Expressway along the village pedestrian and non-motor vehicle risk identification method and system

Abstract

The application relates to the technical field of intelligent traffic, and particularly discloses a highway-side village pedestrian and non-motor vehicle risk identification method and system, which first perceives traffic participants through multispectral and thermal imaging data fusion and generates a state set; then constructs a dynamic interaction relationship topology, calculates a motion trend vector and an interaction intensity coefficient; then calculates an initial risk factor based on road region division, and updates a comprehensive influence factor through a risk propagation mechanism; finally, a risk group and a risk chain are identified, and graded early warning information is generated; through multi-source data fusion and dynamic topology analysis, the application solves the problem of insufficient group risk and chain risk identification caused by semantic understanding loss in the prior art, realizes the leap from single target detection to complex interaction scene risk assessment, and improves the accuracy of risk identification and the timeliness of early warning.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a method and system for identifying risks of pedestrians and non-motorized vehicles in villages along highways. Background Technology

[0002] With the expansion of highway networks, illegal crossings by pedestrians and non-motorized vehicles in villages along the routes have become a significant threat to traffic safety. Existing risk identification technologies mainly rely on video surveillance and radar detection, using target tracking and trajectory prediction to assess collision risk. However, these methods are largely based on independent analysis of single targets, treating pedestrians and vehicles as isolated moving units and calculating collision probabilities solely based on their physical motion states (such as position and speed). Although some research has attempted to introduce multi-sensor fusion technology, its core remains at the level of data complementarity, failing to fundamentally address the complex interactions and interpretation of behavioral intentions among traffic participants. This analytical paradigm based on low-dimensional physical characteristics is ill-suited to handling complex scenarios common in villages along highways, such as group crossings and coordinated violations, and is even less capable of identifying the transmission and amplification effects of risk among multiple participants.

[0003] The most prominent shortcoming of existing technologies lies in the lack of a deep understanding of the semantics of multi-agent interactions in traffic scenarios, resulting in the system's inability to identify potential risks arising from complex relationships between participants. Specifically, when multiple pedestrians or non-motorized vehicles engage in group behavior, traditional methods can only identify the trajectory of individual targets and cannot determine their collaborative intentions; when risks propagate along the participant chain, the system cannot establish causal relationships between events. This inability to anticipate implicit and systemic risks arising from interaction relationships leads to delayed or even missed warnings, failing to meet the safety and control needs of complex traffic scenarios in villages along highways. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for identifying risks of pedestrians and non-motorized vehicles in villages along highways, in order to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] Methods for identifying risks to pedestrians and non-motorized vehicles in villages along highways include the following steps:

[0007] S1: Real-time monitoring of a designated area along the highway, identifying and tracking each traffic participant within the designated area along the highway, and generating a traffic participant status set containing each traffic participant's unique identifier, real-time location, and real-time speed; the traffic participants include: village pedestrians, non-motorized vehicles, and motorized vehicles;

[0008] S2: Based on the set of traffic participant states, construct a dynamic interaction relationship topology for all traffic participants at the current moment; wherein, each traffic participant is treated as a topology node, and a topology edge is established between any two traffic participants with potential interaction behaviors; and each topology node is assigned node attributes, and each edge is assigned edge attributes.

[0009] S3: Based on the dynamic interaction relationship topology, calculate the initial risk factor of each topological node; then, according to the edge attributes of the topological edges, propagate and weight the initial risk factor of each topological node along the topological edges connected to the corresponding topological nodes in a directional manner and aggregate them to update the comprehensive influence factor of each topological node.

[0010] S4: Based on the updated comprehensive impact factors of all topological nodes, identify risk groups or risk chains that meet preset conditions; wherein, a risk group consists of a group of topological nodes that are closely connected by topological edges and whose comprehensive impact factors all exceed the first threshold; a risk chain consists of a group of topological nodes that are connected by topological edges and whose comprehensive impact factors are passed on sequentially and exceed the second threshold.

[0011] S5: In response to the identification of risk groups or risk chains, generate and issue scenario-level early warning information that matches the risk level of the risk group or risk chain.

[0012] As a further aspect of the present invention: the process for generating the traffic participant state set is as follows:

[0013] S11: Simultaneously acquire multispectral image data and thermal imaging data deployed in designated areas along the highway;

[0014] S12: Perform fusion processing on multispectral image data and thermal imaging data to generate a fused feature vector with enhanced spectral and thermal radiation features;

[0015] S13: Based on the fused feature vector, identify each individual traffic participant within a designated area along the highway and assign each individual traffic participant a unique identifier that remains unchanged throughout the entire monitoring period.

[0016] S14: Based on the identity identifier, associate the spatial coordinates and movement speed of each traffic participant in a continuous time sequence to generate a traffic participant state set.

[0017] As a further aspect of the present invention: S2 specifically includes:

[0018] S21: Based on the real-time position and real-time speed in the set of traffic participants' states, calculate a motion trend vector for each topology node. The motion trend vector is used to characterize the future instantaneous movement direction and movement intention of the corresponding traffic participant.

[0019] S22: Calculate the interaction strength coefficient based on the node attributes of the two topological nodes connected by each established topological edge;

[0020] S23: Based on the interaction strength coefficient, dynamically maintain the existence state of the topological edge; when the interaction strength coefficient is lower than the preset maintenance threshold, the corresponding topological edge is removed; when the interaction strength coefficient between two unconnected topological nodes is higher than the preset establishment threshold, a new topological edge is created between the two topological nodes.

[0021] As a further aspect of the present invention: S21 specifically includes:

[0022] S211: Extract historical location points within the most recent complete motion cycle from the continuously updated location sequence of each traffic participant to form a short-term trajectory segment;

[0023] S212: Divide the short-term trajectory segment into two equal-length sub-segments, calculate the geometric centroid of the position points contained in each sub-segment, and determine the direction vector from the centroid of the first sub-segment to the centroid of the last sub-segment.

[0024] S213: Combine the direction of the direction vector with the scalar value of the latest real-time speed of the traffic participant to generate a motion trend vector; where the direction of the direction vector defines the direction of the motion trend vector, and the scalar value of the real-time speed defines the magnitude of the motion trend vector.

[0025] As a further aspect of the present invention: the calculation of the interaction strength coefficient specifically includes:

[0026] S221: Calculate the spatial connection vector from the first topological node to the second topological node, and calculate the first angle between the spatial connection vector and the motion trend vector of the first topological node, and the second angle between the spatial connection vector and the motion trend vector of the second topological node.

[0027] S222: Obtain the real-time velocity of two topological nodes and calculate the scalar value of the relative velocity of the real-time velocity;

[0028] S223: The cosine values ​​of the first included angle and the second included angle are weighted and summed to obtain the convergence status evaluation value. The convergence status evaluation value is then multiplied by the scalar value of the relative velocity to obtain the interaction intensity coefficient.

[0029] As a further aspect of the present invention: S3 specifically includes:

[0030] S31: Determine the initial risk factor of the corresponding topological node based on the type of traffic participant and its specific location within the road area;

[0031] S32: Calculate the risk propagation weight based on the interaction strength coefficient and relative distance contained in the attributes of each topological edge, where the interaction strength coefficient is directly proportional to the risk propagation weight and the relative distance is inversely proportional to the risk propagation weight;

[0032] S33: Collect the propagation risk value of all adjacent topological nodes of each topological node through topological edges. The propagation risk value is the product of the current risk factor of the adjacent topological node and the risk propagation weight of the corresponding topological edge.

[0033] S34: Superimpose the risk factor of the topology node itself with all the propagation risk values ​​gathered, and use the superposition result to update the comprehensive impact factor of the corresponding topology node.

[0034] As a further aspect of the present invention: the process for determining the initial risk factor is as follows:

[0035] S311: Divide the highway driving lanes into multiple strip zones with increasing risk levels based on their distance from the emergency lane, and set a corresponding location risk base for each strip zone;

[0036] S312: Determine the specific strip area where the traffic participant corresponding to the topology node is currently located, and analyze the angular relationship between the motion trend vector and the current strip area boundary line;

[0037] S313: Determine the type base value based on the type of traffic participant, multiply the type base value by the location risk base of the strip area, and then multiply by the lane crossing coefficient determined based on the angle relationship to obtain the initial risk factor.

[0038] As a further aspect of the present invention: S4 specifically includes:

[0039] S41: Traverse all topology nodes and mark the topology nodes whose comprehensive impact factor exceeds the preset activation threshold as risk seed nodes;

[0040] S42: Starting from each risk seed node, trace outward along the topological edge it connects, and include all adjacent topological nodes on the tracing path whose comprehensive impact factor exceeds the first threshold into the same risk group, until a topological node with a comprehensive impact factor below the first threshold is encountered.

[0041] S43: Within the risk group, identify three or more topological nodes that are connected by continuous topological edges and whose comprehensive influence factors show an increasing trend, and construct the corresponding group of topological nodes into a risk chain; wherein, the increasing trend means that along the connection direction of the topological edge, the comprehensive influence factor of the subsequent topological node is greater than the comprehensive influence factor of the preceding topological node.

[0042] As a further aspect of the present invention: S5 specifically includes the following steps:

[0043] S51: Determine the corresponding early warning level based on the proportion of risk seed nodes in the risk group or the length of the risk chain;

[0044] S52: Based on the determined warning level, generate a set of warning instructions that includes the coordinates of the warning area, the risk type, and the risk evolution trend;

[0045] S53: According to the warning instruction set, the following warning actions shall be executed simultaneously: display the corresponding graphic warning symbol on the variable information sign upstream of the warning area, issue a focused alarm in the direction of risk transmission through the roadside directional sound and light device, and send a warning report containing a risk group topology map or risk chain path map to the monitoring center.

[0046] A risk identification system for pedestrians and non-motorized vehicles in villages along highways includes:

[0047] The traffic environment status perception module is used to monitor a designated area along the highway in real time, identify and track each traffic participant within the designated area along the highway, and generate a traffic participant status set containing the unique identifier, real-time location, and real-time speed of each traffic participant; the traffic participants include: village pedestrians, non-motorized vehicles, and motorized vehicles;

[0048] The dynamic interaction relationship construction module constructs a dynamic interaction relationship topology for all traffic participants at the current moment, based on the set of traffic participant states. Each traffic participant is treated as a topology node, and a topology edge is established between any two traffic participants with potential interaction behaviors. Each topology node is assigned node attributes, and each topology edge is assigned edge attributes.

[0049] The risk propagation and impact assessment module calculates the initial risk factor for each topological node based on the dynamic interaction relationship topology. Then, according to the edge attributes of the topological edges, the initial risk factor of each topological node is propagated and weighted along the topological edges connected to the corresponding topological nodes to update and obtain the comprehensive impact factor of each topological node.

[0050] The collaborative risk pattern recognition module identifies risk groups or risk chains that meet preset conditions based on the updated comprehensive impact factors of all topological nodes. Among them, a risk group consists of a group of topological nodes that are closely connected by topological edges and whose comprehensive impact factors all exceed the first threshold; a risk chain consists of a group of topological nodes that are connected by topological edges and whose comprehensive impact factors are passed on sequentially and exceed the second threshold.

[0051] The tiered early warning information generation and release module, in response to the identification of risk groups or risk chains, generates and issues scenario-level early warning information that matches the risk level of the risk group or risk chain.

[0052] The beneficial effects of this invention are:

[0053] (1) By fusing multispectral image data and thermal imaging data and constructing an enhanced fusion feature vector, this invention effectively overcomes the perception limitations of a single sensor under adverse conditions such as backlight, nighttime, or fog and haze. Multispectral data provides rich spectral reflectance features, while thermal imaging data ensures reliable detection of living targets. The weighted fusion of the two makes the identification of key targets such as pedestrians and non-motorized vehicles no longer dependent on single visible light information, significantly reducing the probability of missed detections and false detections. At the same time, by calculating motion trend vectors and constructing dynamic interaction relationship topology, the traditional isolated target detection is upgraded to a quantitative analysis of the interaction relationships between traffic participants, enabling the system to identify complex scenarios such as "group collaborative crossing" and "risk chain transmission," thereby achieving more accurate and stable risk perception in the complex environment along real highways and villages.

[0054] (2) This invention simulates and diffuses risks within a network of traffic participants by constructing a dynamic topology network and designing a risk propagation mechanism. Based on the directional propagation and aggregation of initial risk factors and topology edge weights, the comprehensive impact factor of each node not only reflects its own risk but also includes the risk profile of the interactive network it belongs to. This enables the system to identify potential "risk chains," that is, to identify critical paths within a group where risks are propagating and amplifying. Based on this, early warning is no longer limited to obvious dangerous behaviors that have already occurred, but can be used for early identification and tiered warning of emerging, potentially high-risk groups and propagation chains. This provides a valuable time window for intervention measures before accidents occur (such as issuing warnings in advance and adjusting warning information along the route), achieving a fundamental shift from traditional passive response to proactive, predictive prevention and control. Attached Figure Description

[0055] The invention will now be further described with reference to the accompanying drawings.

[0056] Figure 1 This is a flowchart of the method of the present invention;

[0057] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Example 1, please refer to Figure 1 As shown, the present invention provides a method for identifying pedestrian and non-motorized vehicle risks in villages along highways, comprising the following steps:

[0060] S1: Real-time monitoring of a designated area along the highway, identifying and tracking each traffic participant within the designated area along the highway, and generating a traffic participant status set containing each traffic participant's unique identifier, real-time location, and real-time speed; the traffic participants include: village pedestrians, non-motorized vehicles, and motorized vehicles;

[0061] S2: Based on the set of traffic participant states, construct a dynamic interaction relationship topology for all traffic participants at the current moment; wherein, each traffic participant is treated as a topology node, and a topology edge is established between any two traffic participants with potential interaction behaviors; and each topology node is assigned node attributes, and each edge is assigned edge attributes.

[0062] S3: Based on the dynamic interaction relationship topology, calculate the initial risk factor of each topological node; then, according to the edge attributes of the topological edges, propagate and weight the initial risk factor of each topological node along the topological edges connected to the corresponding topological nodes in a directional manner and aggregate them to update the comprehensive influence factor of each topological node.

[0063] S4: Based on the updated comprehensive impact factors of all topological nodes, identify risk groups or risk chains that meet preset conditions; wherein, a risk group consists of a group of topological nodes that are closely connected by topological edges and whose comprehensive impact factors all exceed the first threshold; a risk chain consists of a group of topological nodes that are connected by topological edges and whose comprehensive impact factors are passed on sequentially and exceed the second threshold.

[0064] S5: In response to the identification of risk groups or risk chains, generate and issue scenario-level early warning information that matches the risk level of the risk group or risk chain.

[0065] In Example 2, in S1, multispectral cameras and thermal imaging cameras deployed along a designated area of ​​the highway collect multispectral image data and thermal imaging data of the area, respectively. The multispectral cameras acquire spectral reflectance information including visible and near-infrared bands, while the thermal imaging cameras acquire thermal radiation intensity information based on the surface temperature distribution of objects.

[0066] The spectral reflectance value of each pixel in the multispectral image data and the thermal radiation intensity value of the corresponding spatial location in the thermal imaging data are normalized to ensure all values ​​fall within the range of 0 to 1. Then, a weighted average is calculated for the normalized multispectral and thermal radiation feature values, with a weighting coefficient of 0.6 for the multispectral feature values ​​and 0.4 for the thermal radiation feature values. Next, the weighted average feature values ​​are concatenated with the thermal radiation feature values ​​according to the order of the multispectral bands, forming a fused feature vector that enhances both spectral and thermal radiation characteristics. The dimension of this fused feature vector is equal to the number of multispectral bands plus one, with each dimension corresponding to the spectral reflectance or thermal radiation characteristic of a specific band.

[0067] The process of identifying individual traffic participants and assigning them identity identifiers based on fused feature vectors is implemented as follows: First, the fused feature vectors are input into a pre-trained target classifier using a support vector machine algorithm with a radial basis function kernel and a classification threshold of 0.7. When the similarity between the feature vector and the template features of a certain type of traffic participant exceeds this threshold, the participant is identified as belonging to that type of traffic participant. For each identified individual traffic participant, the system records their initial appearance feature set, including the main color distribution, contour shape features, and typical size features. Then, the system generates a 14-digit identity identifier for this individual, where the first 6 digits represent the date, the middle 4 digits represent the camera number, and the last 4 digits represent the sequence number. This identity identifier remains constant throughout the monitoring period and does not change with the target location.

[0068] The process of generating a set of traffic participant states based on identifiers includes the following calculation steps: First, for each individual traffic participant with an identifier, the system records its spatial coordinates in each processing cycle (set to 0.1 seconds). These coordinates are obtained through binocular visual ranging, specifically by calculating the disparity value of the same target in the imaging planes of two cameras. Motion speed is calculated using a difference method, that is, dividing the difference in spatial coordinates between the current frame and the previous frame (0.1-second interval) by the time interval to obtain the instantaneous velocity vector. Then, the system establishes a data structure with the identifier as the primary key, associating and storing the spatial coordinates and motion speeds in continuous temporal sequence. For each identifier, the system maintains a circular buffer of length 50 to store the motion state data within the last 5 seconds.

[0069] In S2, the process of constructing a dynamic interaction topology for all traffic participants at the current moment, based on the set of traffic participant states, includes the following steps: First, each traffic participant is treated as a topology node. The attributes of a topology node include the participant's type, real-time location coordinates, and real-time speed vector. The type of traffic participant is represented by a classification identifier: pedestrians are identified by identifier 1, non-motorized vehicles by identifier 2, and motorized vehicles by identifier 3. The real-time location coordinates are in meters and recorded using a Cartesian coordinate system, with the origin set to the lower left corner of the monitored area. The real-time speed vector is in meters per second and includes horizontal and vertical components. Second, a topology edge is established between any two traffic participants with potential interaction behaviors. The attributes of the topology edge include relative distance and relative speed. The relative distance is obtained by calculating the Euclidean distance between the two topology nodes, and the relative speed is obtained by calculating the magnitude of the difference between the speed vectors of the two topology nodes. The criteria for judging potential interaction behaviors are a relative distance of less than 50 meters and a relative speed directional angle of less than 90 degrees. Finally, the topology structure is dynamically updated, and the existence state of the topology edge is adjusted according to the real-time calculated interaction strength coefficient to ensure that the topology reflects the actual interaction of the current traffic participants.

[0070] The process of calculating the motion trend vector for each topological node based on the real-time location and speed in the traffic participant state set includes the following specific steps: First, from the continuously updated location sequence of each traffic participant, historical location points within the most recent complete motion cycle are extracted to form a short-term trajectory segment. A complete motion cycle is defined as 2 seconds, and 20 location points are collected at a sampling frequency of 10 Hz. The location sequence is stored as an array, with each location point containing an x-coordinate and a y-coordinate. Next, the short-term trajectory segment is divided into two equal-length sub-segments, each containing 10 location points. The geometric centroid of the location points in each sub-segment is calculated. The formula for calculating the x-coordinate of the centroid of the first sub-segment is: The formula for calculating the ordinate is: ;in, and It is the first segment in the first child's segment. The coordinates of each location point The x-coordinate representing the centroid of the first sub-fragment. The vertical coordinate represents the centroid of the first sub-segment. Similarly, the horizontal coordinate of the centroid of the last sub-segment is calculated using the following formula: The formula for calculating the ordinate is: Then, determine the direction vector from the centroid of the first sub-segment to the centroid of the last sub-segment, and the horizontal component of the direction vector. The calculation is as follows: Vertical component The calculation is as follows: ; Direction angle of the direction vector The calculation formula is: ; The value ranges from 0 to 360 degrees. Finally, the direction vector is synthesized with the scalar value of the latest real-time speed of the traffic participants to generate the motion trend vector. The direction of the motion trend vector is defined as the direction angle of the direction vector. The magnitude of the motion trend vector Defined as a scalar value of real-time velocity, i.e.: ;in and These are the horizontal and vertical components of the real-time velocity vector. The motion trend vector is ultimately represented in polar coordinates. .

[0071] The process of calculating the interaction strength coefficient based on the node attributes of the two topological nodes connected by each established topological edge is carried out in the following steps: First, calculate the spatial connection vector from the first topological node to the second topological node. The horizontal component of the spatial connection vector... The calculation is as follows: Vertical component The calculation is as follows: ;in These are the coordinates of the first topological node. These are the coordinates of the second topological node. The magnitude LL of the spatial connection vector is calculated as follows: Next, calculate the first angle between the spatial connection vector and the motion trend vector of the first topological node. And the second angle between the spatial connection vector and the motion trend vector of the second topological node. First included angle The calculation formula is: ;in and These are the horizontal and vertical components of the motion trend vector of the first topological node. It is the magnitude of the motion trend vector of the first topological node. The second included angle... The calculation formula is: ;in and These are the horizontal and vertical components of the motion trend vector of the second topological node. This is the magnitude of the motion trend vector of the second topological node. Then, the real-time velocities of the two topological nodes are obtained, and the scalar value of their relative velocity is calculated. Scalar value of relative velocity The calculation formula is as follows: Finally, the cosine values ​​of the first and second included angles are weighted and summed to obtain the convergence status evaluation value. Weighted summation formula: The weighting coefficients are among them. and All values ​​were set to 0.5. The convergence trend evaluation value was then... Scalar value of relative velocity Multiply to obtain the interaction strength coefficient. Interaction strength coefficient It is a unitless scalar with a range from 0 to positive infinity. The larger the value, the stronger the interaction.

[0072] The process of dynamically maintaining the existence state of topological edges based on the interaction strength coefficient includes the following specific steps: First, a topological edge maintenance threshold of 0.3 and an establishment threshold of 0.6 are set. These thresholds are derived from statistical analysis of historical traffic accident data. Specifically, regression analysis is performed on multiple interaction scenario data to determine the correspondence between risk probability and interaction strength coefficient. The maintenance threshold corresponds to the 10th quantile of the interaction strength coefficient, and the establishment threshold corresponds to the 80th quantile of the interaction strength coefficient. For existing topological edges, their interaction strength coefficients are calculated in real time. When the interaction strength coefficient falls below the maintenance threshold, the corresponding topological edge is removed. The removal operation includes deleting the record of that edge from the topological edge set and updating the topological graph. For two unconnected topological nodes, the interaction strength coefficient between them is calculated in real time. When the interaction strength coefficient exceeds the establishment threshold, a new topological edge is created between the two topological nodes. The creation operation includes adding a new record to the topological edge set and initializing the edge's attributes, including relative distance and relative velocity. The dynamic maintenance period for topological edges is set to 0.1 seconds, meaning that every 0.1 seconds, all possible node pairs are traversed, and the above check and update operations are performed. During maintenance, an adjacency matrix is ​​used to store the topological structure, and the matrix elements... Represents a node With nodes Does a topological edge exist between them? If so, then ,otherwise After each update, the adjacency matrix is ​​recalculated to ensure that the topological relationships are consistent with the current traffic conditions.

[0073] In S3, the process of determining the initial risk factor for a topological node based on the type of traffic participant and its specific location within the road area includes the following steps: First, the highway driving lanes are divided into four strip-shaped areas according to their distance from the emergency lane. The emergency lane is marked as an area. The location risk baseline is set to 1.0; adjacent lanes are marked as areas. The location risk baseline is set at 2.0; the middle lane is marked as an area. The location risk baseline is set at 3.0; the innermost lane is marked as an area. The location risk baseline is set to 4.0. The boundary lines of the area divisions are parallel to the road direction, and each area is 3.75 meters wide. Next, the specific strip area where the traffic participant corresponding to the topological node is currently located is determined, and the angular relationship between the motion trend vector and the current strip area boundary line is analyzed. The angle between the motion trend vector and the boundary line is denoted as... This is determined by calculating the angle between the direction of the motion trend vector and the boundary line normal vector. Finally, the base values ​​for each type are determined based on the type of traffic participant: 3.0 for pedestrians, 2.0 for non-motorized vehicles, and 1.0 for motorized vehicles. Location risk base of the strip region Multiply by, then multiply by the lane crossing coefficient determined based on the angle relationship. To obtain the initial risk factor Lane crossing coefficient The calculation formula is: ,in It is the angle between the trend vector and the current boundary line of the strip region, ranging from 0 to 180 degrees. The formula for calculating the initial risk factor is: .

[0074] The process of calculating the risk propagation weight based on the interaction strength coefficient and relative distance contained in the attributes of each topological edge is carried out in the following steps: First, obtain the interaction strength coefficient of the topological edge. and relative distance Relative distance This is Euclidean distance measured in meters. Risk propagation weight. The calculation formula is: The denominator in the formula uses This is to avoid relative distance A division-by-zero error occurs when the value is 0, while ensuring the risk propagation weight. With interaction strength coefficient Proportional to the relative distance Inversely proportional. The calculated risk propagation weights It is a dimensionless scalar value, ranging from 0 to positive infinity. The larger the value, the higher the intensity of risk transmission.

[0075] The process of collecting the propagation risk value of all neighboring topological nodes of each topological node through topological edges includes the following specific steps: First, for each topological node... Identify all its neighboring topological nodes These adjacent nodes are connected to each other via topological edges. They are directly connected. Secondly, for each adjacent node... Calculate its propagation to nodes through topological edges. The risk value of transmission. The calculation formula is: ;in Adjacent nodes Current risk factors It is a connection node and nodes The risk propagation weights of the topological edges are then calculated. Finally, the propagation risk values ​​from all neighboring nodes are summed to obtain the node's risk weight. Total propagation risk value received The calculation formula is: ;in Represents a node The set of all adjacent nodes. The calculation period for the propagation risk value is consistent with the topology update period, set to 0.1 seconds.

[0076] The process of superimposing the risk factor of a topology node itself with all the aggregated propagation risk values, and using the superposition result to update the comprehensive influence factor of the corresponding topology node, is performed in the following steps: First, obtain the topology node. Its own initial risk factors Total transmission risk value Comprehensive Influence Factor The update formula is: ;in This is the risk propagation attenuation coefficient, set at 0.7. Based on historical data statistical analysis, and through regression analysis of 500 risk transmission scenarios, the optimal attenuation coefficient was determined to be 0.7. This coefficient is used to adjust the contribution of the transmission risk value to the comprehensive impact factor. (Comprehensive Impact Factor) This is a dimensionless scalar value; a larger value indicates a higher overall risk level for the topology node. The updated comprehensive impact factor will be used in subsequent risk group and risk chain identification processes. The update cycle of the comprehensive impact factor is consistent with the topology maintenance cycle to ensure the real-time nature of risk assessment.

[0077] In S4, the process of traversing all topology nodes and marking those with a comprehensive impact factor exceeding a preset activation threshold as risk seed nodes includes the following steps: First, the activation threshold is set to 1.5. This threshold is derived from historical accident data analysis. Through statistical analysis of 500 risk scenarios, it was determined that the probability of an accident increases significantly when the comprehensive impact factor reaches 1.5. The traversal process uses a depth-first search algorithm to check the comprehensive impact factor value of each topology node. When the comprehensive impact factor value is greater than or equal to the activation threshold, the node is marked as a risk seed node, and its node identifier, comprehensive impact factor value, and timestamp are recorded. After marking, a set of risk seed nodes is generated, where each seed node contains all its basic attribute information. This process is executed every 0.1 seconds to ensure real-time performance.

[0078] Starting from each risk seed node, the process of constructing a risk group by tracing outwards along the topological edges connected to it is implemented as follows: First, a first threshold of 1.2 is set. This threshold is determined through cluster analysis and can effectively distinguish high-risk node groups. For each risk seed node, a breadth-first search is performed along the topological edges, using it as the starting point. During the search, the comprehensive influence factor value of each neighboring node is checked. When the comprehensive influence factor value is greater than or equal to the first threshold, the node is included in the current risk group, and the process continues to expand outwards from it as a new starting point. The tracing process stops when a node with a comprehensive influence factor value lower than the first threshold is encountered. The final risk group consists of at least one risk seed node and several neighboring nodes. All nodes within a group are directly or indirectly connected through topological edges, and the length of the connection path between any two nodes does not exceed the set maximum path length of 5.

[0079] The process of identifying risk chains within a risk group includes the following specific steps: First, within the identified risk group, search for all node sequences with continuous topological edge connections. For each node sequence, check if its length meets the requirement of being at least 3. Then, verify whether the comprehensive influence factor in the sequence shows a sequentially increasing trend, that is, for any adjacent node pair in the sequence, the comprehensive influence factor value of the later node is greater than that of the earlier node. Simultaneously, there must be a direct topological edge connection between adjacent nodes. Node sequences that meet these conditions are constructed as risk chains. Each risk chain contains at least 3 nodes and at most the total number of nodes in its risk group. During the identification of risk chains, the difference in the comprehensive influence factor values ​​of adjacent nodes in the chain must also be greater than or equal to 0.1 to ensure the significance of the increasing trend.

[0080] The threshold parameters used in this invention are derived from statistical analysis of a large amount of historical data. The activation threshold of 1.5 corresponds to the 85th quantile of the comprehensive impact factor, ensuring that only truly risky nodes are marked. The first threshold of 1.2 corresponds to the 70th quantile of the comprehensive impact factor, used to define the boundaries of the risk group. These threshold settings have been validated, achieving a recall rate of 95% and a precision rate of 88% on the test dataset. The thresholds are updated every 24 hours, dynamically adjusted based on the latest collected data to ensure adaptability to different traffic conditions and environmental circumstances.

[0081] In S5, the process of determining the corresponding early warning level based on the proportion of risk seed nodes in a risk group or the length of a risk chain includes the following steps: First, calculate the proportion of risk seed nodes in the risk group, which is obtained by dividing the number of risk seed nodes by the total number of nodes in the risk group. When this proportion exceeds 50%, the early warning level is determined to be Level 1; when the proportion is between 30% and 50%, the early warning level is determined to be Level 2; and when the proportion is below 30%, the early warning level is determined to be Level 3. Simultaneously, for risk chains, when the chain length reaches 4 or more nodes, the early warning level is determined to be Level 1; and when the chain length is 3 nodes, the early warning level is determined to be Level 2. If both risk groups and risk chains exist simultaneously, the higher early warning level is taken as the final level. The calculation cycle for determining the early warning level is 0.5 seconds to ensure timely response to changes in risk.

[0082] Based on the determined warning level, the process of generating a warning instruction set containing the coordinates of the warning area, risk type, and risk evolution trend is implemented according to the following steps: First, the coordinates of the warning area are obtained by calculating the average coordinates of all nodes in the risk group or risk chain. Specifically, the calculation method is to sum the x-coordinates of all nodes and divide by the number of nodes; the y-coordinates are calculated similarly. Risk types are divided into two categories based on the characteristics of the risk group or risk chain: group clustering and chain transmission. The risk evolution trend is obtained by analyzing the rate of change in the size of the risk group or the rate of change in the length of the risk chain within the last 3 seconds. The rate of change is calculated by dividing the difference between the current value and the value 3 seconds ago by the time interval. The warning instruction set is organized in a specific format and includes five fields: warning level, area center coordinates, risk impact radius, risk type code, and trend change rate.

[0083] According to the warning instruction set, the process of synchronously executing warning actions includes the following specific steps: First, the corresponding graphic warning symbol is displayed on variable message signs 1.5 kilometers upstream of the warning area. A Level 1 warning displays a red exclamation mark icon, a Level 2 warning displays a yellow triangle icon, and a Level 3 warning displays a blue circle icon. Simultaneously, directional audio-visual devices deployed along the roadside emit focused alarms in the direction of risk transmission. The alarm sound intensity increases with the warning level: 80 decibels for Level 1, 70 decibels for Level 2, and 60 decibels for Level 3. Finally, a warning report is sent to the monitoring center. The report includes a topological structure map of the at-risk group or a path map of the risk chain, as well as a record of the risk evolution process over the past 10 minutes. All warning actions are initiated and executed within 0.2 seconds of receiving the instruction set to ensure timely warnings.

[0084] Example 3, please refer to Figure 2 As shown, the pedestrian and non-motorized vehicle risk identification system along the highway includes:

[0085] The traffic environment status perception module is used to monitor a designated area along the highway in real time, identify and track each traffic participant within the designated area along the highway, and generate a traffic participant status set containing the unique identifier, real-time location, and real-time speed of each traffic participant; the traffic participants include: village pedestrians, non-motorized vehicles, and motorized vehicles;

[0086] The dynamic interaction relationship construction module constructs a dynamic interaction relationship topology for all traffic participants at the current moment, based on the set of traffic participant states. Each traffic participant is treated as a topology node, and a topology edge is established between any two traffic participants with potential interaction behaviors. Each topology node is assigned node attributes, and each topology edge is assigned edge attributes.

[0087] The risk propagation and impact assessment module calculates the initial risk factor for each topological node based on the dynamic interaction relationship topology. Then, according to the edge attributes of the topological edges, the initial risk factor of each topological node is propagated and weighted along the topological edges connected to the corresponding topological nodes to update and obtain the comprehensive impact factor of each topological node.

[0088] The collaborative risk pattern recognition module identifies risk groups or risk chains that meet preset conditions based on the updated comprehensive impact factors of all topological nodes. Among them, a risk group consists of a group of topological nodes that are closely connected by topological edges and whose comprehensive impact factors all exceed the first threshold; a risk chain consists of a group of topological nodes that are connected by topological edges and whose comprehensive impact factors are passed on sequentially and exceed the second threshold.

[0089] The tiered early warning information generation and release module, in response to the identification of risk groups or risk chains, generates and issues scenario-level early warning information that matches the risk level of the risk group or risk chain.

[0090] The working principle of this invention is as follows: First, monitoring data of a designated area along a highway is collected using multispectral and thermal imaging cameras. The multispectral image data and thermal imaging data are fused to generate a fused feature vector. Based on this, individual traffic participants are identified and assigned unique identifiers. Real-time position and speed are calculated using binocular visual ranging and differential methods to generate a set of traffic participant states. Next, a dynamic interaction relationship topology is constructed based on this set, abstracting traffic participants as topological nodes. Topological edges are established and dynamically maintained by calculating motion trend vectors and interaction strength coefficients. Then, initial risk factors are calculated based on road area division and participant types. Risk factors are propagated and weighted in the topological network through risk propagation weights to update and obtain a comprehensive impact factor. Furthermore, risk groups and risk chains are identified by setting thresholds. Risk groups consist of closely connected nodes exceeding the risk threshold, and risk chains consist of a continuous sequence of nodes with increasing risk. Finally, the warning level is determined based on the risk characteristics, a warning instruction set is generated, and graded warnings are implemented through variable message signs, directional audio-visual devices, and reports from the monitoring center. This solution achieves accurate identification and early warning of risks to pedestrians and non-motorized vehicles along highways through multi-source data fusion, dynamic topology construction, and risk propagation mechanisms.

[0091] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for identifying risks to pedestrians and non-motorized vehicles in villages along highways, characterized in that... Includes the following steps: S1: Real-time monitoring of a designated area along the highway, identifying and tracking each traffic participant within the designated area along the highway, and generating a traffic participant status set containing each traffic participant's unique identifier, real-time location, and real-time speed; the traffic participants include: village pedestrians, non-motorized vehicles, and motorized vehicles; S2: Based on the set of traffic participant states, construct a dynamic interaction relationship topology for all traffic participants at the current moment; wherein, each traffic participant is treated as a topology node, and a topology edge is established between any two traffic participants with potential interaction behaviors; and each topology node is assigned node attributes, and each edge is assigned edge attributes. S3: Based on the dynamic interaction relationship topology, calculate the initial risk factor of each topological node; then, according to the edge attributes of the topological edges, propagate and weight the initial risk factor of each topological node along the topological edges connected to the corresponding topological nodes in a directional manner and aggregate them to update the comprehensive influence factor of each topological node. S4: Based on the updated comprehensive impact factors of all topological nodes, identify risk groups or risk chains that meet preset conditions; wherein, a risk group consists of a group of topological nodes that are closely connected by topological edges and whose comprehensive impact factors all exceed the first threshold; a risk chain consists of a group of topological nodes that are connected by topological edges and whose comprehensive impact factors are passed on sequentially and exceed the second threshold. S5: In response to the identification of risk groups or risk chains, generate and issue scenario-level early warning information that matches the risk level of the risk group or risk chain.

2. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 1, characterized in that, The process of generating the traffic participant state set is as follows: S11: Simultaneously acquire multispectral image data and thermal imaging data deployed in designated areas along the highway; S12: Perform fusion processing on multispectral image data and thermal imaging data to generate a fused feature vector with enhanced spectral and thermal radiation features; S13: Based on the fused feature vector, identify each individual traffic participant within a designated area along the highway and assign each individual traffic participant a unique identifier that remains unchanged throughout the entire monitoring period. S14: Based on the identity identifier, associate the spatial coordinates and movement speed of each traffic participant in a continuous time sequence to generate a traffic participant state set.

3. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 1, characterized in that, S2 specifically includes: S21: Based on the real-time position and real-time speed in the set of traffic participants' states, calculate a motion trend vector for each topology node. The motion trend vector is used to characterize the future instantaneous movement direction and movement intention of the corresponding traffic participant. S22: Calculate the interaction strength coefficient based on the node attributes of the two topological nodes connected by each established topological edge; S23: Based on the interaction strength coefficient, dynamically maintain the existence state of the topological edge; when the interaction strength coefficient is lower than the preset maintenance threshold, the corresponding topological edge is removed; when the interaction strength coefficient between two unconnected topological nodes is higher than the preset establishment threshold, a new topological edge is created between the two topological nodes.

4. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 3, characterized in that, S21 specifically includes: S211: Extract historical location points within the most recent complete motion cycle from the continuously updated location sequence of each traffic participant to form a short-term trajectory segment; S212: Divide the short-term trajectory segment into two equal-length sub-segments, calculate the geometric centroid of the position points contained in each sub-segment, and determine the direction vector from the centroid of the first sub-segment to the centroid of the last sub-segment. S213: Combine the direction of the direction vector with the scalar value of the latest real-time speed of the traffic participant to generate a motion trend vector; where the direction of the direction vector defines the direction of the motion trend vector, and the scalar value of the real-time speed defines the magnitude of the motion trend vector.

5. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 3, characterized in that, The calculation of the interaction strength coefficient specifically includes: S221: Calculate the spatial connection vector from the first topological node to the second topological node, and calculate the first angle between the spatial connection vector and the motion trend vector of the first topological node, and the second angle between the spatial connection vector and the motion trend vector of the second topological node. S222: Obtain the real-time velocity of two topological nodes and calculate the scalar value of the relative velocity of the real-time velocity; S223: The cosine values ​​of the first included angle and the second included angle are weighted and summed to obtain the convergence status evaluation value. The convergence status evaluation value is then multiplied by the scalar value of the relative velocity to obtain the interaction intensity coefficient.

6. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 1, characterized in that, S3 specifically includes: S31: Determine the initial risk factor of the corresponding topological node based on the type of traffic participant and its specific location within the road area; S32: Calculate the risk propagation weight based on the interaction strength coefficient and relative distance contained in the attributes of each topological edge, where the interaction strength coefficient is directly proportional to the risk propagation weight and the relative distance is inversely proportional to the risk propagation weight; S33: Collect the propagation risk value of all adjacent topological nodes of each topological node through topological edges. The propagation risk value is the product of the current risk factor of the adjacent topological node and the risk propagation weight of the corresponding topological edge. S34: Superimpose the risk factor of the topology node itself with all the propagation risk values ​​gathered, and use the superposition result to update the comprehensive impact factor of the corresponding topology node.

7. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 6, characterized in that, The process for determining the initial risk factors is as follows: S311: Divide the highway driving lanes into multiple strip zones with increasing risk levels based on their distance from the emergency lane, and set a corresponding location risk base for each strip zone; S312: Determine the specific strip area where the traffic participant corresponding to the topology node is currently located, and analyze the angular relationship between the motion trend vector and the current strip area boundary line; S313: Determine the type base value based on the type of traffic participant, multiply the type base value by the location risk base of the strip area, and then multiply by the lane crossing coefficient determined based on the angle relationship to obtain the initial risk factor.

8. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 1, characterized in that, S4 specifically includes: S41: Traverse all topology nodes and mark the topology nodes whose comprehensive impact factor exceeds the preset activation threshold as risk seed nodes; S42: Starting from each risk seed node, trace outward along the topological edge it connects, and include all adjacent topological nodes on the tracing path whose comprehensive impact factor exceeds the first threshold into the same risk group, until a topological node with a comprehensive impact factor below the first threshold is encountered. S43: Within the risk group, identify three or more topological nodes that are connected by continuous topological edges and whose comprehensive influence factors show an increasing trend, and construct the corresponding group of topological nodes into a risk chain; wherein, the increasing trend means that along the connection direction of the topological edge, the comprehensive influence factor of the subsequent topological node is greater than the comprehensive influence factor of the preceding topological node.

9. The method for identifying pedestrian and non-motorized vehicle risks in villages along highways according to claim 1, characterized in that, S5 specifically includes the following steps: S51: Determine the corresponding early warning level based on the proportion of risk seed nodes in the risk group or the length of the risk chain; S52: Based on the determined warning level, generate a set of warning instructions that includes the coordinates of the warning area, the risk type, and the risk evolution trend; S53: According to the warning instruction set, the following warning actions shall be executed simultaneously: display the corresponding graphic warning symbol on the variable information sign upstream of the warning area, issue a focused alarm in the direction of risk transmission through the roadside directional sound and light device, and send a warning report containing a risk group topology map or risk chain path map to the monitoring center.

10. A risk identification system for pedestrians and non-motorized vehicles in villages along highways, characterized in that: The method for identifying pedestrian and non-motorized vehicle risks in villages along highways as described in any one of claims 1-9 includes: The traffic environment status perception module is used to monitor a designated area along the highway in real time, identify and track each traffic participant within the designated area along the highway, and generate a traffic participant status set containing the unique identifier, real-time location, and real-time speed of each traffic participant; the traffic participants include: village pedestrians, non-motorized vehicles, and motorized vehicles; The dynamic interaction relationship construction module constructs a dynamic interaction relationship topology for all traffic participants at the current moment, based on the set of traffic participant states. Each traffic participant is treated as a topology node, and a topology edge is established between any two traffic participants with potential interaction behaviors. Each topology node is assigned node attributes, and each topology edge is assigned edge attributes. The risk propagation and impact assessment module calculates the initial risk factor for each topological node based on the dynamic interaction relationship topology. Then, according to the edge attributes of the topological edges, the initial risk factor of each topological node is propagated and weighted along the topological edges connected to the corresponding topological nodes to update and obtain the comprehensive impact factor of each topological node. The collaborative risk pattern recognition module identifies risk groups or risk chains that meet preset conditions based on the updated comprehensive impact factors of all topological nodes. Among them, a risk group consists of a group of topological nodes that are closely connected by topological edges and whose comprehensive impact factors all exceed the first threshold; a risk chain consists of a group of topological nodes that are connected by topological edges and whose comprehensive impact factors are passed on sequentially and exceed the second threshold. The tiered early warning information generation and release module, in response to the identification of risk groups or risk chains, generates and issues scenario-level early warning information that matches the risk level of the risk group or risk chain.

Images