A method and system for active cooperative warning of an electric bicycle at a signal-free intersection

Abstract

The application discloses a kind of active cooperative early warning method and system for signal intersection electric bicycle, it is related to Internet of Things and edge computing field.The method obtains intersection continuous image frame, and the pixel track sequence of motor vehicle and electric bicycle is converted into multidimensional motion state tensor;According to the track extrapolation in space conflict area according to intersection physical topology, mapping brake envelope surface and extracting double-end space-time overlapping feature;Based on the occlusion profile, construct sight line topology occlusion matrix, evaluate the mutual visible state in overlapping feature, fuse and solve collision approaching potential energy and distribute differentiated intervention weight, package alarm message;Through edge cooperative gateway, alarm message is respectively split and issued to the execution unit in the direction of motor vehicle and electric bicycle, analyze weight and trigger corresponding sound and light intervention sequence.The system is used to realize the above method.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things and edge computing technology, specifically to a method and system for proactive collaborative early warning of electric bicycles at unsignalized intersections. Background Technology

[0002] The urban road network contains numerous level crossings without traffic lights. Electric bicycles, a primary mode of transportation for residents, are characterized by their high speed, agile maneuverability, and large numbers. At unsignalized intersections, motor vehicles and electric bicycles frequently intersect. Due to unclear right-of-way divisions and the randomness of traffic participants' behavior, collisions are highly likely to occur between motor vehicle drivers and electric bicycle riders in conflict zones, especially when visibility is limited or anticipation is insufficient.

[0003] Existing intersection warning systems typically rely on fixed sensors to detect approaching vehicles and trigger omnidirectional audible and visual alarms. These systems treat motor vehicles and non-motorized vehicles as point masses with similar motion characteristics, failing to consider the physical differences between the greater braking inertia of motor vehicles and the shorter braking distance of electric bicycles. This leads to collision risk projections deviating from actual physical laws. Furthermore, intersections are often surrounded by obstructions such as green belts, buildings, or illegally parked vehicles. Existing methods generate warning commands based solely on the absolute spatial distance between vehicles, neglecting the visibility of other road users. When motor vehicles and electric bicycles are in mutually obstructing blind spots, a uniform alarm output cannot provide targeted collision avoidance intervention for high-risk targets. This can easily cause visual fatigue and decreased vigilance among road users, resulting in poor effectiveness and specificity of the warning system in complex intersection environments. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an active cooperative early warning method and system for electric bicycles at unsignalized intersections to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for active cooperative early warning of electric bicycles at unsignalized intersections, comprising the following steps: S1, acquiring continuous image frames of the unsignalized intersection, tracking motor vehicle targets and electric bicycle targets, extracting pixel trajectory sequences of the motor vehicle targets and electric bicycle targets, and converting the pixel trajectory sequences into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate; S2, dividing the spatial conflict zone according to the physical topology of the intersection, extrapolating the trajectory into the spatial conflict zone based on the multidimensional motion state tensor, mapping the braking envelope of the motor vehicle and the braking envelope of the electric bicycle respectively, and extracting the spatiotemporal overlap features of the double-ended braking envelope in the spatial conflict zone; 3. Construct a line-of-sight topology occlusion matrix based on the static occlusions and dynamic target contours at the intersection, evaluate the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlap features, fuse the mutual visibility state with the double-ended braking envelope surface to calculate the collision approximation potential energy, assign differentiated intervention weights, and encapsulate motor vehicle warning messages and electric bicycle warning messages according to the differentiated intervention weights; S4. Through the edge collaborative gateway, split and send the motor vehicle warning messages and electric bicycle warning messages to the first execution unit in the direction of motor vehicles and the second execution unit in the direction of electric bicycles, respectively. The first execution unit and the second execution unit respectively parse the differentiated intervention weights and trigger the corresponding audio-visual intervention sequence according to the differentiated intervention weights.

[0006] In a preferred embodiment, the specific process of acquiring continuous image frames of an unsignalized intersection, tracking motor vehicle targets and electric bicycle targets, and extracting pixel trajectory sequences of motor vehicle targets and electric bicycle targets is as follows: identifying feature anchor points of motor vehicle targets and electric bicycle targets within continuous image frames, establishing a correlation chain between anchor points through time-domain feature matching; fitting the displacement path of the target motion center in the image coordinate system to form a pixel trajectory sequence that evolves with the image frame sequence.

[0007] In a preferred embodiment, the specific process of converting the pixel trajectory sequence into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate is as follows: a projective transformation matrix between the image coordinate system and the real coordinate system is established through camera calibration parameters, and the pixel trajectory sequence is mapped into a position vector sequence in the real coordinate system; the position vector sequence is processed by order difference to extract instantaneous velocity and acceleration, and the heading deflection rate is extracted based on the rate of change of the tangent azimuth angle of the position vector sequence in the horizontal plane, and encapsulated to form a multidimensional motion state tensor.

[0008] In a preferred embodiment, the spatial conflict zone is divided according to the physical topology of the intersection, and the trajectory is extrapolated into the spatial conflict zone based on the multidimensional motion state tensor to map the braking envelope of motor vehicles and the braking envelope of electric bicycles respectively. The specific process is as follows: the geometric topological boundary of the intersection center area is defined to form the spatial conflict zone. The braking envelope of motor vehicles is represented by a one-dimensional longitudinal occupancy interval generated by mapping the instantaneous velocity and predetermined braking acceleration of the motor vehicle target in the spatial conflict zone. The braking envelope of electric bicycles is represented by a two-dimensional fan-shaped scanning occupancy area generated by mapping the heading deflection rate and lateral displacement trend of the electric bicycle target in the spatial conflict zone.

[0009] In a preferred embodiment, the specific process of extracting the spatiotemporal overlap features of the dual-end braking envelope surface in the spatial conflict zone is as follows: synchronizing the time axis references of the motor vehicle target and the electric bicycle target during trajectory extrapolation, and retrieving the geometric intersection portion of the motor vehicle braking envelope surface and the electric bicycle braking envelope surface in the spatial conflict zone; extracting the overlap displacement amount in the spatial dimension and the overlap duration in the time axis of the geometric intersection portion as spatiotemporal overlap features characterizing the degree of collision risk.

[0010] In a preferred embodiment, the specific process of constructing a line-of-sight topology occlusion matrix based on static obstructions and dynamic target contours at an intersection, and evaluating the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlap features, is as follows: Extracting the driver's viewpoint spatial coordinates of the motor vehicle target and the overall circumscribed geometric contour of the electric bicycle target; connecting the driver's viewpoint spatial coordinates and the edges of the overall circumscribed geometric contour to generate a line-of-sight ray beam; extracting the three-dimensional boundary information of fixed obstacles in the background environment and the three-dimensional boundary information of the dynamic target contour within continuous image frames as an occlusion mask; calculating the spatial interference distribution of the line-of-sight ray beam and the occlusion mask in the real coordinate system to generate a line-of-sight topology occlusion matrix; and calibrating the mutual visibility state between the motor vehicle target and the electric bicycle target on the time axis section corresponding to the spatiotemporal overlap features based on the interference occlusion and transmission state within the line-of-sight topology occlusion matrix.

[0011] In a preferred embodiment, the specific process of fusing the mutually visible state with the double-ended braking envelope to calculate the collision approximation potential energy, assigning differentiated intervention weights, and encapsulating the vehicle alarm message and the electric bicycle alarm message according to the differentiated intervention weights is as follows: Extract the approximation gradient parameters of the vehicle braking envelope and the electric bicycle braking envelope within the spatiotemporal overlap features; establish a mapping lookup association table, perform joint feature encoding mapping of the approximation gradient parameters and the mutually visible state to output the collision approximation potential energy; based on the energy level interval division results of the collision approximation potential energy, assign a first-level intervention weight to the vehicle target and a second-level intervention weight to the electric bicycle target, the first-level intervention weight and the second-level intervention weight constitute differentiated intervention weights; according to the low-power local area IoT communication protocol frame format, encapsulate the first-level intervention weights into a vehicle alarm message using channel encoding, and encapsulate the second-level intervention weights into an electric bicycle alarm message using channel encoding.

[0012] In a preferred embodiment, the specific process of splitting and sending vehicle alarm messages and electric bicycle alarm messages to the first execution unit in the direction of the vehicle and the second execution unit in the direction of the electric bicycle respectively through the edge collaborative gateway is as follows: extract the motion state tensor heading of the vehicle target to the corresponding physical space entrance associated hardware medium access control address, and the motion state tensor heading of the electric bicycle target to the corresponding physical space entrance associated hardware medium access control address; call the radio frequency transceiver chip to establish a star-shaped spread spectrum modulation communication link, and through the star-shaped spread spectrum modulation communication link, send the vehicle alarm message to the first execution unit according to the corresponding physical space entrance associated hardware medium access control address, and send the electric bicycle alarm message to the second execution unit according to the corresponding physical space entrance associated hardware medium access control address.

[0013] In a preferred embodiment, the first execution unit and the second execution unit respectively parse the differentiated intervention weights, and the specific process of triggering the corresponding audio-visual intervention sequence according to the differentiated intervention weights is as follows: The first execution unit restores the motor vehicle alarm message through the internal baseband radio frequency decoding module to extract the first-level intervention weight within the differentiated intervention weight, and uses the pulse width modulation port to output a drive level signal to trigger the LED array connected locally to the first execution unit to generate a low-frequency flashing sequence; The second execution unit restores the electric bicycle alarm message through the internal baseband radio frequency decoding module to extract the second-level intervention weight within the differentiated intervention weight, and uses the digital output pin to drive the LED array connected locally to the second execution unit in parallel to generate a high-frequency strobe sequence and drive the electroacoustic conversion device to generate an emergency alarm audio sequence. The low-frequency flashing sequence, the high-frequency strobe sequence and the emergency alarm audio sequence are combined to form an audio-visual intervention sequence.

[0014] An active cooperative early warning system for electric bicycles at unsignalized intersections, used to execute the aforementioned active cooperative early warning method for electric bicycles at unsignalized intersections, includes: a state tensor module, used to acquire continuous image frames of the unsignalized intersection, track motor vehicle targets and electric bicycle targets, extract pixel trajectory sequences of the motor vehicle targets and electric bicycle targets, and convert the pixel trajectory sequences into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate; and a spatiotemporal overlap module, used to divide spatial conflict zones according to the physical topology of the intersection, extrapolate the trajectories into the spatial conflict zones based on the multidimensional motion state tensor, map the braking envelope surfaces of motor vehicles and electric bicycles respectively, and extract the spatiotemporal overlap characteristics of the braking envelope surfaces of motor vehicles and electric bicycles within the spatial conflict zones. The system comprises: a warning encapsulation module, used to construct a line-of-sight topology occlusion matrix based on the outlines of static obstructions and dynamic targets at intersections; an assessment of the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlap features; fusion of the mutual visibility state with the braking envelopes of motor vehicles and electric bicycles to calculate the collision approximation potential energy; allocation of differentiated intervention weights; and encapsulation of motor vehicle warning messages and electric bicycle warning messages based on the differentiated intervention weights; and a warning execution module, used to split and distribute motor vehicle warning messages and electric bicycle warning messages to a first execution unit in the direction of motor vehicles and a second execution unit in the direction of electric bicycles respectively through an edge collaborative gateway. The first and second execution units respectively parse the differentiated intervention weights and trigger the corresponding audio-visual intervention sequences based on the differentiated intervention weights.

[0015] The present invention has the following beneficial effects:

[0016] (1) An active collaborative early warning method for electric bicycles at unsignalized intersections, in the perception and conflict inference stage, converts the pixel trajectory in continuous image frames into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate, providing basic state parameters for the motion evolution of the intersection target; based on this, the trajectory is extrapolated into the spatial conflict zone according to the multidimensional motion state tensor, mapping the braking envelope of motor vehicles and the braking envelope of electric bicycles respectively. Thus, it can overcome the problem of physical characteristic deviation caused by treating motor vehicles and electric bicycles as single points in the prior art, and by extracting the spatiotemporal overlap features of the double-ended braking envelope in the spatial conflict zone, the collision prediction is more in line with the actual physical operation law of the large inertia of motor vehicles and the flexible deflection of electric bicycles.

[0017] (2) An active collaborative early warning system for electric bicycles at unsignalized intersections, in the risk assessment and collaborative execution phase, constructs a line-of-sight topology occlusion matrix based on the static occlusions and dynamic target contours at the intersection to assess the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlap features. The mutual visibility state is then fused with the double-ended braking envelope surface to calculate the collision approximation potential energy, thereby allocating differentiated intervention weights. Subsequently, the corresponding alarm messages are split and distributed to execution units in different directions via an edge collaborative gateway and a low-power local area network. This improves upon existing early warning schemes that rely solely on absolute distance to trigger omnidirectional alarms, resulting in poor targeting and visual fatigue. It achieves precise directional collaborative intervention based on blind spot conditions and asymmetric physical characteristics.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of an active cooperative early warning method for electric bicycles at unsignalized intersections according to the present invention.

[0020] Figure 2 This is a schematic diagram of the mapping between spatiotemporal conflict and braking envelope surface.

[0021] Figure 3 This is a schematic diagram of the nonlinear mapping of the collision approximation potential energy.

[0022] Figure 4 This is a flowchart of an active cooperative early warning system for electric bicycles at unsignalized intersections according to the present invention. Detailed Implementation

[0023] This application provides an active collaborative early warning method and system for electric bicycles at unsignalized intersections, which solves the problems of existing intersection early warning schemes failing to distinguish the physical braking characteristics of traffic participants and ignoring blind spot conditions, resulting in poor early warning targeting and low intervention effectiveness.

[0024] The overall approach of the scheme in this application embodiment is as follows: The edge collaborative gateway acquires video data from the intersection and extracts the multi-dimensional motion state tensors of motor vehicles and electric bicycles. Trajectory extrapolation is performed within the spatial conflict zone divided according to the physical topology of the intersection to generate the braking envelope surface of motor vehicles and the braking envelope surface of electric bicycles, so as to obtain the spatiotemporal overlap characteristics of the double-ended braking envelope surface in the spatial conflict zone. Then, a line-of-sight topology occlusion matrix is ​​constructed by combining the intersection occlusion contour, and the mutual visibility state of motor vehicles and electric bicycles in the spatiotemporal overlap state is evaluated. The mutual visibility state is combined with the double-ended braking envelope surface to calculate the collision approximation potential energy. Differentiated intervention weights are generated based on the collision approximation potential energy and the message is encapsulated. Finally, the motor vehicle alarm message and the electric bicycle alarm message are sent to the first execution unit and the second execution unit in the corresponding direction through a low-power local area Internet of Things. The first execution unit and the second execution unit parse the differentiated intervention weights and trigger independent audio-visual intervention sequences.

[0025] Example 1; please refer to Figure 1 This invention provides a technical solution: a method for active cooperative early warning of electric bicycles at unsignalized intersections, comprising the following steps: S1, acquiring continuous image frames of the unsignalized intersection, tracking motor vehicle targets and electric bicycle targets, extracting pixel trajectory sequences of the motor vehicle targets and electric bicycle targets, and converting the pixel trajectory sequences into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate; S2, dividing the spatial conflict zone according to the physical topology of the intersection, extrapolating the trajectory into the spatial conflict zone based on the multidimensional motion state tensor, mapping the braking envelope surface of the motor vehicle and the braking envelope surface of the electric bicycle respectively, and extracting the spatiotemporal overlap features of the double-ended braking envelope surface in the spatial conflict zone; S3, ... Based on the static occlusions and dynamic target contours at the intersection, a line-of-sight topology occlusion matrix is ​​constructed to evaluate the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlap features. The mutual visibility state is fused with the double-ended braking envelope to calculate the collision approximation potential energy, and differentiated intervention weights are assigned. Based on the differentiated intervention weights, motor vehicle alarm messages and electric bicycle alarm messages are encapsulated. S4, The motor vehicle alarm messages and electric bicycle alarm messages are split and sent to the first execution unit in the direction of motor vehicles and the second execution unit in the direction of electric bicycles, respectively, through the edge collaborative gateway. The first execution unit and the second execution unit respectively parse the differentiated intervention weights and trigger the corresponding audio-visual intervention sequence according to the differentiated intervention weights.

[0026] In this implementation scheme, step S1 mainly achieves basic state perception and physical quantification of traffic participants at the intersection. By acquiring continuous images of the intersection, the system identifies and records the moving coordinate points of motor vehicles and electric bicycles in the two-dimensional image frame; this coordinate point array is the pixel trajectory sequence. The system, combined with camera calibration parameters, maps this coordinate point array into a multidimensional motion state tensor. The multidimensional motion state tensor is a mathematical structure that integrates and encapsulates the instantaneous velocity, acceleration / deceleration rate, and deflection angle of the target in three-dimensional physical space. This process completes the extraction of basic kinematic parameters, providing objective data input for subsequent collision evolution prediction.

[0027] Step S2 primarily involves calculating and extrapolating the future physical intersection risks between motor vehicles and electric bicycles. Based on the actual geographical boundaries of the intersection and the vehicle trajectories, the system delineates specific physical areas where vehicle intersections may occur as spatial conflict zones. Using the multidimensional motion state tensor obtained in the previous step, the system predicts the future driving paths of the vehicles and, combined with vehicle braking performance, calculates and generates the braking envelope surfaces of both motor vehicles and electric bicycles. The braking envelope surface refers to the maximum physical space occupied by the target vehicle at its current speed and acceleration state until it comes to a complete stop after emergency braking. By comparing the spatiotemporal overlap characteristics of these two braking envelope surfaces within the spatial conflict zone—that is, calculating the overlapping area and duration of the braking space occupied by the two vehicles within the same time period—the actual physical intersection overlap state is confirmed. To more intuitively illustrate the above spatial extrapolation process, this embodiment provides a spatiotemporal conflict and braking envelope surface mapping diagram, such as... Figure 2 As shown in the figure, this diagram illustrates the geometric occupancy of the vehicle braking envelope (longitudinal distribution data) and the electric bicycle braking envelope (sector distribution data) within the intersection's spatial conflict zone under the same prediction time axis. The overlapping portion of the two data clusters in the figure represents the potential collision risk area, intuitively reflecting the spatiotemporal overlap characteristics extracted by the system after extrapolating the trajectories of targets with different physical attributes.

[0028] Step S3 primarily involves the cognitive assessment and orientation strategy formulation for blind spot risks at intersections. The system extracts the dynamic outlines of fixed obstructions such as roadside trees and buildings, as well as large vehicles, to construct a spatial data structure reflecting the line-of-sight or obstruction relationship between targets, namely, a line-of-sight topology occlusion matrix. This matrix is ​​used to determine whether drivers of motor vehicles and riders of electric bicycles are in each other's blind spots when predicted to meet, thus obtaining their mutual visibility status. The system then superimposes the line-of-sight occlusion situation with the aforementioned spatiotemporal overlap features to calculate and output a collision approach potential energy characterizing the degree of risk proximity. The collision approach potential energy is an independent numerical indicator that quantifies and integrates spatial proximity rate and line-of-sight obstruction degree. The quantitative differences in collision approach potential energy under different visibility states can be characterized by a nonlinear mapping graph, such as... Figure 3As shown in the figure, this diagram illustrates the mapping relationship between the approach gradient, relative distance, and collision approach potential energy. The diagram contains two layers of data distribution: the lower layer, with a gently distributed data set, represents the baseline potential energy when traffic participants are mutually visible; the upper layer, exhibiting a steep, exponentially increasing trend, represents the potential energy distribution in blind spots (obstructed vision). This mapping relationship reflects the system's numerical amplification mechanism of collision approach potential energy under obstructed vision conditions. Based on the calculation results of this indicator, the system assigns parameter settings reflecting the severity of the current danger faced by motor vehicles and electric bicycles—that is, differentiated intervention weights—and encapsulates these into independent alarm communication messages.

[0029] Step S4 primarily implements the distributed, targeted distribution of warning commands and the closed-loop execution by physical devices. The system utilizes an edge collaborative gateway for communication distribution. An edge collaborative gateway is a communication hub hardware node deployed at the front end of an intersection, possessing local data processing capabilities and supporting low-power local area network (LAN) protocols. The gateway sends vehicle warning messages to the first execution unit facing the oncoming vehicle direction and electric bicycle warning messages to the second execution unit facing the non-motorized vehicle lane direction. After receiving the messages, the execution hardware at both ends parses out the differentiated intervention weights carried within and converts them into a combination of flashing lights at a specific frequency and audio with a specific waveform—a sound-light intervention sequence. By driving local peripherals to output independent sound-light intervention sequences, customized anti-collision warning actions are completed for targets in different directions.

[0030] Specifically, the process of acquiring continuous image frames of an unsignalized intersection, tracking motor vehicle targets and electric bicycle targets, and extracting the pixel trajectory sequences of motor vehicle targets and electric bicycle targets is as follows: identify the feature anchor points of motor vehicle targets and electric bicycle targets within continuous image frames, establish the association chain between anchor points through time-domain feature matching; fit the displacement path of the target motion center in the image coordinate system to form a pixel trajectory sequence that evolves with the image frame sequence.

[0031] In this implementation scheme, the system performs compliant desensitization processing on the raw video stream data collected from the front end. After masking the privacy regions of faces and license plates within the image frame using a mask matrix, the image frame is input into a pre-set convolutional neural network to output bounding boxes for motor vehicles and electric bicycles. Texture extrema points of the target surface are extracted within the bounding boxes as feature anchor points. Let the set of feature anchor points of the target in the k-th frame be denoted as . For feature matching between adjacent frames, a cost function is constructed that combines spatial and appearance features. The association is performed, and the calculation process is as follows: ;in, This represents the matching cost between the m-th feature anchor in frame k and the n-th feature anchor in frame (k+1). This represents the two-dimensional pixel coordinates of the m-th feature anchor point in the k-th frame; This represents the two-dimensional pixel coordinates of the nth feature anchor point in the (k+1)th frame; Represents the Euclidean distance operation; This represents the local appearance feature vector of the anchor point extracted through the feature layer of the neural network; This represents the cosine similarity between two feature vectors. Indicates the spatial distance weighting coefficient; This represents the appearance similarity weighting coefficient. Spatial distance weighting coefficient. The target's scale is dynamically determined based on its position within the image region. Let D be the diagonal pixel length of the target's bounding box. The calculation relationship is then as follows: ,and ;in, This represents a preset reference scale constant. Based on the constructed cost function matrix, the Hungarian algorithm is used to find the matching pairs corresponding to the minimum cost to form an association chain. The geometric center coordinates of all matching anchor points of the same target in each frame are calculated. A local weighted regression algorithm is used to smooth the high-frequency jitter of the center coordinates, resulting in a smoothed pixel trajectory sequence that evolves with the frame number.

[0032] Specifically, the process of converting the pixel trajectory sequence into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate is as follows: A projective transformation matrix between the image coordinate system and the real coordinate system is established through camera calibration parameters, mapping the pixel trajectory sequence into a position vector sequence in the real coordinate system; the position vector sequence is processed by order difference to extract instantaneous velocity and acceleration, and the heading deflection rate is extracted based on the rate of change of the tangent azimuth angle of the position vector sequence in the horizontal plane, which is then encapsulated to form a multidimensional motion state tensor.

[0033] In this implementation scheme, the system obtains the pre-calibrated camera intrinsic parameter matrix and distortion coefficients offline, performs radial and tangential distortion correction on the pixel trajectory sequence, and then applies a projective transformation matrix to map it to the real physical coordinate system of the ground plane. The coordinate mapping calculation process is as follows: ;in, This represents the physical x-coordinate of the q-th sampling point in the ground plane coordinate system after mapping; This represents the physical ordinate of the q-th sampling point in the ground plane coordinate system after mapping; This indicates that the parameters include the external rotation and translation of the camera. The first-order projective transformation matrix; This represents the x-coordinate of the pixel at the q-th sampling point in the pixel trajectory sequence; This represents the ordinate of the pixel at the q-th sampling point in the pixel trajectory sequence. The mapped position vector sequence is obtained. Then, the second-order central difference algorithm is used to extract kinematic features. The calculation process of the instantaneous velocity vector is as follows: The calculation process for the instantaneous acceleration vector is as follows: ;in, This represents the instantaneous velocity vector at the q-th sampling time. This represents the position vector of the (q+1)th sampling point; This represents the position vector of the (q-1)th sampling point; This represents the sampling interval constant of consecutive image frames in the time domain; Let represent the instantaneous acceleration vector at the q-th sampling time. The process of calculating the rate of change of the tangent azimuth angle of the position vector sequence in the horizontal plane and extracting the instantaneous heading deflection rate is as follows: ;in, This represents the heading deflection rate at the q-th sampling time. After extraction, , and The data formats are aligned and cascaded to obtain a multidimensional motion state tensor.

[0034] Specifically, the spatial conflict zone is divided according to the physical topology of the intersection. Trajectory extrapolation is performed into the spatial conflict zone based on the multidimensional motion state tensor. The specific process of mapping the braking envelope of motor vehicles and the braking envelope of electric bicycles is as follows: The geometric topological boundary of the intersection center area is defined to form the spatial conflict zone. The braking envelope of motor vehicles is represented by a one-dimensional longitudinal occupancy interval generated by mapping the instantaneous velocity and predetermined braking acceleration of the motor vehicle target in the spatial conflict zone. The braking envelope of electric bicycles is represented by a two-dimensional fan-shaped scanning occupancy area generated by mapping the heading deflection rate and lateral displacement trend of the electric bicycle target in the spatial conflict zone.

[0035] In this implementation scheme, the hardware gateway parses the high-precision vector topology file of the unsignalized intersection, extracts the physical coordinates of the four right-angled or chamfered edges of the intersection, connects them to form a closed polygonal geometric boundary set, and marks it as the spatial conflict zone. For the braking envelope of a motor vehicle, it is considered as a one-dimensional longitudinally extending space dominated by the vehicle's physical inertia. The formula for predicting the coordinate vector of the front boundary within is: ;in, Indicates the variable in the prediction time. The coordinate vector of the vehicle's front end occupancy; The initial physical coordinate vector representing the moment when the vehicle just touches the boundary of the conflict zone; Indicates the scalar speed of the vehicle at the initial stage of entry; Indicates the rated longitudinal braking deceleration of a motor vehicle; This represents the time step operator used for integration. Considering the highly random lateral turning of electric bicycles at intersections, their braking envelope is constructed as a two-dimensional sector-shaped scanning region that expands with time. The lateral sweep angle boundary formula for this region is: And its longitudinal reach depth formula is: ;in, This represents the dynamic lateral movement angular domain of the electric bicycle at time \tau; This indicates the initial heading angle of the electric bicycle; This represents the observed initial instantaneous yaw rate; This indicates the acceleration at the handlebar's maximum steering angle. Indicates the farthest longitudinal slip depth constrained by braking capacity; This represents the initial scalar speed of the electric bicycle; This indicates the calibrated longitudinal braking deceleration of the electric bicycle. Among the parameters mentioned above, the calibrated longitudinal braking deceleration... and The numerical method for determining the value is as follows: multiply the factory-standard braking performance parameters of the corresponding vehicle model by the real-time road friction state compensation coefficient. Dynamic calculations are performed to correct for deformation of the physical envelope surface under rain and snow conditions, ultimately resulting in... and exist The internal envelope forms a two-dimensional sector-shaped scanning occupancy area. .

[0036] Specifically, the process of extracting the spatiotemporal overlap features of the dual-end braking envelope in the spatial conflict zone is as follows: Using the temporal axis reference of the synchronous motor vehicle target and the electric bicycle target during trajectory extrapolation, the geometric intersection of the motor vehicle braking envelope and the electric bicycle braking envelope in the spatial conflict zone is retrieved; the overlap displacement in the spatial dimension and the duration of overlap in the time axis of the geometric intersection are extracted as spatiotemporal overlap features characterizing the degree of collision risk.

[0037] In this implementation scheme, after obtaining the dynamic braking envelope surface at both ends, the system uses a unified hardware clock source in the edge processor to forcibly synchronize the predicted time axis reference of the motor vehicle and the electric bicycle, with a time synchronization step size. Discretization is performed within the prediction domain, and the geometric intersection set for each discrete time slice is extracted, expressed by the following formula: ;in, Indicates the synchronization time The geometrically intersecting set of two-dimensional space; The rigid physical projection boundary box of a motor vehicle body is generated by expanding the one-dimensional longitudinal occupancy area of ​​the vehicle body to both sides along its standard width. Indicates that they belong to the same time period The two-dimensional sector-shaped scanning area occupied by the electric bicycle; This represents the Boolean intersection operation of spatial coordinates. Subsequently, the system quantitatively separates the spatiotemporal overlap features, and the formula for calculating the duration of overlap on the time axis is: ;in, This indicates the cumulative length of time during which there is a substantial risk of collision between the two vehicles; Indicates the threshold time for the deduction; Represents the step function of intersecting states, when the geometrically intersecting sets... The value is 1 when the spatial area is greater than zero, and 0 otherwise. The formula for calculating the overlapping displacement in the spatial dimension is: ;in, This represents the total length of slip in the potential collision physical region during the intersection; and These represent the start and end synchronization timestamps of the step function first becoming 1 and finally becoming 0, respectively. This represents the instantaneous drift velocity vector of the centroid of the geometrically intersecting set in the ground physical coordinate system. This represents the mathematical operation for calculating the vector magnitude. The derivation here involves a time threshold. A specific method for determining this is disclosed: its value is based on the shortest straight-line distance from the intersection center coordinates to the physical boundary of the visual obstruction, divided by the legally mandated maximum speed limit for that road segment, and thus accurately extracted. and This serves as an absolute physical quantitative indicator for subsequent system assessment of the severity of collision risks.

[0038] Specifically, the process of constructing a line-of-sight topology occlusion matrix based on static occlusions and dynamic target contours at intersections, and evaluating the mutual visibility state of motor vehicles and electric bicycles within spatiotemporal overlap features, is as follows: The driver's viewpoint spatial coordinates of the motor vehicle target and the overall circumscribed geometric contour of the electric bicycle target are extracted; the edges of the driver's viewpoint spatial coordinates and the overall circumscribed geometric contour are connected to generate a line-of-sight ray beam; the three-dimensional boundary information of fixed obstacles in the background environment and the three-dimensional boundary information of the dynamic target contour within continuous image frames are extracted as occlusion masks; the spatial interference distribution of the line-of-sight ray beam and the occlusion mask in the real coordinate system is calculated to generate a line-of-sight topology occlusion matrix; and the mutual visibility state between the motor vehicle target and the electric bicycle target is calibrated on the time axis section corresponding to the spatiotemporal overlap features based on the interference occlusion and transmission state within the line-of-sight topology occlusion matrix.

[0039] In this implementation, the system deploys a 3D semantic segmentation network at the edge, pre-trained on an open-source urban traffic scene dataset and processed with desensitized masks for face and license plate image regions. This network extracts the 3D boundary information of fixed obstacles and dynamic target contours within continuous image frames, and performs voxelization on this information to serve as a spatial occlusion mask. The system also extracts the driver's viewpoint spatial coordinates of the vehicle target. And extract the set of edge vertices of the overall circumscribed geometric contour of the electric bicycle target. In the real coordinate system, a line-of-sight ray is generated by connecting the driver's viewpoint spatial coordinates with the edge of the overall circumscribed geometric contour. The equation for a single line-of-sight ray is expressed as: ;in, Represents the spatial coordinate vector of the driver's viewpoint in a motor vehicle; , , These represent the horizontal, vertical, and lateral coordinate components of the viewpoint in the real coordinate system, respectively. The vector represents the coordinates of the h-th edge vertex of the overall circumscribed geometric contour of the electric bicycle; h represents a positive integer index identifier with a value from 1 to 8; g represents a linear step scalar with a value between 0 and 1. Let represent the three-dimensional spatial coordinate vector of the point corresponding to the step scalar g on the h-th line of sight. The spatial interference distribution between the line of sight beam and the blocking mask is calculated, and the transmittance equation for each ray is obtained as follows: ;in, This represents the physical transmittance of the h-th line of sight; This represents the voxel opacity density function of the occlusion mask at the corresponding 3D spatial coordinates. The view topology occlusion matrix is ​​generated by arranging the physical transmittance of all edge vertices into row vectors. On the time axis section corresponding to the spatiotemporal overlap features, based on the line-of-sight topological occlusion matrix... The numerical mean of the matrix elements is used to determine the mutual visibility status between motor vehicle and electric bicycle targets: when the algebraic mean of the matrix elements is greater than the visibility determination threshold... At that time, mutually visible state variables The value is 1, and the value is 0 otherwise. This is the visibility threshold. A method for determining this is disclosed: the edge computing unit reads the real-time simulated ambient illuminance data collected by the intersection's photoelectric sensors, and calculates the real-time fluctuation by subtracting the nonlinear attenuation compensation term corresponding to the simulated ambient illuminance data from the baseline transmittance constant for clear weather conditions. This enables the system to automatically tighten the criteria for determining blind zone transmittance in low-visibility scenarios such as nighttime or rain and snow.

[0040] Specifically, the process of fusing the mutually visible state with the double-ended braking envelope to calculate the collision approximation potential energy, assigning differentiated intervention weights, and encapsulating the vehicle alarm message and electric bicycle alarm message based on the differentiated intervention weights is as follows: Extract the approximation gradient parameters of the vehicle braking envelope and the electric bicycle braking envelope within the spatiotemporal overlap features; establish a mapping lookup association table, perform joint feature encoding mapping of the approximation gradient parameters and the mutually visible state to output the collision approximation potential energy; based on the energy level interval division results of the collision approximation potential energy, assign a first-level intervention weight to the vehicle target and a second-level intervention weight to the electric bicycle target, the first-level intervention weight and the second-level intervention weight constitute differentiated intervention weights; according to the low-power local area IoT communication protocol frame format, encapsulate the first-level intervention weights into a vehicle alarm message using channel coding, and encapsulate the second-level intervention weights into an electric bicycle alarm message using channel coding.

[0041] In this implementation scheme, the system extracts the approximation gradient parameters of the braking envelope surfaces of motor vehicles and electric bicycles within the spatiotemporal overlap features. and the relative physical distance of the geometric centroid A nonlinear joint feature encoding mapping lookup table is established to fuse the approximation gradient parameters with the mutually visible state variables calibrated in the preceding process, and the collision approximation potential energy is calculated. The joint feature encoding mapping formula is as follows: ;in, This represents the collision approximation potential energy value output by the joint feature encoding map; The approximation gradient parameter represents the boundary of the overlapping boundary of the double-ended braking envelope surface; Indicates the relative physical distance between the centroid and the ground. This represents the preset minimum safety boundary constant to prevent the denominator from approaching zero and overflowing. Indicates the sensitivity index of braking physical kinetic energy; The blind zone potential energy penalty amplification factor represents the blind zone potential energy amplification factor when the line of sight is completely blocked in a mutually visible state. This represents the mutually visible state quantities specified in the previous calibration. The braking kinetic energy sensitivity index is used here. A method for determining the physical mass of a motor vehicle is disclosed: The system parses the vehicle's outline and appearance features from the front-end video stream and retrieves the local offline vehicle curb weight database. It then extracts the corresponding physical mass benchmark value of the motor vehicle, calculates the quotient of this benchmark value and a preset standard mass constant for electric bicycles, and assigns this quotient value to... This results in a steeper nonlinear growth rate in the collision approach potential energy of large-mass vehicles. The system calculates and assigns weights based on the energy level interval division of the collision approach potential energy: the first-level intervention weight assigned to the vehicle target is... The second-level intervention weight assigned to the electric bicycle target is ;in, This indicates the first-level intervention weight; This indicates the second-level intervention weight; This represents the preset steady-state response weighting coefficients on the vehicle side; This represents the preset high-sensitivity response weight coefficient for the electric bicycle side. After completing the allocation of differentiated intervention weights, the system calls the low-power local area IoT communication protocol stack to... The media access control address of the hardware node associated with the physical lane of the target vehicle is encoded using the data link layer frame payload, and encapsulated to generate a vehicle alarm message; simultaneously, The medium access control address of the hardware node associated with the direction of travel of the electric bicycle is processed by corresponding channel coding, encapsulated to generate an electric bicycle alarm message, and sent to the physical layer radio frequency transmission queue.

[0042] Specifically, the process of splitting and sending vehicle alarm messages and electric bicycle alarm messages to the first execution unit in the direction of the vehicle and the second execution unit in the direction of the electric bicycle respectively through the edge collaborative gateway is as follows: Extract the motion state tensor heading of the vehicle target to the corresponding physical space entrance associated hardware medium access control address, and the motion state tensor heading of the electric bicycle target to the corresponding physical space entrance associated hardware medium access control address; call the radio frequency transceiver chip to establish a star-shaped spread spectrum modulation communication link, and through the star-shaped spread spectrum modulation communication link, send the vehicle alarm message to the first execution unit according to the corresponding physical space entrance associated hardware medium access control address, and send the electric bicycle alarm message to the second execution unit according to the corresponding physical space entrance associated hardware medium access control address.

[0043] In this implementation scheme, the edge collaborative gateway extracts the velocity vector components of the motor vehicle and electric bicycle targets in the multi-dimensional motion state tensor, and inverts their directions to generate a reverse trajectory vector representing the direction of arrival of the targets. The gateway traverses all pre-marked physical space entrances at the intersection and calculates the spatial alignment between the reverse trajectory vector and the preset outward normal vector of each physical space entrance. The calculation formula is as follows: ;in, This represents the spatial alignment between the target's reverse trajectory vector and the p-th physical space entrance; This represents the reverse trajectory vector obtained after reversing the instantaneous velocity direction of the target by 180 degrees; This represents the normal vector that is pre-assigned to the p-th physical space entrance and is perpendicular to the cross section of that entrance and extends outwards. This represents the magnitude of the calculated spatial vector. Gateway selection. When the maximum value is obtained, the physical space entry index corresponding to that index is retrieved from the local flash memory routing table, along with the hardware media access control address bound to that index. After addressing is complete, the underlying parameters of the star spread spectrum modulation communication link are configured using the RF transceiver chip. The formula for calculating the spread spectrum modulation symbol period is as follows: ;in, The symbol duration parameter represents a single RF modulation. The spreading factor represents the communication link; This indicates the configured physical bandwidth of the radio frequency channel. The spreading factor is disclosed here. One method for determining this is to read the offline calibration straight-line distance from the gateway to the first or second execution unit at the corresponding target physical space entrance, and then use the formula... Dynamic calculation settings are performed; among which, This indicates the actual calibrated straight-line distance between the gateway and the execution unit; Indicates the reference attenuation distance constant; This represents the signal-to-noise ratio compensation coefficient obtained based on the current frequency band noise floor. This represents the floor function. After the underlying parameters are configured, the RF transceiver chip will use the corresponding hardware media access control address to distinguish between the vehicle alarm message and the electric bicycle alarm message. Modulated transmission is performed.

[0044] Specifically, the first execution unit and the second execution unit respectively analyze the differentiated intervention weights, and the specific process of triggering the corresponding audio-visual intervention sequence according to the differentiated intervention weights is as follows: The first execution unit restores the motor vehicle alarm message through the internal baseband radio frequency decoding module to extract the first-level intervention weight within the differentiated intervention weights, and uses the pulse width modulation port to output the drive level signal to trigger the LED array connected locally in the first execution unit to generate a low-frequency flashing sequence; The second execution unit restores the electric bicycle alarm message through the internal baseband radio frequency decoding module to extract the second-level intervention weight within the differentiated intervention weights, and uses the digital output pin to drive the LED array connected locally in the second execution unit in parallel to generate a high-frequency strobe sequence and drive the electroacoustic conversion device to generate an emergency alarm audio sequence. The low-frequency flashing sequence, the high-frequency strobe sequence and the emergency alarm audio sequence are combined to form the audio-visual intervention sequence.

[0045] In this implementation scheme, the first execution unit uses its internal baseband RF decoding module to restore the received RF carrier signal to a digital baseband signal, from which the first-level intervention weight for the vehicle's direction is extracted. The microcontroller then calculates the pulse width modulation duty cycle of the LED array based on this weight, using the following formula: ;in, This indicates the pulse width modulation duty cycle output to the driver circuit; Indicates the preset hardware base driver duty cycle; This represents the duty cycle modulation gain constant; This represents the first-level intervention weight extracted from the vehicle alarm message. The first execution unit uses a timer to output a level signal according to this duty cycle, driving the local array to emit light and form a low-frequency flashing sequence. Similarly, the second execution unit decodes and restores the electric bicycle alarm message and extracts the second-level intervention weight. The high-frequency strobe light frequency and the emergency alarm audio frequency, which are controlled in parallel by its digital output pin, are calculated using the following formulas: as well as ;in, This indicates the strobe frequency of the lights output towards the electric bicycle; Indicates the maximum flicker frequency supported by the hardware; The negative feedback suppression coefficient represents the flicker frequency; This represents the second-level intervention weight extracted from the electric bicycle alarm message; Indicates the alarm audio frequency that drives the electroacoustic conversion device to produce sound; Indicates the fundamental frequency of the dynamic alarm audio; This represents the modulation stretching factor of the audio frequency. The fundamental frequency of the dynamic alarm audio is disclosed here. One method for determining this is as follows: The second execution unit acquires the background noise level of the intersection output by the microphone in the residing environment in real time through an onboard analog-to-digital converter, and uses a formula... Calculated; where, This indicates the standard penetration frequency band value specified at the factory. This represents the ambient noise level at the intersection, which is collected in real time and filtered through a low-pass filter. This represents the noise floor constant under the benchmark test environment. The second execution unit, based on the calculated frequency parameters, synchronously drives the lights and audio peripherals to synthesize a complete heterogeneous audio-visual intervention sequence for physical output.

[0046] Example 2; please refer to Figure 4An active cooperative early warning system for electric bicycles at unsignalized intersections is disclosed, used to execute an active cooperative early warning method for electric bicycles at unsignalized intersections as described in the embodiments. The system includes: a state tensor module for acquiring continuous image frames of the unsignalized intersection, tracking motor vehicle targets and electric bicycle targets, extracting pixel trajectory sequences of the motor vehicle targets and electric bicycle targets, and converting the pixel trajectory sequences into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate; and a spatiotemporal overlap module for dividing spatial conflict zones according to the physical topology of the intersection, extrapolating trajectories into the spatial conflict zones based on the multidimensional motion state tensor, mapping the braking envelope surfaces of motor vehicles and electric bicycles respectively, and extracting the spatiotemporal relationships between the braking envelope surfaces of motor vehicles and electric bicycles within the spatial conflict zones. The system includes an overlapping feature module and an alarm encapsulation module. The alarm encapsulation module constructs a line-of-sight topology occlusion matrix based on the outlines of static obstructions and dynamic targets at intersections. It assesses the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlapping features, fuses the mutual visibility state with the braking envelopes of motor vehicles and electric bicycles to calculate the collision approximation potential energy, assigns differentiated intervention weights, and encapsulates motor vehicle and electric bicycle alarm messages based on these weights. The warning execution module splits and distributes the motor vehicle and electric bicycle alarm messages to a first execution unit in the motor vehicle direction and a second execution unit in the electric bicycle direction, respectively, via an edge collaborative gateway. The first and second execution units parse the differentiated intervention weights and trigger corresponding audio-visual intervention sequences based on these weights.

[0047] In this implementation scheme, the state tensor module serves as the system's fundamental perception data processing unit, responsible for executing the image parsing and physical state quantization processes in the method steps. This module receives continuous image frames transmitted from the front-end camera device via a physical interface, runs a target tracking algorithm to continuously locate motor vehicles and electric bicycles, and records their two-dimensional coordinate change sequences to extract pixel trajectories. Subsequently, the module calls its internally configured camera calibration parameters and coordinate transformation algorithm to perform projective transformation and time-domain order difference operations on the pixel trajectories, directly outputting a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate. This completes the transformation from visual pixel stream to objective physical motion feature parameters, which are then transmitted to the downstream processing layer.

[0048] The spatiotemporal overlap module, serving as the system's collision boundary deduction unit, is responsible for executing the trajectory prediction and intersection conflict zone calculation processes in the method steps. This module receives a multidimensional motion state tensor and, combined with pre-imported high-precision vector map data of the intersection, delineates the geometric boundaries of the spatial conflict zone. Within this region, the module uses calculus to extrapolate the motion states of the motor vehicle and the electric bicycle forward according to a set time step, calculating and generating the braking envelope surfaces of the motor vehicle and the electric bicycle based on set braking deceleration parameters. Then, through Boolean intersection operations, the module compares these two braking envelope surfaces within the spatial conflict zone, extracting their overlapping data along the spatial displacement and time axis, forming an objective spatiotemporal overlap feature parameter output.

[0049] The alarm encapsulation module, serving as the system's risk decision-making and protocol packaging unit, is responsible for the blind zone transmission calculation and intervention strategy generation processes in the execution method. This module extracts the 3D boundary of fixed obstructions and the dynamic outline data of the target, generates a line-of-sight topology occlusion matrix through spatial ray interferometry, and compares the matrix transmittance to determine the mutual visibility status of the motor vehicle and electric bicycle during their intersection. After obtaining the visibility status, the module establishes a nonlinear lookup mapping association table, fuses it with the envelope surface approximation parameters in the spatiotemporal overlap features, and calculates and outputs the collision approximation potential energy value. Finally, based on the potential energy value, the module assigns asymmetric differentiated intervention weights to both ends, and strictly follows the data link layer frame format of low-power local area IoT to encode and encapsulate the weights into independent motor vehicle alarm messages and electric bicycle alarm messages.

[0050] The early warning execution module, acting as the physical layer output control unit of the system, is responsible for the targeted distribution of messages and the driving of peripheral devices in the execution method steps. The main control unit (edge ​​collaborative gateway) reads the generated alarm message, matches it with the medium access control address of the corresponding physical direction hardware, establishes a spread spectrum modulation link through the RF transceiver chip, and completes the targeted message distribution. After receiving the RF carrier, the first execution unit deployed in the motor vehicle lane and the second execution unit deployed in the electric bicycle lane reconstruct the data frame through their internal baseband decoding module, extracting the specific differentiated intervention weights. The main control microcontroller of each execution unit then calculates the corresponding pulse width duty cycle and output frequency based on the extracted weight values, controls the pin level changes in parallel, drives the external LED array and speaker to emit sound, and outputs the corresponding audio-visual intervention sequence.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0056] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for active cooperative warning of an electric bicycle at a signal-free intersection, characterized in that, Includes the following steps: S1. Acquire continuous image frames of unsignalized intersections, track motor vehicle targets and electric bicycle targets, extract pixel trajectory sequences of motor vehicle targets and electric bicycle targets, and convert the pixel trajectory sequences into multidimensional motion state tensors containing instantaneous velocity, acceleration and heading deflection rate. S2. Based on the physical topology of the intersection, the spatial conflict zone is divided. Based on the multidimensional motion state tensor, the trajectory is extrapolated into the spatial conflict zone. The braking envelope of motor vehicles and the braking envelope of electric bicycles are mapped respectively. The spatiotemporal overlap features of the double-ended braking envelope in the spatial conflict zone are extracted. S3. Construct a line-of-sight topology occlusion matrix based on the static occlusions and dynamic target contours at the intersection, evaluate the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlap features, fuse the mutual visibility state with the double-ended braking envelope surface to calculate the collision approximation potential energy, assign differentiated intervention weights, and encapsulate motor vehicle alarm messages and electric bicycle alarm messages according to the differentiated intervention weights. S4. The alarm messages for motor vehicles and electric bicycles are split and sent to the first execution unit in the direction of motor vehicles and the second execution unit in the direction of electric bicycles through the edge collaborative gateway. The first execution unit and the second execution unit respectively parse the differentiated intervention weights and trigger the corresponding sound and light intervention sequence according to the differentiated intervention weights.

2. The active cooperative pre-warning method for electric bicycles at signal-free intersections according to claim 1, characterized in that: The specific process of acquiring continuous image frames at an unsignalized intersection, tracking motor vehicle targets and electric bicycle targets, and extracting the pixel trajectory sequences of the motor vehicle targets and electric bicycle targets is as follows: Identify feature anchor points of motor vehicle targets and electric bicycle targets within consecutive image frames, and establish the association chain between anchor points through time-domain feature matching; The displacement path of the target's center of motion in the image coordinate system is fitted to form a pixel trajectory sequence that evolves with the image frame sequence.

3. The active cooperative early warning method for electric bicycles at unsignalized intersections according to claim 2, characterized in that: The specific process of converting a pixel trajectory sequence into a multidimensional motion state tensor containing instantaneous velocity, acceleration, and heading deflection rate is as follows: By establishing a projective transformation matrix between the image coordinate system and the real coordinate system through camera calibration parameters, the pixel trajectory sequence is mapped to a position vector sequence in the real coordinate system. The instantaneous velocity and acceleration are extracted by performing order difference processing on the position vector sequence, and the heading deflection rate is extracted based on the rate of change of the tangent azimuth angle of the position vector sequence in the horizontal plane. These are then encapsulated to form a multidimensional motion state tensor.

4. The active cooperative pre-warning method for electric bicycles at signal-free intersections according to claim 1, characterized in that: Based on the physical topology of the intersection, spatial conflict zones are divided. Trajectory extrapolation is performed into the spatial conflict zones based on the multidimensional motion state tensor, and the specific process of mapping the braking envelope surfaces of motor vehicles and electric bicycles is as follows: The geometric topological boundary of the intersection center area is defined to form a spatial conflict zone. The vehicle braking envelope is represented by a one-dimensional longitudinal occupancy interval generated by mapping the instantaneous velocity and predetermined braking acceleration of the vehicle target within the spatial conflict zone. The braking envelope of an electric bicycle is characterized by a two-dimensional sector-shaped scanning occupancy region generated by mapping the heading deflection rate and lateral displacement trend of the electric bicycle target within the spatial conflict zone.

5. The active cooperative pre-warning method for electric bicycles at signal-free intersections according to claim 1, characterized in that: The specific process for extracting the spatiotemporal overlap features of the double-ended braking envelope surface within the spatial conflict zone is as follows: Synchronize the time axis references of motor vehicle targets and electric bicycle targets during trajectory extrapolation, and retrieve the geometric intersection of the braking envelope surfaces of motor vehicles and electric bicycles within the spatial conflict zone; The overlapping displacement of the geometrically intersecting parts in the spatial dimension and the duration of overlap in the time axis are extracted as spatiotemporal overlapping features to characterize the degree of collision risk.

6. The active cooperative pre-warning method for electric bicycles at signal-free intersections according to claim 1, characterized in that: The specific process of constructing a line-of-sight topology occlusion matrix based on static occlusions and dynamic target contours at intersections, and evaluating the mutual visibility of motor vehicles and electric bicycles within spatiotemporal overlapping features, is as follows: Extract the driver's viewpoint spatial coordinates of the motor vehicle target and the overall external geometric contour of the electric bicycle target, and connect the driver's viewpoint spatial coordinates and the edge of the overall external geometric contour to generate a line-of-sight ray beam; Extract the 3D boundary information of fixed obstacles in the background environment and the 3D boundary information of dynamic target contours within continuous image frames as occlusion masks; The spatial interference distribution of the line-of-sight ray beam and the occlusion mask in the real coordinate system is calculated to generate the line-of-sight topology occlusion matrix. On the time axis section corresponding to the spatiotemporal overlap features, the mutual visibility state between motor vehicle targets and electric bicycle targets is calibrated based on the interference occlusion and transmission state within the line-of-sight topology occlusion matrix.

7. The active cooperative pre-warning method for signal-free intersection of electric bicycle according to claim 1, characterized in that: The process of fusing the mutually visible states with the double-ended braking envelope to calculate the collision approximation potential energy, assigning differentiated intervention weights, and encapsulating the motor vehicle alarm message and the electric bicycle alarm message based on the differentiated intervention weights is as follows: Extract the approximation gradient parameters of the braking envelope of motor vehicles and the braking envelope of electric bicycles within the spatiotemporal overlap features; Establish a mapping lookup association table, perform joint feature encoding mapping between the approximation gradient parameters and mutually visible states, and output the collision approximation potential energy. Based on the energy level range division of the collision approach potential energy, a first-level intervention weight is assigned to the motor vehicle target, and a second-level intervention weight is assigned to the electric bicycle target. The first-level intervention weight and the second-level intervention weight constitute differentiated intervention weights. According to the low-power local IoT communication protocol frame format, the first-level intervention weights are channel-coded and encapsulated into motor vehicle alarm messages, and the second-level intervention weights are channel-coded and encapsulated into electric bicycle alarm messages.

8. The active cooperative pre-warning method for signal-free intersection of electric bicycle according to claim 1, characterized in that: The specific process of using an edge collaborative gateway to separately split and distribute vehicle alarm messages and electric bicycle alarm messages to the first execution unit in the direction of vehicles and the second execution unit in the direction of electric bicycles is as follows: Extract the motion state tensor heading of the motor vehicle target to the corresponding physical space entrance associated hardware medium access control address, and extract the motion state tensor heading of the electric bicycle target to the corresponding physical space entrance associated hardware medium access control address. A star-shaped spread spectrum modulation communication link is established by calling the radio frequency transceiver chip. Through the star-shaped spread spectrum modulation communication link, the motor vehicle alarm message is routed and sent to the first execution unit according to the corresponding physical space entry associated hardware medium access control address, and the electric bicycle alarm message is routed and sent to the second execution unit according to the corresponding physical space entry associated hardware medium access control address.

9. The active cooperative pre-warning method for signal-free intersection of electric bicycle according to claim 1, characterized in that: The first execution unit and the second execution unit respectively analyze the differentiated intervention weights, and the specific process of triggering the corresponding audio-visual intervention sequence based on the differentiated intervention weights is as follows: The first execution unit restores the vehicle alarm message through the internal baseband radio frequency decoding module, extracts the first-level intervention weight within the differentiated intervention weight, and outputs a drive level signal through the pulse width modulation port to trigger the local LED array of the first execution unit to generate a low-frequency flashing sequence. The second execution unit restores the electric bicycle alarm message through the internal baseband RF decoding module, extracts the second-level intervention weight within the differentiated intervention weight, and uses the digital output pin to drive the locally connected light-emitting diode array of the second execution unit in parallel to generate a high-frequency flashing sequence and drive the electroacoustic conversion device to generate an emergency alarm audio sequence. The low-frequency flashing sequence, the high-frequency flashing sequence and the emergency alarm audio sequence are combined to form an audio-visual intervention sequence.

10. A signal-free intersection e-bike active cooperative warning system for performing the signal-free intersection e-bike active cooperative warning method of any one of claims 1-9, characterized in that, include: The state tensor module is used to acquire continuous image frames of unsignalized intersections, track motor vehicle targets and electric bicycle targets, extract pixel trajectory sequences of motor vehicle targets and electric bicycle targets, and convert the pixel trajectory sequences into a multi-dimensional motion state tensor containing instantaneous velocity, acceleration and heading deflection rate. The spatiotemporal overlap module is used to divide the spatial conflict zone according to the physical topology of the intersection, extrapolate the trajectory into the spatial conflict zone based on the multidimensional motion state tensor, map the braking envelope of motor vehicles and the braking envelope of electric bicycles respectively, and extract the spatiotemporal overlap features of the braking envelope of motor vehicles and the braking envelope of electric bicycles in the spatial conflict zone. The alarm encapsulation module is used to construct a line-of-sight topology occlusion matrix based on the outlines of static occluders and dynamic targets at intersections, evaluate the mutual visibility state of motor vehicles and electric bicycles within the spatiotemporal overlap features, fuse the mutual visibility state with the braking envelope of motor vehicles and the braking envelope of electric bicycles to calculate the collision approximation potential energy, assign differentiated intervention weights, and encapsulate motor vehicle alarm messages and electric bicycle alarm messages according to the differentiated intervention weights. The early warning execution module is used to split and send motor vehicle alarm messages and electric bicycle alarm messages to the first execution unit in the direction of motor vehicles and the second execution unit in the direction of electric bicycles through the edge collaborative gateway. The first execution unit and the second execution unit respectively parse the differentiated intervention weights and trigger the corresponding sound and light intervention sequences according to the differentiated intervention weights.

Images