Scenario-based risk prediction methods and related devices that integrate individual risks of traffic participants
By acquiring trajectory data and skeleton key points of traffic participants, and using graph kernel method and time series classification model to construct scene graph structure, the problem of insufficient prediction of future risks in complex traffic environments by traditional methods is solved, and accurate prediction of future scene risks and safety decision support are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING INST OF TECH
- Filing Date
- 2026-04-21
- Publication Date
- 2026-07-14
AI Technical Summary
Existing technologies struggle to accurately predict future scenario risks in complex traffic environments with multi-agent interactions. Traditional methods rely on strong assumptions and have weak generalization capabilities, while deep learning methods suffer from inconsistent risk labels and scarce data, resulting in insufficient characterization of future risk evolution.
By acquiring trajectory data of various types of traffic participants and the skeleton key point sequence of vulnerable traffic participants, we use graph kernel method and temporal classification model to determine behavioral intentions and states, construct scene graph structure and extract spatiotemporal joint network features, and integrate spatial interaction features and temporal evolution features for risk prediction.
It enables accurate prediction of risks in future H-frame scenarios, provides a more forward-looking basis for safety decisions, and improves the safety and explainability of autonomous driving systems.
Smart Images
Figure CN122390465A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of autonomous driving, and in particular to a method and device for predicting scenario risks by integrating the individual risks of traffic participants. Background Technology
[0002] As autonomous driving evolves from L2 / L2+ to L3 / L4, vehicles need to make higher-level safety decisions in complex traffic environments. However, real-world road scenarios are often characterized by complex multi-agent interactions, high uncertainty, and frequent long-tail events, such as cutting in, sudden braking, intersection yield conflicts, and potential hazards caused by obstructions. These risks typically do not arise instantaneously at the current moment but accumulate gradually through the movement and interaction of multiple traffic participants, culminating in a concentrated outbreak at some point in the future. Therefore, relying solely on current risk assessments is insufficient to meet the needs of autonomous driving systems for early warning and proactive avoidance. Scenario risk prediction aims to predict the changing trend of risk levels over a future period based on historical trajectory and other observational information, providing more forward-looking safety constraints and cost basis for planning decisions. It also supports closed-loop simulation evaluation and safety verification, improving the system's safety and interpretability in complex scenarios.
[0003] Existing research, methods based on rule-based indicators such as TTC (Total Traffic Conversion) have the advantages of computational simplicity and strong interpretability, but they rely on strong assumptions such as uniform linear motion, making it difficult to cover multi-agent interactions. Data-driven deep learning methods can learn richer behavioral patterns and interaction relationships, but issues such as inconsistent risk label definitions, data scarcity, and distribution bias limit their generalization ability and stability. Furthermore, some methods still focus on single-vehicle risk or current-moment judgment, lacking characterization of future risk evolution. In addition, even with the introduction of graph networks or attention mechanisms for interaction modeling, challenges remain, such as static graph structures, insufficient dynamic expression of interactions, and the dilution of key high-risk factors by scene-level aggregation, resulting in insufficient characterization of scenario risks triggered by a few key participants. Summary of the Invention
[0004] The purpose of this application is to provide a scenario risk prediction method and related device that integrates the individual risks of traffic participants. By performing differentiated behavioral analysis and individual risk characterization on different types of traffic participants, and performing spatiotemporal interaction modeling of multiple traffic participants at the scenario level, the method can achieve the prediction output of future H-frame scenario risks.
[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a scenario risk prediction method that integrates individual risks of traffic participants, including: Acquire trajectory data of various types of traffic participants in the current scene, as well as the skeleton key point sequence of vulnerable traffic participants. For vulnerable traffic participants, based on the skeleton key point sequence and trajectory data, the crossing intention behavior of the vulnerable traffic participants is determined using a graph kernel method and a classifier. For non-vulnerable traffic participants, the relative motion state behavior of non-vulnerable traffic participants is determined based on trajectory data and a time-series classification model. Construct a scene graph structure; the scene graph structure uses each traffic participant as a node, the spatial distance between each traffic participant as an edge, and the historical trajectory data of each node and the crossing intention behavior of vulnerable traffic participants as node features; Based on a spatiotemporal joint network, feature extraction is performed on the scene graph structure; the features include spatial interaction features and temporal evolution features. Based on spatial interaction characteristics and temporal evolution characteristics, the scenario risk prediction results are determined.
[0006] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the scenario risk prediction method that integrates individual risks of traffic participants as described above.
[0007] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the scenario risk prediction method that integrates individual risks of traffic participants as described above.
[0008] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the scenario risk prediction method that integrates individual risks of traffic participants as described above.
[0009] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a scenario risk prediction method and related apparatus that integrates individual risks of traffic participants. By acquiring trajectory data of various types of traffic participants in the current scenario and the skeletal key point sequences of vulnerable traffic participants, it solves the problem that traditional methods relying solely on trajectory data cannot accurately capture the behavioral intentions of vulnerable traffic participants. By using a graph kernel method and classifier to determine the crossing intention behavior of vulnerable traffic participants, and by using a temporal classification model to determine the relative motion state behavior of non-vulnerable traffic participants, it solves the problem of large differences in behavioral patterns among different types of traffic participants and the difficulty in uniformly modeling them, thus achieving a differentiated and refined representation of individual risks of traffic participants. By constructing a scenario graph structure with traffic participants as nodes, spatial distance as edges, and historical trajectories and crossing intentions as node features, and extracting spatial interaction features and temporal evolution features based on a spatiotemporal joint network, it solves the problems of insufficient dynamic expression of scenario interactions and the dilution of key high-risk factors in existing methods. By integrating spatial interaction features with temporal evolution features and inputting them into a risk assessment model to determine the risk level of a scenario, this approach addresses the problems of traditional rule-based methods having strong assumptions and weak generalization ability, as well as the inconsistencies in risk labels and insufficient characterization of future risk evolution in some deep learning methods. It achieves accurate prediction output of future H-frame scenario risks, providing a more forward-looking safety decision-making basis for early warning and proactive avoidance in autonomous driving systems. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is an application environment diagram of a scenario risk prediction method that integrates individual risks of traffic participants, as described in one embodiment of this application. Figure 2 A flowchart illustrating a scenario risk prediction method that integrates individual risks of traffic participants, as provided in an embodiment of this application; Figure 3 This is a schematic diagram of a vulnerable traffic participant diagram provided in an embodiment of this application; Figure 4 A schematic diagram illustrating an implementation method for a scenario risk prediction method that integrates individual risks of traffic participants, as provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0013] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0014] The scenario risk prediction method that integrates individual risks of traffic participants provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send trajectory data of various types of traffic participants in the current scene, as well as the skeleton key point sequence of vulnerable traffic participants, to server 104. Upon receiving the trajectory data and skeleton key point sequence of vulnerable traffic participants, server 104, for each type of traffic participant and the skeleton key point sequence of vulnerable traffic participants, determines the travel intention behavior of the vulnerable traffic participants based on the skeleton key point sequence and trajectory data, using a graph kernel method and classifier. For non-vulnerable traffic participants, it determines their relative motion state behavior based on the trajectory data and a temporal classification model. A scene graph structure is constructed; features are extracted from the scene graph structure based on a spatiotemporal joint network; these features include spatial interaction features and temporal evolution features. Based on the spatial interaction features and temporal evolution features, a scene risk prediction result is determined. Server 104 can then feed back the obtained scene risk prediction result to terminal 102. Furthermore, in some embodiments, the scenario risk prediction method that integrates individual traffic participant risks can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly perform scenario risk prediction that integrates individual traffic participant risks based on the trajectory data of various types of traffic participants in the current scenario and the skeleton key point sequence of vulnerable traffic participants. Alternatively, the server 104 can obtain the trajectory data of various types of traffic participants in the current scenario and the skeleton key point sequence of vulnerable traffic participants from the data storage system, and perform scenario risk prediction that integrates individual traffic participant risks based on the trajectory data of various types of traffic participants in the current scenario and the skeleton key point sequence of vulnerable traffic participants.
[0015] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.
[0016] In one exemplary embodiment, such as Figure 2As shown, a scenario risk prediction method integrating individual risks of traffic participants is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 206. Wherein: Step 201: Obtain trajectory data of various types of traffic participants in the current scene, as well as the skeleton key point sequence of vulnerable traffic participants. Step 202: For vulnerable traffic participants, based on the skeleton key point sequence and trajectory data, and using the graph kernel method and classifier, determine the crossing intention behavior of the vulnerable traffic participants. Step 203: For non-vulnerable traffic participants, based on trajectory data and a time-series classification model, determine the relative motion state behavior of non-vulnerable traffic participants. Step 204: Construct a scene graph structure; the scene graph structure uses each traffic participant as a node, the spatial distance between each traffic participant as an edge, and the historical trajectory data of each node and the crossing intention behavior of vulnerable traffic participants as node features. Step 205: Based on the spatiotemporal joint network, perform feature extraction on the scene graph structure; the features include spatial interaction features and temporal evolution features; Step 206: Determine the scenario risk prediction results based on spatial interaction characteristics and temporal evolution characteristics.
[0017] In one exemplary embodiment, such as Figure 4 As shown, when performing steps 201-203, the following can be done: S1. Collect trajectory data of various types of traffic participants in the current scene and perform data preprocessing to obtain trajectory data containing the coordinates, IDs and timestamps of each traffic participant in the current scene, and extract the skeleton diagram of vulnerable traffic participants from the camera image. S2, use corresponding methods to conduct behavioral analysis for different types of traffic participants; S3. The obtained traffic participant behavior is input into the scenario risk model that integrates the individual risks of traffic participants, thereby optimizing the model's ability to predict individual risks and improving the scenario risk prediction capability. Vulnerable road users refer to cyclists and pedestrians. Corresponding methods are used for behavioral analysis of different types of road users: graph kernel-based methods are used for behavioral analysis and identification of vulnerable road users, while data-driven deep learning methods are used for behavioral analysis and identification of vehicles. A scenario risk model that integrates individual road user risk can simultaneously predict both individual and scenario risk, with individual risk serving scenario risk.
[0018] In S1, the acquired trajectory data format specifically includes: the trajectories of all traffic participants within the bev range centered on the vehicle in the historical observation sequence T frames, and the spatiotemporal skeleton diagrams of vulnerable traffic participants extracted from multi-view camera images. The skeleton diagram of cyclists has three more nodes representing the riding vehicle compared to the skeleton diagram of pedestrians.
[0019] Specifically, the trajectory data includes the position coordinates, IDs, and timestamps of each traffic participant in the current scene. Multi-view camera images and point cloud information are collected using onboard sensors. Data preprocessing yields position coordinates, which include the center positions (x, y) of all vehicles, including the current vehicle, in a Cartesian coordinate system (x, y), in meters, accurate to three decimal places. IDs are used to distinguish and track different traffic participants in the scene; Arabic numerals can be used to label different traffic participants, and different participants use different ID numbers. The timestamp information refers to the time corresponding to each frame of data collected by the sensor, with the start time of data collection as the timing origin, and milliseconds as the unit. For each traffic participant, the trajectory data is represented by a time-series set of coordinate points composed of the traffic participant's position coordinates collected under the aforementioned different timestamp information. The traffic participant ID is also used to distinguish trajectory data from different traffic participants.
[0020] like Figure 3 As shown, the skeleton diagram of vulnerable traffic participants takes key human nodes, including the nose, right shoulder joint, right hip joint, right knee joint, right ankle joint, left shoulder joint, left hip joint, left knee joint, and left ankle joint. It also takes the midpoint of the non-motorized vehicle handlebar, the contact point between the front wheel and the ground, and the contact point between the rear wheel and the ground.
[0021] In S2, the behavior of vulnerable road users can be reflected by the movement of the skeleton. An SVM classifier based on a multidimensional scaling Laplacian graph kernel is used for behavior recognition, with behavior labels categorized as crossing, not crossing, and having a crossing tendency. The multidimensional scaling Laplacian graph kernel considers the structure of graphs at different scales by constructing a hierarchical structure of nested subgraphs, and compares the subgraphs using the feature space Laplacian graph kernel, transforming the kernel defined between nodes of two subgraphs into a kernel between the graphs themselves. This is applied to the input cyclist graph data. Its Laplace operator is ,in for The corresponding adjacency matrix, Let be the degree matrix. For a graph... and picture Let their node sets be respectively , Let the union of nodes be denoted as Let its regularized Laplace operator be denoted as and Then the corresponding normal distributions of the nodes are respectively and The calculation steps for the multidimensional scaling Laplacian kernel are as follows: 1) Represent each node as a component corresponding to a permutation-invariant local vertex feature. A dimensional vector, through performing a linear transformation, represents each graph as a distribution of selected features. The generalized feature space Laplacian graph kernel is defined as follows: ; In the formula, , It is the identity matrix. From the kernel matrix The maximal orthogonal eigenvector group and its corresponding eigenvalues are generated. n1 and n2 are the number of nodes in graph G1 and graph G2, respectively. Q1 and Q2 are the feature projection matrices of graph G1 and graph G2, respectively. γ is the regularization hyperparameter. S1 and S2 are the regularization covariance matrices of graph G1 and graph G2, respectively.
[0022] 2) Compare nested subgraphs using the Laplacian kernel of the generalized feature space. For Generate a nested adjacency sequence , each The generated graph The induced graph is denoted as Then the multidimensional scaling Laplace subgraph kernel Defined as: ; The multidimensional scaling Laplacian kernel between the two graphs is defined as: ; In the formula, This represents the generalized Laplacian graph kernel of the feature space, with the kernel of the (l-1)th layer subgraph as the kernel matrix. This represents the multidimensional scaling Laplacian kernel.
[0023] SVM classifies data by selecting an appropriate hyperplane. For a typical binary classification problem, given a dataset... , for 3D eigenvectors This is the output of the SVM model. For datasets with two-dimensional features, the goal of SVM is to find the decision boundary between two clusters; for datasets with high-dimensional features, the decision boundary evolves into a hyperplane, and data points falling on either side of the hyperplane can be classified into different classes. The goal of SVM is to select, through training, a hyperplane that can separate different classes as much as possible. For the cyclist spatiotemporal graph model dataset in this embodiment, a similarity metric matrix is used. Each column vector in Alternative Inside The classification model obtained through training The expression is: ; In the formula, For hyperplane The normal vector, For the intercept, both are undetermined parameters and are determined by the following optimization objective: ; In the formula, These are slack variables.
[0024] For vehicle traffic participants, a sequence of vehicle trajectory coordinates for a continuous T frames is input, and a Transformer-based temporal classification model is constructed to extract motion dynamic features and output vehicle behavior labels, which include three categories: approaching, relatively stationary, and moving away.
[0025] In S3, this embodiment proposes a scenario risk prediction method that integrates individual risks of traffic participants. The current traffic scenario is modeled as graph-structured data from a bird's-eye view (BEV) perspective to achieve explicit expression of the interaction relationships among multiple traffic participants and spatiotemporal joint modeling of scenario risks. Specifically, a local BEV range is established with the vehicle as the origin of the coordinate system. The BEV range is set to 90m in front of the vehicle, 30m behind the vehicle, and 45m to the left and right. This range is used to filter the set of traffic participants participating in scenario risk prediction, avoiding redundant calculations caused by introducing excessively distant targets and improving the real-time performance and robustness of the modeling.
[0026] In the graph structure data, each traffic participant corresponds to a node in the graph. The input features of each node include the traffic participant's historical trajectory observation sequence and the behavioral label information obtained in step S2. The historical trajectory observation sequence is 18 frames long and can reflect the movement trend and dynamic changes of traffic participants in a recent period. The behavioral labels are used to characterize the current behavioral intention or movement state of traffic participants. For example, the behavioral labels for vehicle traffic participants include approaching, relatively stationary, and moving away, while the behavioral labels for vulnerable traffic participants include crossing, not crossing, and having a tendency to cross. This enables a semantic representation of the individual risk of traffic participants and serves as important input information for scene risk prediction.
[0027] The edge connections between nodes are determined by the spatial distance between traffic participants, representing the interaction relationships between them. Specifically, when the relative distance between any two traffic participants at the same time is less than 5m, an edge is added between their corresponding nodes. These edges are undirected and unweighted, used to construct an undirected, unweighted graph structure, thus only expressing the adjacency constraint of "potential interaction relationship" and avoiding the introduction of complex edge attribute designs that could affect the stability of model training.
[0028] Based on the aforementioned graph structure input, this embodiment constructs a spatiotemporal joint modeling network with a Graph Convolutional Network (GCN) and a Long Short-Term Memory (LSTM) network at its core. The GCN aggregates neighborhood information of nodes in the graph structure to extract spatial interaction features between traffic participants; the LSTM encodes the temporal features of nodes across an 18-frame historical sequence to extract the evolution of traffic participants' motion states over time. Through spatiotemporal feature fusion, this embodiment can generate scene-level feature representations and further output scene risk prediction results for the next 5 frames, achieving forward-looking prediction of short-term risk trends.
[0029] To achieve quantitative definition and supervised training of scenario risks, this embodiment employs K-means clustering to generate scenario risk categories or risk levels. Specifically, a risk representation vector is constructed based on the vehicle's motion attributes, including vehicle speed, acceleration, lateral acceleration, rate of change of heading angle, and braking intensity. K-means clustering is used to classify samples into three risk levels: safe, dangerous, and alert. The clustering results are then used as scenario risk labels for supervised training, thus forming a scenario risk labeling system that can be used for training and evaluation. The goal of K-means is to minimize the within-cluster squared error. .
[0030] In the formula, N is the sample size. It is a 0-1 type indicator variable. It is the center of the k-th cluster.
[0031] This application also provides an application scenario in which the above-mentioned scenario risk prediction method integrating individual traffic participant risks is applied. Specifically, the scenario risk prediction method integrating individual traffic participant risks provided in this embodiment can be applied in intelligent driving assistance systems (ADAS) as the core algorithm of its environmental perception and risk assessment module. When a vehicle is driving on urban roads, highways, or in complex traffic scenarios, this method can process multi-source data collected by onboard sensors in real time, including camera images, LiDAR point clouds, etc. By extracting and analyzing the trajectory data of traffic participants (such as pedestrians, cyclists, other vehicles, etc.), it can accurately identify their behavioral intentions (such as whether pedestrians have a tendency to cross, and whether vehicles are approaching or moving away). Subsequently, a graph structure model containing individual risk characteristics of traffic participants is constructed from a bird's-eye view. Spatiotemporal joint modeling is performed using graph convolutional networks and long short-term memory networks, comprehensively considering the spatial interaction relationships and temporal evolution patterns among traffic participants, and finally outputting the scenario risk level (safe, alert, dangerous) in the next short period of time (such as the next 5 frames). This risk prediction result can provide key decision support for intelligent driving systems. For example, when a "dangerous" level is predicted, the system can trigger active braking, emergency avoidance, or issue a strong warning signal to the driver in advance, thereby effectively reducing the probability of traffic accidents and improving the safety and reliability of intelligent driving.
[0032] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 5 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores scenario risk prediction results data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a scenario risk prediction method that integrates individual risks of traffic participants.
[0033] Those skilled in the art will understand that Figure 5The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0034] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0035] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0036] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of the relevant data are carried out in compliance with the relevant data protection laws and policies of the country where the location is located, and with the authorization granted by the owner of the corresponding device.
[0037] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0038] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0039] In summary, this application has the following technical effects: 1) By introducing skeletal motion information of vulnerable traffic participants and combining it with the graph kernel method for behavior recognition, the ability to identify high-risk behaviors such as crossing has been improved; 2) By employing differentiated behavioral analysis strategies for vehicles and vulnerable traffic participants, an effective integrated expression of individual risks among various types of traffic participants can be achieved; 3) By combining graph structure modeling with spatiotemporal joint learning of GCN-LSTM, the spatial interaction relationships among multiple traffic participants and their evolution over time can be explicitly characterized, thereby improving the accuracy and robustness of scenario risk prediction. 4) This application outputs a future H-frame risk sequence, which can provide forward-looking risk constraints and safety references for autonomous driving planning and decision-making, and has high engineering application value.
[0040] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0041] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A scenario risk prediction method that integrates individual risks of traffic participants, characterized in that, include: Acquire trajectory data of various types of traffic participants in the current scene, as well as the skeleton key point sequence of vulnerable traffic participants. For vulnerable traffic participants, based on the skeleton key point sequence and trajectory data, the crossing intention behavior of the vulnerable traffic participants is determined using a graph kernel method and a classifier. For non-vulnerable traffic participants, the relative motion state behavior of non-vulnerable traffic participants is determined based on trajectory data and a time-series classification model. Construct a scene graph structure; the scene graph structure uses each traffic participant as a node, the spatial distance between each traffic participant as an edge, and the historical trajectory data of each node and the crossing intention behavior of vulnerable traffic participants as node features; Based on a spatiotemporal joint network, feature extraction is performed on the scene graph structure; the features include spatial interaction features and temporal evolution features. Based on spatial interaction characteristics and temporal evolution characteristics, the scenario risk prediction results are determined.
2. The scenario risk prediction method based on the integration of individual risks of traffic participants as described in claim 1, characterized in that, The vulnerable traffic participants include pedestrians and cyclists; the skeletal keypoint sequence of the cyclist includes pedestrian skeletal nodes and riding tool nodes.
3. The scenario risk prediction method based on the integration of individual risks of traffic participants as described in claim 2, characterized in that, The trajectory data includes the coordinates, ID, and timestamp information of the traffic participants.
4. The scenario risk prediction method based on the integration of individual risks of traffic participants as described in claim 3, characterized in that, The graph kernel method is a multidimensional scaling Laplacian graph kernel, used to calculate the structural similarity between different skeleton graphs; the classifier is a support vector machine, and the behavioral labels output by the classifier include crossing, not crossing, and having a crossing trend.
5. The scenario risk prediction method based on the integration of individual risks of traffic participants as described in claim 4, characterized in that, The temporal classification model is a Transformer-based model; when the traffic participant is a vehicle, the input of the model is a sequence of vehicle trajectory coordinates for T consecutive frames, and the output is a vehicle behavior label; the vehicle behavior label includes approaching, relatively stationary, and moving away.
6. The scenario risk prediction method based on the integration of individual risks of traffic participants as described in claim 5, characterized in that, The spatiotemporal joint network includes a graph convolutional network and a long short-term memory network; the graph convolutional network is used to aggregate node neighborhood information to extract spatial interaction features, and the long short-term memory network is used to encode the temporal evolution features of nodes within a historical time window to extract motion evolution patterns.
7. The scenario risk prediction method based on the integration of individual risks of traffic participants as described in claim 6, characterized in that, Based on spatial interaction characteristics and temporal evolution characteristics, the scenario risk prediction results are determined, specifically including: By fusing spatial interaction features with temporal evolution features, comprehensive scene features are obtained. The scene comprehensive features are input into the risk assessment model to obtain the scene risk level; the scene risk level includes safe, dangerous, and alert; the risk assessment model adopts a multilayer perceptron, whose input layer receives the fused scene comprehensive features, processes the scene comprehensive features through nonlinear transformation of the hidden layer, and outputs the probability distribution of each risk level by the output layer, and determines the scene risk prediction result based on the probability distribution.
8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement a scenario risk prediction method that integrates individual risks of traffic participants, as described in any one of claims 1-7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements a scenario risk prediction method that integrates individual risks of traffic participants, as described in any one of claims 1-7.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements a scenario risk prediction method that integrates individual risks of traffic participants, as described in any one of claims 1-7.