Road intersection vulnerable group type classification method and system facing road surface disturbance

Abstract

The present disclosure provides a road intersection vulnerable group type classification method and system facing road surface disturbance, relates to the technical field of urban road traffic operation monitoring, and comprises the following steps: a distributed optical fiber structure observation network is arranged in a target monitoring area, time-space vibration signals are acquired, and the time-space vibration signals are preprocessed; event trigger detection is carried out based on the preprocessed time-space vibration signals, event segments are clustered and separated under mixed traffic conditions, and a single-target event sequence is obtained; structural domain features, rhythm domain features and energy domain features are extracted from the structural domain, the rhythm domain and the energy domain respectively for the single-target event sequence; the structural domain features, the rhythm domain features and the energy domain features are fused to obtain a fusion feature vector; and the fusion feature vector is used to calculate posterior probability on an extensible category set, and a finally recognized vulnerable group category is output. The present disclosure can realize accurate discrimination of traffic disturbance under a complex intersection environment.

Description

Classification Method and System for Vulnerable Groups at Road Intersections Facing Road Disturbance Technical Field

[0001] This disclosure relates to the field of urban road traffic operation monitoring technology, specifically to a classification method and system for vulnerable groups at road intersections facing road surface disturbances. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] Urban road intersections are typical scenarios where pedestrians, non-motorized vehicles, and motorized vehicles intertwine. Traffic participants employ diverse modes of transportation, their operational status changes rapidly, and they are affected by factors such as obstructed visibility, uncertain yielding rules, and complex traffic organization. Vulnerable traffic groups face higher exposure risks in these scenarios. To support traffic safety governance and refined management at intersections, continuous monitoring of vulnerable traffic groups is typically necessary, along with obtaining their specific characteristics. This information can then be used for applications such as traffic flow characteristic statistics, conflict risk identification, yielding pattern analysis, and evaluation of governance measures.

[0004] Existing traffic sensing methods at intersections primarily employ video, millimeter-wave radar, geomagnetic sensors, and coils for target detection and classification. Among these, video solutions are susceptible to variations in lighting, rain, snow, fog, and occlusion interference. In densely populated conditions with mixed pedestrian and vehicle traffic, target occlusion and tracking loss are common, and high maintenance costs are also a concern. Millimeter-wave radar provides relatively stable speed data acquisition, but its fine-grained differentiation of vulnerable traffic groups relies on target scattering characteristics and trajectory stability, making it prone to category confusion in low-speed, parallel, and intersecting traffic situations. Geomagnetic sensors and coils are more suitable for vehicle detection, but their adaptability to the perception and classification of vulnerable traffic groups such as pedestrians and non-motorized vehicles is insufficient, making it difficult to meet the requirements of all-weather, wide-coverage, and stable classification in the complex environment of intersections.

[0005] Distributed fiber optic sensing technology possesses advantages such as resistance to electromagnetic interference, long-distance continuous monitoring, covert deployment, and environmental tolerance, providing a new approach to urban road traffic operation sensing. However, existing traffic sensing applications based on distributed fiber optics still have the following limitations:

[0006] (1) It focuses on extracting coarse-grained information such as whether there is a target passing through, speed or flow. If it relies on a single feature such as speed or energy threshold for discrimination, typical confusion may occur: for example, running pedestrians and low-speed cycling non-motorized vehicles may overlap in speed range; two-wheeled and three-wheeled categories also lack stable separability in speed.

[0007] (2) The signal superposition caused by multiple targets passing in parallel or crossing will further increase the difficulty of target separation and type identification. Therefore, under the condition that vibration and disturbance signals are the main observation quantities in distributed optical fiber, there is still a lack of an engineering-feasible solution for how to achieve scalable, interpretable and robust type identification and classification of vulnerable traffic groups in the intersection scenario. Summary of the Invention

[0008] To address the aforementioned issues, this disclosure proposes a classification method and system for vulnerable groups at road intersections oriented towards road surface disturbances. It constructs a distributed optical fiber structure observation system, and based on the observation mechanism of road surface disturbances using distributed optical fibers, it suppresses the influence of non-target factors on signal characteristics. By employing a multi-domain joint representation of the structural domain, rhythm domain, and energy domain, it significantly reduces category confusion caused by overlapping speed ranges between different groups, thereby achieving accurate identification and classification of vulnerable group types at road intersections.

[0009] According to some embodiments, the present disclosure adopts the following technical solutions:

[0010] Classification methods for vulnerable groups at road intersections facing road surface disturbances include:

[0011] A distributed optical fiber structure observation network was deployed in the target monitoring area and spatiotemporal calibration was performed.

[0012] Acquire spatiotemporal vibration signals collected by a distributed optical fiber structure observation network and preprocess them;

[0013] Event trigger detection is performed based on preprocessed spatiotemporal vibration signals, and event segments are clustered and separated under mixed conditions to obtain single-target event sequences.

[0014] For a single-target event sequence, structural domain features, rhythm domain features, and energy domain features are extracted from the structural domain, rhythm domain, and energy domain, respectively. Multi-domain feature fusion is performed on the structural domain features, rhythm domain features, and energy domain features to obtain a fused feature vector.

[0015] Based on the fused feature vectors, the posterior probability is calculated on an expandable set of categories, and the final identified category of the vulnerable group is obtained.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions:

[0017] A classification system for vulnerable groups at road intersections oriented towards road surface disturbance includes:

[0018] The channel deployment module is used to deploy a distributed optical fiber structure observation network in the target monitoring area and perform spatiotemporal calibration.

[0019] The signal acquisition module is used to acquire spatiotemporal vibration signals collected by the distributed optical fiber structure observation network and to preprocess them.

[0020] The target separation module is used to perform event trigger detection based on preprocessed spatiotemporal vibration signals, and to cluster and separate event segments under mixed conditions to obtain single-target event sequences.

[0021] The feature extraction and fusion module is used to extract structural domain features, rhythm domain features, and energy domain features from the structural domain, rhythm domain, and energy domain respectively from a single target event sequence, and to perform multi-domain feature fusion on the structural domain features, rhythm domain features, and energy domain features to obtain a fused feature vector.

[0022] The classification module is used to calculate the posterior probability on a scalable set of categories based on the fused feature vectors, and outputs the final identified category of the vulnerable group.

[0023] According to some embodiments, the present disclosure adopts the following technical solutions:

[0024] A computer program product includes a computer program that, when executed by a processor, implements the aforementioned method for classifying vulnerable groups at road intersections oriented towards road surface disturbance.

[0025] According to some embodiments, the present disclosure adopts the following technical solutions:

[0026] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned method for classifying vulnerable groups at road intersections to address road surface disturbances.

[0027] According to some embodiments, the present disclosure adopts the following technical solutions:

[0028] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for classifying vulnerable groups at road intersections to address road surface disturbances.

[0029] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0030] This disclosed method for classifying vulnerable traffic groups at road intersections based on road surface disturbances addresses the complex mixed-traffic scenarios of vulnerable traffic groups at urban road intersections. It constructs a distributed fiber optic observation system consisting of a "main monitoring channel and a gate channel," and establishes a unified spatiotemporal benchmark through coupling consistency constraints and mileage-road coordinate mapping. This method enables continuous acquisition and event localization of traffic disturbances in complex intersection environments, providing comparable and reusable data input for subsequent type identification.

[0031] The classification method for vulnerable groups at road intersections oriented to road surface disturbances disclosed herein, compared to visual perception schemes that are susceptible to the effects of light, shading, rain, and fog, is based on the observation mechanism of road surface disturbances through distributed optical fibers. Combined with multi-scale noise reduction, narrowband suppression, and background baseline recursion, this method can effectively suppress the influence of non-target factors such as wind, rain, far-field micro-vibrations, and narrowband interference on signal characteristics, and helps to improve the stability and maintainability of long-term operation in complex intersection scenarios.

[0032] The disclosed method for classifying vulnerable groups at road intersections with road surface disturbances uses spatiotemporal energy statistics and dynamic thresholds to trigger traffic events, and constructs separation clues using information such as gate passage delay and speed constraints. Under the condition of signal superposition caused by parallel and cross traffic of pedestrians and non-motorized vehicles, it can segment mixed events into single-target event sequences, thereby reducing the risk of missed detection, false detection and misjudgment caused by multi-target aliasing.

[0033] The disclosed classification method for vulnerable groups at road intersections oriented to road surface disturbance adopts a multi-domain joint representation of structural domain, rhythmic domain, and energy domain: the structural domain is used to characterize the differences in wheel structure and contact geometry, the rhythmic domain is used to characterize the differences in gait period and stability, and the energy domain is used to characterize the differences in the time-frequency distribution of impact and rolling. The fusion of multi-domain features makes the classification independent of a single speed index, and can significantly reduce the category confusion caused by the overlap of speed ranges between jogging pedestrians and low-speed non-motorized vehicles, and between vehicles with different wheel sets.

[0034] This disclosed method for classifying vulnerable groups at road intersections to address road surface disturbances performs posterior probability decisions on an scalable set of categories, outputs target category labels and corresponding confidence levels, and supports unknown, pending confirmation outputs or triggers calibration mechanisms when confidence levels are insufficient. This improves the interpretability and reliability of the results from the output end, making it easier to directly serve applications such as statistics on vulnerable traffic groups at intersections, operational characteristic analysis, conflict risk assessment, and governance effectiveness evaluation. Attached Figure Description

[0035] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0036] Figure 1 is a flowchart of a method for classifying vulnerable groups at road intersections oriented to road surface disturbance, according to an embodiment of the present disclosure.

[0037] Figure 2 is a schematic diagram of the layout of the distributed optical fiber main monitoring channel and the gate line channel at an urban road intersection according to an embodiment of this disclosure. Detailed Implementation

[0038] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0040] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] Terminology Explanation

[0042] Coupling consistency control and coupling coefficient: refers to the normalization and comparison of the response amplitude at different mileage locations along the distributed optical fiber under the reference excitation, constructing a dimensionless index coupling coefficient that characterizes the coupling strength between the optical fiber and the measured medium and the consistency along the line, and using this coefficient as the control quantity for channel consistency constraints and calibration.

[0043] Mileage to road coordinate mapping function: refers to the function relationship obtained by fitting the fiber optic mileage coordinates to the road plane coordinates using the fiber optic mileage and the corresponding road plane coordinates of the calibration point as samples. It is used to realize a unified conversion benchmark from the spatiotemporal signal position of mileage to the geometric coordinates of the road.

[0044] The GCC-PHAT method for estimating gate passage delay refers to constructing a generalized cross-correlation based on the frequency domain representation of adjacent gate channel signals, and normalizing the cross-power spectrum using a phase transformation weighting method. The time offset corresponding to the cross-correlation peak is used as the estimated value of the passage delay of adjacent gates.

[0045] Maximum likelihood clustering for multi-objective separation: Under the mixed condition of superimposed event fragments, the feature vectors of the fragments are used as observation samples to establish a mixed probability model, and the model parameters and sample assignments are estimated by the maximum likelihood criterion, thereby achieving clustering and segmentation of mixed event fragments to obtain single-objective event sequences.

[0046] Quantization of the structural domain, rhythm domain, and energy domain: This refers to taking a single-target passage event sequence as the processing object, and then parameterizing the gate response morphology and wheel assembly contact differences in the structural domain, quantifying the periodicity and stability of gait or wheel assembly excitation in the rhythm domain, and quantifying statistical quantities such as time-frequency energy distribution and frequency band energy proportion in the energy domain. The quantization results of the three domains together constitute the fusion feature input for type recognition and classification decision.

[0047] Vulnerable Road Users (VRUs) are road users who lack structural protection and are more likely to be injured or suffer more serious consequences in traffic conflicts due to limited physical / cognitive abilities or high exposure. They typically include pedestrians, non-motorized vehicle riders (bicycles / electric bicycles, etc.) and some micro-mobility users, with children and the elderly being more typical representatives.

[0048] Example 1

[0049] One embodiment of this disclosure provides a method for classifying vulnerable groups at road intersections prone to road disturbances, the method comprising the following steps:

[0050] Step 1: Deploy a distributed fiber optic observation network in the target monitoring area and perform spatiotemporal calibration;

[0051] Step 2: Acquire the spatiotemporal vibration signals collected by the distributed optical fiber structure observation network and preprocess them;

[0052] Step 3: Based on the preprocessed spatiotemporal vibration signal, perform event trigger detection, and cluster and separate event segments under mixed conditions to obtain a single-target event sequence;

[0053] Step 4: Extract structural domain features, rhythm domain features, and energy domain features from the structural domain, rhythm domain, and energy domain respectively for a single target event sequence. Perform multi-domain feature fusion on the structural domain features, rhythm domain features, and energy domain features to obtain a fused feature vector.

[0054] Step 5: Calculate the posterior probability on the scalable category set based on the fused feature vector, and output the final identified vulnerable group category.

[0055] As one embodiment, this disclosure presents a classification method for vulnerable groups at road intersections oriented towards road surface disturbances. Addressing the complex mixed-traffic scenarios of vulnerable groups at urban road intersections, it triggers traffic events through spatiotemporal energy statistics and dynamic thresholds, and constructs separation cues using information such as gate passage delay and speed constraints. Under conditions of signal superposition caused by pedestrians and non-motorized vehicles traveling side-by-side and crossing, it can segment mixed events into single-target event sequences, thereby reducing the risk of missed detections, false detections, and misjudgments caused by multi-target aliasing. Employing a multi-domain joint representation of structural, rhythmic, and energy domains significantly reduces category confusion caused by overlapping speed ranges between running pedestrians and low-speed non-motorized vehicles, and vehicles with different wheel sets. Furthermore, it performs posterior probability decision-making on an expandable category set, outputting the target category label and corresponding confidence level. The specific implementation process is as follows:

[0056] Step 1: Deploy a distributed fiber optic observation network in the target monitoring area and perform spatiotemporal calibration;

[0057] In this step, main monitoring channels along the direction of traffic are set up in areas such as zebra crossings and non-motorized vehicle lanes, and transverse gate channels are set up to form a structural observation network;

[0058] By establishing a unified spatiotemporal benchmark through coupled consistency control, gate spacing design, and mileage-to-road coordinate mapping, spatiotemporal calibration is achieved, providing comparable input for event localization, wheel assembly structure, and gait rhythm discrimination. Specific steps include:

[0059] Step 11: Coupling consistency control, calculate channel coupling coefficient:

[0060]

[0061]

[0062] in, mileage along the optical fiber The raw signal collected at the location; For normalized signals; For position The amplitude of the response to the reference stimulus; The reference segment response amplitude; is the coupling coefficient.

[0063] Step 12: Door line spacing design, determined by separable constraints at maximum speed:

[0064]

[0065]

[0066] in, The distance between adjacent door lines; The preset maximum throughput speed; The minimum separable time interval; The minimum number of separable sampling points; The sampling frequency.

[0067] Step 13: Spatiotemporal calibration, fitting the mileage-to-road coordinate mapping function:

[0068]

[0069] in, For the first Fiber optic mileage at each calibration point; For the corresponding road plane coordinates; This is a function that maps mileage to coordinates. The function to be estimated; The number of calibration points.

[0070] As an example, Figure 2 illustrates the channel layout and parameter settings of a distributed optical fiber structure observation network. The structure observation network includes a main monitoring channel and gate line channels. The main monitoring channel is deployed along the target's travel direction to continuously acquire the spatiotemporal vibration signals of the target within the monitoring area. The gate line channels (2a, 2b, and 2c as shown in Figure 2) are deployed in a direction approximately perpendicular to the travel direction to form a gate line corridor and provide arrival time information for the target passing through the gate lines. The spacing between adjacent gate line channels is denoted as d, which is used to constrain the time delay estimation window and, together with the arrival time difference, calculate the target's passing speed.

[0071] In this embodiment, a set of calibration points {P} for mileage-coordinate mapping is constructed. i (s i ,x i ,y i )}, where s i To determine the mileage of the calibration point on the optical fiber, (x i ,y i Let f(s) represent the position coordinates of the calibration point in the road plane coordinate system. A mapping function f(s) is obtained by fitting the calibration point set, thereby realizing the mapping from fiber optic mileage to road spatial location, which is used to locate detected event segments in road space.

[0072] For example, a zebra crossing at an intersection is selected as the target monitoring area. The zebra crossing is 12 m long, and the road width is 3.5 m. Gate access channels 2a, 2b, and 2c are installed at the entrance, middle, and exit of the monitoring area, respectively, with a spacing of d = 1.5 m between adjacent gate access channels. The main monitoring channel runs through the entire monitoring area. Let the distributed optical fiber sampling frequency be f. s =1000 Hz, preset target maximum speed v max =8m / s, then the gate channel spacing d and the sampling frequency f s This method ensures stable estimation of the passage delay of adjacent gate lines within the allowable delay range, and calculates the target speed accordingly to assist in event segment clustering and separation under mixed traffic conditions. Furthermore, at least three calibration points P1, P2, and P3 are selected (located near the entrance, gate line 2b, and the exit area, respectively), and the fiber optic mileage s at each calibration point is measured. i and road plane coordinates (x i ,y i By fitting the data, f(s) is obtained, thereby realizing the spatial localization of a single-target event sequence and subsequent type classification output.

[0073] Step 2: Acquire the spatiotemporal vibration signals collected by the distributed optical fiber structure observation network and preprocess them;

[0074] In this step of the disclosure, two-dimensional mileage-time signals from the main monitoring channel and the gate channel are acquired. Multi-scale noise reduction, narrowband interference suppression, and background baseline recursive updates are performed on the signals to ensure feature integrity and threshold stability. Quality indicators such as signal-to-noise ratio are calculated to suppress environmental disturbances and constrain subsequent fusion weights and decision confidence. Specific steps include:

[0075] Step 21: Multi-scale noise reduction, wavelet threshold shrinkage to suppress random noise:

[0076]

[0077] in, These are wavelet coefficients; The coefficient after threshold shrinkage; For the first Scale threshold.

[0078] Step 22: Narrowband suppression, band-stop filtering to eliminate narrowband high-frequency interference:

[0079]

[0080] in, for Fourier transform; This is the filtered spectrum; For the frequency response of the band-stop filter; by Inverse transform yields the filtered time-domain signal .

[0081] Step 23: Baseline recursive update, recursively updating the background mean and variance baselines:

[0082]

[0083]

[0084] in, , The mean and variance of the background window samples; Forgetting factor; The baseline is the background mean. The background variance baseline.

[0085] Step 24: Signal quality evaluation, calculating the signal-to-noise ratio for confidence modulation:

[0086]

[0087] in, For event window power, Power for the background window; This refers to the signal-to-noise ratio.

[0088] Step 3: Based on the preprocessed spatiotemporal vibration signal, perform event trigger detection, and cluster and separate event segments under mixed conditions to obtain a single-target event sequence;

[0089] This disclosure performs event-triggered detection and separation of multiple targets, forming single-target events and gate passage sequences. Energy statistics are constructed on the preprocessed signal, and passage events are triggered based on dynamic thresholds. Initial velocity is obtained by estimating passage delay using the gate channel. Under mixed-traffic conditions, event segments are clustered and separated to obtain single-target event sequences, providing input for subsequent multi-domain feature construction. Specific steps include:

[0090] Step 31: Event-triggered detection, constructing energy statistics and dynamic thresholds:

[0091]

[0092]

[0093] in, It is an energy statistic; A set of spatial indexes for the monitoring area; Number of channels in the monitoring area; , These are the background mean baseline and the background variance baseline, respectively. For threshold coefficient; when Events are triggered at certain times; For dynamic thresholds; This is the filtered time-domain signal.

[0094] Step 32: Door line latency estimation, using GCC-PHAT to improve robustness:

[0095]

[0096] in, The time delay between two adjacent gate lines; This is the frequency domain representation of the gate line signal.

[0097] Step 33: Initial velocity estimation, calculated from gate spacing and passage time delay:

[0098]

[0099] in, For initial velocity estimation; This refers to the spacing between the door lines; To allow for time delay.

[0100] Step 34: Multi-objective separation, obtaining single-objective event sequences through maximum likelihood clustering:

[0101] The result obtained in step 33 Construct fragment feature vectors from event fragments Assuming these fragments are made by A target is generated, and a Gaussian mixture model is established:

[0102]

[0103] Clustering is performed by maximizing the log-likelihood function:

[0104]

[0105] in, For segment feature vectors; The target number; For parameters of the hybrid model; The number of fragments; after separation, a single-target event sequence is obtained. .

[0106] Step 4: Extract structural domain features, rhythm domain features, and energy domain features from the structural domain, rhythm domain, and energy domain respectively for a single target event sequence. Perform multi-domain feature fusion on the structural domain features, rhythm domain features, and energy domain features to obtain a fused feature vector.

[0107] This disclosure quantifies target types from the structural, rhythmic, and energy domains respectively, forming a fused feature vector. The structural domain characterizes the wheel assembly structure and contact geometry; the rhythmic domain characterizes gait period, stride length, and stability; and the energy domain characterizes time-frequency energy distribution differences. These three domain features together constitute the fused feature vector F, used to distinguish different types of vulnerable traffic groups and support category expansion. Specific steps include:

[0108] Step 41: Velocity fine estimation, least squares fitting of arrival times for multiple gate lines:

[0109]

[0110] in, For precise speed estimation; The number of gate lines involved in the fitting; The arrival time; The coordinate distance is the gate line coordinate distance, and its coordinate mapping method is shown in step 13.

[0111] Step 42: Extract structural domain features and calculate the number of gate line main peaks and the full width at half maximum (FWHM) of the combined peaks:

[0112]

[0113]

[0114] in, This refers to the number of peaks at the gate line. It is a set of local maxima; For peak significance; For threshold; The width and height are equal to half the height, satisfying the following conditions. , The peak value is the main peak value.

[0115] Step 43: Extract rhythm domain features, calculate step frequency, and derive stride length and stability:

[0116]

[0117]

[0118]

[0119] in, It is a single-target event sequence; The search interval for gait cycles; This refers to the step frequency.

[0120] Further define stride length:

[0121]

[0122] in, Indicates precise speed estimation; Indicates step frequency.

[0123] Further define rhythm stability as:

[0124]

[0125] in, It is a sliding window step frequency sequence; For rhythm stability.

[0126] Step 44: Extract energy domain features and calculate the time-frequency spectrum and frequency band energy ratio:

[0127]

[0128]

[0129] in, This is the short-time Fourier transform spectrum; For window functions; For the first One frequency band; This represents the percentage of energy in the frequency band.

[0130] Step 45: Construct multi-domain feature vectors and fuse them as classification input:

[0131]

[0132]

[0133]

[0134] in, Features of structural domains; It is a characteristic of the rhythmic domain; This is a characteristic of the energy domain.

[0135] Furthermore, adaptive fusion weights are defined for multi-domain fusion, and a fused feature vector is constructed as the classification input:

[0136]

[0137]

[0138] in, To The result after standardization; For the quality indicators of this domain (which can be derived from...) (Construction of peak properties, etc.) The modulation coefficient; For weighting; This is the final fused feature vector.

[0139] Step 5: Calculate the posterior probability on the scalable category set based on the fused feature vector, and output the final identified vulnerable group category.

[0140] This disclosure calculates the posterior probability and outputs the target class and confidence level, supporting unknown classes and online calibration. Based on the fused feature vector F, it calculates the posterior probability on a scalable class set C, outputting the final class y and confidence level c. When the decision confidence level does not meet preset requirements, it outputs an unknown, pending confirmation class, or triggers rule constraint verification to improve the reliability and engineering usability of the output results. Specific steps include:

[0141] Step 51: Define the category set and construct an scalable category space for vulnerable transportation groups:

[0142]

[0143] in, It is a set of vulnerable transportation groups, including at least pedestrians (youth / elderly / children, etc.), bicycles, electric non-motorized vehicles, tricycles, wheelchairs / assisted vehicles, etc., and allows for the expansion and addition of new categories.

[0144] Step 52: Calculate the posterior probability and output the probability of each class based on Softmax:

[0145]

[0146] in, For class posterior probability; To fuse feature vectors, For category parameters. W q b is the weight vector corresponding to the q-th class; qThis is the bias term corresponding to the q-th class.

[0147] Step 53: Output the categories and determine the final categories and confidence levels:

[0148]

[0149]

[0150] in, For the final output category; To output the confidence level; denoted as the posterior probability of the category.

[0151] Step 54: Unknown Class Rules and Online Calibration Updates:

[0152]

[0153]

[0154] in, The confidence threshold; Category unknown / to be confirmed; For the first Sub-event fusion characteristics; Online estimation of characteristic statistics; This refers to the update rate.

[0155] Example 2

[0156] One embodiment of this disclosure provides a method for classifying vulnerable groups at road intersections prone to road disturbance. Taking a city road intersection in a certain urban area as an example, the zebra crossing length... non-motorized vehicle lane width One main monitoring channel and two gate line channels are deployed; sampling frequency The specific implementation process of this disclosed method is as follows:

[0157] Step 1: Deploy distributed fiber optic main monitoring channels and gate line channels:

[0158] Take the spacing between adjacent door lines This meets the requirement that the time delay of high-speed targets can be separated.

[0159] Step 2: Signal Acquisition

[0160] Vibration signals are collected using a distributed fiber optic sensing system. sampling frequency This indicates that continuous spatiotemporal sequence data can be obtained.

[0161] Step 3: Preprocessing and baseline estimation:

[0162] Take the background mean Standard deviation The noise reduction and background baseline are completed, indicating that the threshold can be adaptively set according to the environment.

[0163] Step 4: Calculate the event trigger threshold :

[0164] Take the threshold coefficient , This indicates that the event trigger threshold has been determined.

[0165] Step 5: Passage Event Determination:

[0166] Calculate peak energy , measured This indicates the existence of a passage event and the process has entered the classification process.

[0167] Step 6: Calculate the speed through :

[0168] Gate arrival delay measured , This indicates that the target speed is within the range where fast pedestrians / slow cyclists may overlap.

[0169] Step 7: Calculate structural features :

[0170] Gate line main peak number measured This indicates that the wheel set structure characteristics are more consistent with two-wheeled vehicles than with gait pulse trains.

[0171] Step 8: Calculate rhythmic characteristics :

[0172] No stable gait period was detected. This indicates that the rhythm domain does not support pedestrian gait features.

[0173] Step 9: Calculate energy characteristics :

[0174] Calculate the high-frequency energy ratio and measure This indicates that the event energy distribution is closer to the characteristics of rolling / mechanical excitation.

[0175] Step 10: Output the category and confidence level:

[0176] Constructing fusion features The Softmax output corresponds to the maximum posterior probability of a two-wheeled vehicle, and the confidence level is measured. This indicates that the final output category is two-wheeled vehicle; if a confidence threshold is taken... ,but It is not marked as pending confirmation.

[0177] Example 3

[0178] One embodiment of this disclosure provides a classification system for vulnerable groups at road intersections oriented towards road surface disturbance, including:

[0179] The channel deployment module is used to deploy a distributed optical fiber structure observation network in the target monitoring area and perform spatiotemporal calibration.

[0180] The signal acquisition module is used to acquire spatiotemporal vibration signals collected by the distributed optical fiber structure observation network and to preprocess them.

[0181] The target separation module is used to perform event trigger detection based on preprocessed spatiotemporal vibration signals, and to cluster and separate event segments under mixed conditions to obtain single-target event sequences.

[0182] The feature extraction and fusion module is used to extract structural domain features, rhythm domain features, and energy domain features from the structural domain, rhythm domain, and energy domain respectively from a single target event sequence, and to perform multi-domain feature fusion on the structural domain features, rhythm domain features, and energy domain features to obtain a fused feature vector.

[0183] The classification module is used to calculate the posterior probability on a scalable set of categories based on the fused feature vectors, and outputs the final identified category of the vulnerable group.

[0184] Example 4

[0185] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for classifying vulnerable groups at road intersections oriented towards road surface disturbance.

[0186] Example 5

[0187] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the aforementioned method for classifying vulnerable groups at road intersections to address road surface disturbances.

[0188] Example 6

[0189] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for classifying vulnerable groups at road intersections oriented to road surface disturbance.

[0190] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. 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, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0191] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0192] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A classification method for vulnerable groups at road intersections oriented towards road surface disturbance, characterized in that, include: A distributed optical fiber structure observation network was deployed in the target monitoring area and spatiotemporal calibration was performed. Acquire spatiotemporal vibration signals collected by a distributed optical fiber structure observation network and preprocess them; Event trigger detection is performed based on preprocessed spatiotemporal vibration signals, and event segments are clustered and separated under mixed conditions to obtain single-target event sequences. The process of event trigger detection based on preprocessed spatiotemporal vibration signals, and clustering and separating event segments under mixed conditions to obtain single-target event sequences, includes: performing event trigger detection on preprocessed spatiotemporal vibration signals, constructing energy statistics, and triggering passage events based on dynamic thresholds; estimating the passage delay using gate channel to obtain initial velocity; performing multi-target clustering and separation of event segments under mixed conditions, and obtaining single-target event sequences through maximum likelihood clustering; extracting structural domain features, rhythmic domain features, and energy domain features from the structural domain, rhythmic domain, and energy domain of the single-target event sequence, respectively, and performing multi-domain feature fusion to obtain a fused feature vector; the process of extracting structural domain features, rhythmic domain features, and energy domain features from the structural domain, rhythmic domain, and energy domain of the single-target event sequence, respectively, and performing multi-domain feature fusion to obtain a fused feature vector, includes: extracting structural domain features, rhythmic domain features, and energy domain features from the structural domain, rhythmic domain, and energy domain of the single-target event sequence, respectively, and performing multi-domain feature fusion to obtain a fused feature vector, including ..., respectively, and performing multi-domain feature fusion to obtain a fused feature vector, including: extracting structural domain features, rhythmic domain features, and energy domain features from the structural domain, rhythmic domain, and energy domain, respectively, and performing multi-domain feature fusion to obtain a fused feature vector, including: extracting structural domain features, rhythmic domain features, and energy domain features from the structural domain, rhythmic domain, and energy The system characterizes the wheel assembly structure and contact geometry in the domain, the gait cycle, stride length, and stability in the rhythm domain, and the time-frequency energy distribution differences in the energy domain. It calculates the number of gate peaks and the half-width at half-maximum (FWHM) of the peaks to obtain structural domain features, calculates the step frequency and derives stride length and stability to obtain rhythm domain features, and calculates the time-frequency spectrum and frequency band energy proportions to obtain energy domain features. It defines adaptive fusion weights for multi-domain fusion and constructs a fusion feature vector as the classification input. Based on the fusion feature vector, it calculates the posterior probability on an expandable category set and outputs the final identified vulnerable group category. The calculation of the posterior probability on the expandable category set based on the fusion feature vector and outputting the final identified vulnerable group category includes: calculating the posterior probability and outputting the target category and confidence level, supporting unknown classes and online calibration; calculating the posterior probability on the expandable category set based on the fusion feature vector and outputting the final category and confidence level; when the decision confidence level does not meet preset requirements, it outputs an unknown, pending confirmation category, or triggers rule constraint verification.

2. The method for classifying vulnerable groups at road intersections to address road surface disturbance as described in claim 1, characterized in that, The deployment of a distributed optical fiber structure observation network and spatiotemporal calibration in the target monitoring area includes: deploying optical fiber sensors, constructing a main monitoring channel and gate line channels, ensuring consistent coupling and completing spatiotemporal calibration, deploying a distributed optical fiber main monitoring channel in the zebra crossing and non-motorized vehicle lane area at intersections, and deploying transverse gate line channels to form structural observation; constraining the consistency of channel response and completing the mileage-to-road coordinate mapping.

3. The method for classifying vulnerable groups at road intersections to address road surface disturbance as described in claim 1, characterized in that, The acquisition and preprocessing of spatiotemporal vibration signals collected by the distributed optical fiber structure observation network includes: acquiring spatiotemporal vibration signals from the main monitoring channel and the gate channel; performing preprocessing operations such as multi-scale noise reduction, narrowband interference suppression, and background baseline recursive update on the signals; and calculating the signal-to-noise ratio as a quality indicator for confidence modulation.

4. A classification system for vulnerable groups at road intersections oriented towards road surface disturbance, characterized in that, The method for classifying vulnerable groups at road intersections oriented towards road surface disturbance, as described in any one of claims 1-3, includes: a channel deployment module for deploying a distributed optical fiber structure observation network in the target monitoring area and performing spatiotemporal calibration; a signal acquisition module for acquiring spatiotemporal vibration signals collected by the distributed optical fiber structure observation network and preprocessing them; a target separation module for performing event trigger detection based on the preprocessed spatiotemporal vibration signals, clustering and separating event segments under mixed conditions to obtain single-target event sequences; a feature extraction and fusion module for extracting structural domain features, rhythmic domain features, and energy domain features from the structural domain, rhythmic domain, and energy domain of the single-target event sequences, respectively, and performing multi-domain feature fusion on the structural domain features, rhythmic domain features, and energy domain features to obtain a fused feature vector; and a classification module for calculating the posterior probability on an expandable category set based on the fused feature vector and outputting the finally identified vulnerable group category.

5. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for classifying vulnerable groups at road intersections oriented to road surface disturbance as described in any one of claims 1-3.

6. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the method for classifying vulnerable groups at road intersections oriented towards road surface disturbance as described in any one of claims 1-3.

7. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for classifying vulnerable groups at road intersections oriented to road surface disturbance as described in any one of claims 1-3.

Images