and 232 further optimize the process of network commissioning. The following are explanations and specific examples of the respective nouns.

Wherein step 231 determines a spatiotemporal state description commonality score for two intelligent traffic participant debugging examples based on the spatiotemporal state description vector debugging examples.

For example, assume that there are two vehicles whose spatiotemporal state describe vectors, including their position, speed, and direction of travel. The similarity of the two vectors can be calculated to obtain a commonality score. If this score is high, it indicates that the driving conditions of the two vehicles are very similar; and conversely, the running states of the two wheels are greatly different.

Further, step 232 determines a first network debug cost function for two intelligent traffic participant debug examples based on the spatiotemporal state descriptive commonality score and the a priori traffic state perspectives.

For example, the gap between the commonality score and the a priori traffic state perspective (i.e., the real case) may be compared as a first network debug cost function. If the difference is small, the prediction result of the network is very close to the real situation, otherwise, the prediction result of the network is greatly different from the real situation.

By introducing a spatiotemporal state descriptive commonality score, it is possible to more effectively measure whether the travel states of two traffic participant objects are similar. Meanwhile, a more accurate network debugging cost function can be determined based on the commonality score and the prior traffic state viewpoint. In this way, the traffic state prediction risk processing network can be trained and optimized more effectively, so that the prediction accuracy of the traffic state prediction risk processing network is improved. After repeated training, the network can more accurately predict future traffic states, thereby helping city managers to take measures in advance and improving traffic conditions.

In some examples, the first network debug cost function has a first quantitative relationship with a spatiotemporal state descriptive commonality score of the two intelligent traffic participant debug examples when the a priori traffic state perspective is a non-congestion associated state label. And when the prior traffic state viewpoint is a congestion aggravated state label or a congestion slowed state label, the first network debugging cost function and the space-time state description commonality scores of the two intelligent traffic participation object debugging examples have a second quantization relation.

In this technical solution, the prior traffic state view is a preset judgment or understanding of the traffic condition, such as a non-congestion related state, a congestion aggravated state, a congestion slowed down state, and the like. The first network debugging cost function is a quantization standard for evaluating the difference between the model prediction result and the actual situation.

The non-congestion related status label refers to a condition that traffic conditions are smooth and congestion does not occur. For example, on a morning day, if the number of vehicles on a primary road is normal and the speed remains within the normal range, then this condition may be defined as a non-congestion associated state.

The congestion acceleration state flag refers to a case where traffic congestion is deteriorating. For example, during a late peak period, if the number of vehicles on a primary road increases sharply and the average speed of the vehicles decreases significantly, this condition may be defined as a congestion-aggravated condition.

The congestion reduction status label refers to a situation in which traffic congestion is being reduced. For example, if the number of vehicles on a main road starts to gradually decrease after a late peak period, and the average speed of the vehicles starts to gradually recover, this situation may be defined as a congestion slowing state.

In this technical solution, according to the difference of the prior traffic state viewpoints, the first network debug cost function and the space-time state description commonality score (possibly a measure reflecting the similarity or association of two objects in space-time) of two intelligent traffic participation object debug examples have different quantization relations. That is, the model considers the current traffic condition in the process of prediction and optimization, and adjusts the processing mode and optimization strategy for different traffic conditions accordingly.

The method and the device can dynamically adjust the optimization strategy of the model according to the actual traffic condition, so that the prediction result of the model is more in line with the actual condition, and the accuracy and the efficiency of traffic condition prediction are improved.

In other examples, when the prior traffic state perspective is a non-congestion associated state label, the spatiotemporal state descriptive commonality score of the two intelligent traffic participation object debugging examples has a second quantized relationship with a contribution weight value of the first network debugging cost function and a path spatial distribution difference of the two intelligent traffic participation object debugging examples; and when the prior traffic state viewpoint is a congestion acceleration state label or a congestion slowing state label, the space-time state description commonality score of the two intelligent traffic participation object debugging examples has a first quantized relationship with the contribution weight value of the first network debugging cost function and the path space distribution difference of the two intelligent traffic participation object debugging examples.

In the above example, the a priori traffic status perspective: this is a known traffic state label for training a predictive model. Here, it may be a "non-congestion associated state", "congestion acceleration state", or "congestion relief state".

The spatiotemporal state descriptive commonality score is a similarity score calculated based on the spatiotemporal state descriptive vectors of the two intelligent traffic participant debugging examples.

The first network debugging cost function is a cost function for network debugging, which is determined based on the space-time state description commonality score and the prior traffic state viewpoint.

The path spatial distribution difference refers to a spatial distribution difference of two intelligent traffic participant debugging examples. For example, if two automobile commissioning examples are both traveling on the same road, then their path spatial distribution varies little; conversely, if they travel on different roads, their path spatial distributions are greatly different.

In this technical solution, according to the difference of the prior traffic state views, the contribution weight value of the space-time state description commonality score to the first network debugging cost function has different quantization relations. Such relationships may be described by some mathematical formula or machine learning model, such as linear functions, logistic regression, etc.

For example, when the first-check traffic state view is "non-congestion association state", if the path spatial distribution of two intelligent traffic participants is greatly different, the weight value of the spatiotemporal state description commonality score may be lowered; conversely, if their path spatial distribution varies little, this weight value may be increased. The opposite strategy may be adopted when the a priori traffic status views are "congestion acceleration status" or "congestion mitigation status".

By the method, the time-space state description commonality score can be utilized more effectively, so that the first network debugging cost function is further optimized, and the accuracy and stability of the prediction model are improved. For urban traffic management departments, the method can help the urban traffic management departments to more accurately predict future traffic conditions and take measures in advance, so that the urban traffic problem is effectively solved.

In some possible embodiments, the traffic state prediction branch includes a cascade of feature mining nodes, feature mapping nodes, and feature prediction nodes. Based on this, the obtaining in step 240, through the traffic state prediction branch, a traffic state prediction thermodynamic diagram obtained by performing traffic state prediction analysis on the two intelligent traffic participation object debug examples based on the spatiotemporal state description vector debug example, includes steps 241-243.

Step 241, obtaining initial space-time state description features obtained by feature mining of the space-time state description vector debugging examples of the two intelligent traffic participation object debugging examples through the feature mining node.

And step 242, acquiring a space-time state numerical value mapping feature obtained by performing interval numerical value mapping on the initial space-time state description feature through the feature mapping node.

And 243, obtaining a traffic state prediction thermodynamic diagram obtained by carrying out traffic state discrimination operation on the space-time state numerical value mapping characteristics through the characteristic prediction nodes.

In this technical solution, the traffic state prediction branch includes a feature mining node, a feature mapping node, and a feature prediction node in cascade. Their roles are as follows, respectively.

Feature mining node: the node is mainly responsible for feature extraction of the input data, namely, converting an original space-time state description vector debugging example into features which can be understood and processed by a model. For example, if the two intelligent traffic participants are vehicles, feature mining may include information about the speed, location, direction of travel, etc. of the vehicles.

Feature mapping node: the node is responsible for performing interval numerical mapping on the initial space-time state description characteristics acquired by the characteristic mining node, namely converting the characteristic values into a unified standardized interval (such as 0-1) so as to facilitate subsequent calculation and comparison.

Feature prediction node: the node is responsible for carrying out traffic state discrimination operation based on the numerical characteristics after the characteristic mapping, and generating a traffic state prediction thermodynamic diagram. This thermodynamic diagram may visually display predicted traffic conditions, such as which places may be congested, etc.

For example, two vehicles (intelligent traffic participants) are traveling on a road, and the feature mining nodes first extract their speed, location, etc. The feature map node then normalizes this information to a uniform numerical interval. Finally, the feature prediction node performs traffic state discrimination operation based on the numerical features to generate a thermodynamic diagram for displaying the predicted traffic state.

By the design, complex space-time data can be effectively processed through the steps of feature mining, feature mapping, feature prediction and the like, and an intuitive traffic state prediction result can be generated. This is very valuable information to both traffic authorities and drivers, helping them make decisions ahead of time, avoiding possible traffic jams.

In some embodiments, the improving of the network parameters of the traffic state prediction risk processing network based on the first network commissioning cost function and the second network commissioning cost function in step 250 includes steps 251-253.

Step 251, integrating the first network debugging cost function and the second network debugging cost function according to the set confidence level to obtain a global network debugging cost variable of the traffic state prediction risk processing network.

And step 252, feeding back the global network debugging cost variable in the traffic state prediction risk processing network to obtain the loss function change of each network parameter.

Step 253, improving the network parameters based on the loss function variation.

In the above embodiments, the first network debugging cost function and the second network debugging cost function refer to mathematical expressions for evaluating and optimizing the performance of the neural network. These cost functions may calculate an error or "cost" of the network by comparing the predicted value to the actual value. In the context of this, the first network and the second network may represent a traffic state prediction model and a risk processing model.

The global network debugging cost variable is a variable obtained by integrating cost functions of the first network and the second network and according to the set confidence level. It represents the overall optimization objective for the entire traffic state prediction risk processing network.

The loss function is a function that measures the prediction error of the model, and the change of the loss function represents the change of the prediction error of the model in the training process. Network parameters: these are elements of the neural network that can be learned, including weights and biases, etc., and can be adjusted by an optimization algorithm such as gradient descent to minimize the loss function.

For example, assume that there is one traffic state prediction model (first network) and one risk processing model (second network), each model having its own cost function. A higher confidence level may be given to the predictive model because it is the basis for the overall system. Then, the two cost functions are combined together to obtain a global network debugging cost variable.

Then, a back propagation algorithm is used to feed back the global network debugging cost variables, so that the loss function change of each network parameter is calculated. And finally, according to the change of the loss function, improving network parameters by using optimization methods such as gradient descent and the like, so that the performance of the whole traffic state prediction risk processing network is improved.

Therefore, the performance of the traffic state prediction risk processing network is effectively improved by integrating cost functions of a plurality of networks and adjusting network parameters by using a back propagation and optimization algorithm. Specifically, the network can learn and predict the traffic state better by gradually reducing the loss function of the network parameters, so that necessary risk treatment measures are adopted in advance, and the road traffic safety is improved.

In some independent embodiments, after performing traffic state prediction analysis on the two intelligent traffic participation objects based on the first spatiotemporal state description vector and the second spatiotemporal state description vector described in step 140 to obtain traffic state prediction risks of the two intelligent traffic participation objects, the method further includes steps 151-156.

Step 151, acquiring a linked object track data stream set according to the traffic state prediction risks of the two intelligent traffic participation objects, wherein the linked object track data stream set comprises uninterrupted P linked object track data streams, and P is an integer greater than or equal to 1.

Step 152, obtaining a path conflict track data stream set according to the linkage object track data stream set, wherein the path conflict track data stream set comprises uninterrupted P path conflict track data streams.

Step 153, acquiring a linkage object track space-time element matrix set through a first space-time element identification component included in a traffic scheduling strategy generation network based on the linkage object track data flow set, wherein the linkage object track space-time element matrix set comprises P linkage object track space-time element matrices.

And 154, based on the path conflict track data flow set, acquiring a path conflict track space-time element matrix set through a second space-time element identification component included in the traffic scheduling strategy generation network, wherein the path conflict track space-time element matrix set comprises P path conflict track space-time element matrices.

Step 155, based on the linkage object track space-time element matrix set and the path conflict track space-time element matrix set, acquiring a traffic scheduling mode variable corresponding to the linkage object track data stream set through a traffic scheduling decision component included in the traffic scheduling policy generation network.

And 156, determining the traffic scheduling strategy of the linkage object track data flow set according to the traffic scheduling mode variable.

In this technical solution, steps 151 to 156 mainly relate to nouns such as a linkage object track data stream set, a path conflict track data stream set, a linkage object track space-time element matrix set, a path conflict track space-time element matrix set, and a traffic scheduling mode variable.

Linkage object trajectory data stream set: a set of continuous trajectory data streams obtained from the predicted risk of traffic conditions of intelligent traffic participants. The linked object may be any entity that participates in the traffic system and can affect the traffic state, such as a vehicle, a pedestrian, etc. Each trajectory data stream represents the path of movement of one linked object over a particular period of time.

Path collision trace data stream set: and finding out paths which are likely to collide, namely paths with traffic risks, based on the linked object track data stream set.

Linkage object track space-time element matrix set: based on the linked object trajectory data stream set, a set of space-time element matrices are obtained by the first space-time element recognition component. These matrices reflect the motion of the individual linked objects in time and space.

Path collision trajectory space-time element matrix set: and a set of space-time element matrixes acquired by the second space-time element recognition component based on the path conflict track data stream set. These matrices reflect the motion of paths in time and space where collisions may exist.

Traffic scheduling mode variable: and generating one or more variables acquired by a traffic scheduling decision component of the network through a traffic scheduling strategy based on the linkage object track space-time element matrix set and the path conflict track space-time element matrix set. These variables are used to represent and determine the mode or strategy of traffic scheduling.

For example, in a smart city transportation system, step 151 involves collecting trajectory data streams of all traffic-involved entities such as vehicles, pedestrians, etc. Step 152 involves finding those trajectories where traffic conflicts may occur, such as when two vehicles are about to meet at the same location. Next, steps 153 and 154 will generate corresponding sets of spatio-temporal element matrices based on the two types of trajectory data streams, respectively. Finally, step 155 and step 156 then generate a network according to the traffic scheduling policy, and acquire and determine the traffic scheduling mode or policy.

Thus, traffic risk can be effectively predicted, complex track data flows can be acquired and processed, and an appropriate traffic scheduling strategy can be generated according to the data. This helps to improve the safety and efficiency of the traffic system, enabling more intelligent and automated traffic management.

In some possible embodiments, based on the linkage object track space-time element matrix set and the path conflict track space-time element matrix set in step 155, obtaining, by a traffic scheduling decision component included in the traffic scheduling policy generating network, a traffic scheduling mode variable corresponding to the linkage object track data flow set, including: based on the linkage object track space-time element matrix set, generating first road port node attention branches included in a network through the traffic scheduling strategy to obtain P first space-time element relation networks, wherein each first space-time element relation network corresponds to one linkage object track space-time element matrix; based on the path conflict track space-time element matrix set, generating second intersection node attention branches included in the network through the traffic scheduling strategy to obtain P second space-time element relation networks, wherein each second space-time element relation network corresponds to one path conflict track space-time element matrix; combining the P first space-time element relation networks and the P second space-time element relation networks to obtain P target space-time element relation networks, wherein each target space-time element relation network comprises a first space-time element relation network and a second space-time element relation network; and acquiring traffic scheduling mode variables corresponding to the linked object track data flow set through the traffic scheduling decision component included in the traffic scheduling policy generation network based on the P target space-time element relation networks.

In the above embodiment, the linking object trajectory space-time element matrix set and the path collision trajectory space-time element matrix set: the two sets refer to data structures that describe and convey the state of motion of traffic participants (e.g., vehicles, pedestrians, etc.) in time and space. The link object track space-time element matrix comprises the motion information of each traffic participant, and the path conflict track space-time element matrix specifically describes the situation possibly causing traffic conflict.

The first intersection node pays attention to the branch and the second intersection node pays attention to the branch: the two branches refer to modules for processing different types of track data (such as linkage object track and path conflict track) in the traffic scheduling strategy generation network.

The space-time element relation network is a network structure describing the relation between space-time elements, such as describing the relative position and speed relation between vehicles.

The traffic scheduling mode variable is a decision variable obtained according to the space-time element relation network and used for guiding traffic scheduling.

For example, having multiple linked objects (such as vehicles) and their possible path conflict trajectories, a corresponding space-time element matrix is first generated based on this information. And then, respectively processing the space-time element matrixes through the first intersection node attention branch and the second intersection node attention branch to obtain a corresponding space-time element relation network.

And then, combining the obtained first time-space element relation network and the second time-space element relation network to obtain a target time-space element relation network. And finally, according to the target space-time element relation networks, acquiring corresponding traffic scheduling mode variables through a traffic scheduling decision component, wherein the variables can be used for guiding traffic scheduling.

Thus, by describing and handling the movement states of traffic participants and possible conflict situations in detail, decision variables are generated that can effectively guide traffic scheduling. The method not only can help to improve the traffic efficiency of the road, but also can effectively avoid traffic collision, thereby improving the safety of road traffic.

Further, there is also provided a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the above-described method.

In the embodiments provided in the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus and method embodiments described above are merely illustrative, for example, flow diagrams and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

In addition, functional modules in the embodiments of the present invention may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.

The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a network device, or the like) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for processing intelligent traffic big data based on 3D visual intelligence, which is characterized by being applied to an intelligent traffic big data processing system, the method comprising:

Based on the first space-time state description vector and the second space-time state description vector, carrying out traffic state prediction analysis on two intelligent traffic participation objects to obtain traffic state prediction risks of the two intelligent traffic participation objects;

wherein, based on the first space-time state description vector and the second space-time state description vector, performing traffic state prediction analysis on two intelligent traffic participation objects to obtain traffic state prediction risks of the two intelligent traffic participation objects, including:

Determining traffic state prediction risks of the two intelligent traffic participation objects based on the maximum discrimination coefficient of the traffic state discrimination coefficients;

when the traffic state prediction risks of the two intelligent traffic participation objects are non-congestion association state labels, the maximum discrimination coefficient of the traffic state discrimination coefficients and the space-time state description commonality scores of the two intelligent traffic participation objects have a first quantization relationship; when the traffic state prediction risks of the two intelligent traffic participation objects are congestion aggravated state labels or congestion slowing state labels, the maximum discrimination coefficient of the traffic state discrimination coefficients and the space-time state description commonality scores of the two intelligent traffic participation objects have a second quantization relationship;

when the traffic state prediction risks of the two intelligent traffic participation objects are non-congestion association state labels, the space-time state description commonality scores of the two intelligent traffic participation objects have a second quantization relationship with the contribution weight values of the traffic state discrimination coefficients and the path space distribution differences of the two intelligent traffic participation objects; when the traffic state prediction risks of the two intelligent traffic participation objects are congestion aggravation state labels or congestion slowing state labels, the space-time state description commonality scores of the two intelligent traffic participation objects have a first quantization relation with the contribution weight values of the traffic state discrimination coefficients and the path space distribution differences of the two intelligent traffic participation objects;

Before acquiring the 3D visual traffic image information to be subjected to traffic state prediction risk determination, the method further comprises the following steps:

2. The 3D visual intelligence based intelligent traffic big data processing method according to claim 1, wherein before acquiring a space-time state description vector debug example obtained by space-time state description mining of the intelligent traffic participation object debug example by the space-time state description mining branch, the method further comprises:

3. The intelligent traffic big data processing method based on 3D visual intelligence according to claim 2, wherein masking traffic participation units or traffic segments in the 3D visual traffic image information debug example comprises:

4. The 3D visual intelligence based intelligent traffic big data processing method of claim 1, wherein determining a first network debug cost function of two intelligent traffic participant debug examples based on the spatiotemporal state description vector debug examples and the a priori traffic state perspectives comprises:

determining a first network debug cost function of the two intelligent traffic participant debug examples based on the spatiotemporal state descriptive commonality score and the prior traffic state perspective;

when the prior traffic state view is a non-congestion association state label, the first network debugging cost function and the space-time state description commonality scores of the two intelligent traffic participation object debugging examples have a first quantization relation; when the prior traffic state view is a congestion aggravating state label or a congestion slowing state label, the first network debugging cost function and the space-time state description commonality scores of the two intelligent traffic participation object debugging examples have a second quantization relation;

When the prior traffic state view is a non-congestion association state label, the space-time state description commonality score of the two intelligent traffic participation object debugging examples has a second quantization relation with the contribution weight value of the first network debugging cost function and the path space distribution difference of the two intelligent traffic participation object debugging examples; and when the prior traffic state viewpoint is a congestion acceleration state label or a congestion slowing state label, the space-time state description commonality score of the two intelligent traffic participation object debugging examples has a first quantized relationship with the contribution weight value of the first network debugging cost function and the path space distribution difference of the two intelligent traffic participation object debugging examples.

5. The intelligent traffic big data processing method based on 3D visual intelligence according to claim 1, wherein the traffic state prediction branch comprises a feature mining node, a feature mapping node and a feature prediction node which are cascaded;

6. The 3D visual intelligence based intelligent traffic big data processing method according to claim 1, wherein improving network parameters of the traffic state prediction risk processing network based on the first network debugging cost function and the second network debugging cost function comprises:

The network parameter is modified based on the change in the loss function.

7. An intelligent traffic big data processing system is characterized by comprising a processor and a memory; the processor is communicatively connected to the memory, the processor being configured to read a computer program from the memory and execute the computer program to implement the method of any of claims 1-6.

8. A computer readable storage medium, characterized in that a program is stored thereon, which program, when being executed by a processor, implements the method of any of claims 1-6.