Traffic track monitoring method based on distributed optical fiber acoustic sensing

Abstract

The invention belongs to the field of traffic monitoring, and particularly relates to a traffic track monitoring method based on distributed optical fiber acoustic sensing, which comprises the following steps: preprocessing an original phase signal to obtain a phase signal of a target channel; performing signal envelope extraction on the phase signals, and performing screening to obtain a candidate track point set; density-based spatial clustering is carried out on points in the candidate trajectory point set, and the points meeting a density connectivity condition are divided into a plurality of trajectory clusters; aiming at each track cluster, constructing an anti-outlier track fitting model to obtain an initial fitting straight line corresponding to each track cluster, and carrying out secondary elimination on outliers to obtain a final fitting straight line; all the final fitting straight lines are verified and merged, final effective track straight lines are obtained, all the final effective track straight lines are connected according to the time sequence, a complete vehicle track is drawn, the method is used for track recognition in the complex traffic environment, and robustness and stability in environments such as severe weather and noise interference are improved.

Description

Technical Field

[0001] This application belongs to the field of traffic monitoring, specifically involving a traffic trajectory monitoring method based on distributed fiber optic acoustic sensing. It is particularly suitable for high-precision vehicle trajectory extraction in complex noise environments and can be widely applied to scenarios such as smart city traffic management, traffic flow statistics, and vehicle speed estimation. Background Technology

[0002] Distributed fiber acoustic sensing (DAS) technology has significant potential in fields such as urban traffic monitoring due to its advantages of low cost, wide coverage, anti-interference, and high resolution. This technology can capture vehicle vibration signals using existing optical cables to achieve vehicle speed estimation and traffic flow statistics, making up for the shortcomings of traditional traffic monitoring (high cost of surveillance cameras, large deviation of mobile phone signal data, and susceptibility of geomagnetic sensors to environmental influences).

[0003] However, distributed fiber optic acoustic sensing systems face two major challenges in traffic monitoring: First, complex noise interference, including inherent system noise such as light source frequency shift and electrical noise, as well as external noise such as environmental interference and the superposition of vibrations from multiple vehicles, leading to a decrease in the signal-to-noise ratio; second, the frequency components of vehicle vibration signals are complex, with effective components randomly distributed across different modes, and outliers easily mixed into the trajectory points. Existing vehicle trajectory detection methods often employ the least squares method for trajectory fitting, which is sensitive to outliers and prone to fitting deviations under complex noise and outlier interference, resulting in insufficient trajectory extraction accuracy and affecting the accuracy of subsequent vehicle speed estimation and traffic flow statistics.

[0004] Existing noise suppression methods (such as Butterworth filtering and median filtering combinations) have limited effectiveness against non-stationary noise. Traditional "modal decomposition-correlation coefficient" strategies are difficult to adapt to complex signals, are prone to missing effective components or misjudgment, and affect the accuracy of vehicle speed estimation. There is an urgent need for multimodal phase noise suppression schemes adapted to traffic scenarios. Summary of the Invention

[0005] The technical problem to be solved by this application is to provide a traffic trajectory monitoring method based on distributed fiber optic acoustic sensing, which solves the problem that existing noise suppression technologies are difficult to deal with non-stationary noise in traffic scenarios and that complex signal modes are prone to losing effective components.

[0006] This application provides a vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing, comprising: The raw phase signal output from the distributed optical fiber sensing system is acquired, and the raw phase signal is preprocessed to obtain the phase signal of the target channel. The phase signal is subjected to signal envelope extraction, and the signal envelope is filtered according to a preset energy threshold to obtain a set of candidate trajectory points; Density-based spatial clustering is performed on the points in the candidate trajectory point set, and the points that meet the density connectivity condition are divided into multiple trajectory clusters; For each trajectory cluster, the minimum absolute value deviation method is used to fit a straight line, and a trajectory fitting model against outliers is constructed to obtain the initial fitting line corresponding to each trajectory cluster. Outliers are then removed twice to obtain the final fitting line. Based on the preset similarity criteria, all the final fitted lines are verified and merged to obtain the final valid trajectory lines. All the final valid trajectory lines are then connected in chronological order to finally draw the complete vehicle trajectory.

[0007] Further, the original phase signal is preprocessed, including: Select one or more channels from multiple channels that correspond to the lane area to be monitored as target channels; Based on the determined channel length and sampling rate, the phase signal of the target channel is extracted.

[0008] Further, the phase signal is subjected to signal envelope extraction, and the signal envelope is filtered according to a preset energy threshold to obtain a candidate trajectory point set, including: Perform a Hilbert transform on the phase signal of each target channel to obtain the corresponding analytic signal, and calculate the signal envelope of each target channel by calculating the amplitude of the analytic signal; Initial high-value points are extracted based on envelope threshold, and the initial high-value points are filtered in multiple rounds. The multiple rounds of filtering include: excluding initial abnormal channels, filtering significant points based on time windows, and removing abnormal interference channels with an excessive number of significant points. The time-channel coordinates of the significantly high-value points obtained after filtering are normalized and converted into pixel coordinates to construct a binary image; The binary image is subjected to Hough transform to detect potential straight lines, and points that are less than a preset threshold from the straight lines are selected as candidate trajectory point sets.

[0009] Furthermore, density-based spatial clustering is performed on the points in the candidate trajectory point set, dividing the points that meet the density connectivity condition into multiple trajectory clusters, including: Each point in the candidate trajectory point set is used to construct a two-dimensional coordinate point with a timestamp and channel number as a clustering input sample; The neighborhood radius and minimum number of contained points are set for density clustering, wherein the neighborhood radius is calibrated based on the fiber channel spacing and signal sampling frequency, and the minimum number of contained points is determined based on the minimum signal duration of the traffic target passing through the fiber monitoring segment; Calculate the Euclidean distance between any two sample points, and determine whether the two points are neighbors based on the relationship between the Euclidean distance and the neighborhood radius; Traverse all input samples and label them as core points, boundary points, or noise points based on the number of points contained in the neighborhood of each sample point. Starting from the core point, all its density-reachable neighboring points are recursively merged to form an initial cluster, and noise points are removed to obtain multiple trajectory candidate point clusters corresponding to the movement of traffic targets.

[0010] Furthermore, the labeling rules include: labeling sample points whose number of points in their neighborhood is greater than or equal to the minimum number of points as core points; labeling sample points whose number of points in their neighborhood is less than the minimum number of points but belong to the neighborhood of a certain core point as boundary points; and labeling sample points that are neither core points nor boundary points as noise points.

[0011] Furthermore, after obtaining multiple candidate trajectory point clusters corresponding to the movement of the traffic target, the method further includes: Detect whether there are overlapping clusters of candidate trajectory points within the same time interval; For candidate point clusters of trajectory that overlap, the overlap degree between point clusters is calculated, whereby the overlap degree is the ratio of the number of overlapping points to the total number of points in the smaller point cluster of the two point clusters. When the overlap is greater than a preset overlap threshold, the overlapping trajectory candidate point clusters are merged into the same trajectory candidate point cluster.

[0012] Furthermore, the minimum absolute value deviation method is used for line fitting to construct a trajectory fitting model against outliers, obtaining the initial fitted line corresponding to each trajectory cluster, including: For coordinate points within each trajectory cluster Establish a linear fitting model ,in The slope of the line. The intercept is... For timestamps, Number the channel; Construct the LAD fitting loss function by minimizing the fitting error of all points in absolute value form; The slope is obtained by solving the minimum value of the LAD fitting loss function using an iterative optimization algorithm. With intercept The optimal solution is used to determine the initial fitted line corresponding to each trajectory cluster.

[0013] Furthermore, the secondary removal of outliers includes: Calculate the vertical distance from each point within the trajectory cluster to the initial fitted line; Set a distance threshold, and identify points whose vertical distance is greater than the distance threshold as outliers and remove them; Based on the remaining trajectory points after removal, the straight line fitting of the LAD fitting loss function is re-executed to obtain the final trajectory cluster fitting straight line.

[0014] Furthermore, based on the preset similarity criteria, all the final fitted lines are verified and merged, including: calculating the similarity index between any two fitted lines, wherein the similarity index includes the slope difference, intercept difference, and the proportion of overlapping time axis intervals. A similarity threshold is set. When two fitted lines meet the similarity threshold, the two fitted lines are determined to belong to the same vehicle trajectory segment and are merged. The validity of the merged fitted lines is verified, and the fitted lines that meet the line coverage requirements are selected as the final valid trajectory lines.

[0015] Furthermore, by connecting all the final valid trajectory lines in chronological order, the complete vehicle trajectory is finally drawn, including: The direction of movement of the traffic target is determined by the slope of the final effective trajectory line. A positive slope indicates movement along the direction of increasing channel number, while a negative slope indicates movement along the direction of decreasing channel number.

[0016] Compared with the prior art, the advantages of this application are as follows: This application can be used for track recognition in complex traffic environments, improving robustness and stability in environments such as severe weather and noise interference. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method provided in the embodiments of this application; Figure 2 This is a grayscale image of the original data from a certain location on August 3, 2024, provided in an embodiment of this application. Figure 3 This is a clustering result diagram of data from a certain location on August 3, 2024, provided in an embodiment of this application; Figure 4 This is a graph showing the linear fitting result of the clustering results provided in the embodiments of this application; Figure 5 This is a diagram showing the final result of vehicle trajectory fitting provided in this application embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] See Figure 1 As shown in the figure, the vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing provided in this application uses the minimum absolute deviation method (LAD) as the core fitting technology, and includes the following steps: S101, Acquire the raw phase signal output by the distributed optical fiber sensing system, and preprocess the raw phase signal to obtain the phase signal of the target channel; S102, extract the signal envelope of the phase signal, and filter the signal envelope according to a preset energy threshold to obtain a set of candidate trajectory points; S103, Perform density-based spatial clustering on the points in the candidate trajectory point set, and divide the points that meet the density connectivity condition into multiple trajectory clusters; S104. For each trajectory cluster, the minimum absolute value deviation method is used to fit a straight line, and a trajectory fitting model against outliers is constructed to obtain the initial fitting straight line corresponding to each trajectory cluster. Outliers are removed twice to obtain the final fitting straight line. S105, based on the preset similarity criterion, verifies and merges all the final fitted straight lines to obtain the final effective trajectory straight lines, and connects all the final effective trajectory straight lines in chronological order to finally draw the complete vehicle trajectory.

[0020] In step S101, which is the signal acquisition and preprocessing stage, a distributed fiber optic sensing system is used to acquire the raw phase signal in the traffic monitoring scenario. The raw phase signal is then processed by block loading to determine the channel length and sampling rate, accurately extracting the phase signal of the target channel. Phase unwrapping technology is used to eliminate interference in the signal. To mitigate jump errors, a multi-stage filtering strategy is employed to suppress noise: Butterworth bandpass filters are used to filter inherent system noise and external environmental interference, followed by median filtering to further smooth the signal; finally, downsampling is performed on the filtered data to reduce the amount of data while preserving effective signal characteristics, resulting in high-quality preprocessed data.

[0021] By collecting and preprocessing the data obtained from the fiber optic transmission system, we can obtain, for example... Figure 2 The grayscale image shown extracts the signal from the target channel.

[0022] The transfer function design of the Butterworth bandpass filter is based on the amplitude-frequency response formula: , in, For order, The cutoff frequency, The square of the amplitude-frequency response, Given the angular frequency of the input signal, this design enables precise filtering of signals within a specific frequency range.

[0023] Step S102 involves extracting the candidate trajectory point set: extracting the signal envelope from the phase signal and filtering the signal envelope according to a preset energy threshold to obtain the candidate trajectory point set, including: Perform a Hilbert transform on the phase signal of each target channel to obtain the corresponding analytic signal, and calculate the signal envelope of each target channel by calculating the amplitude of the analytic signal; Initial high-value points are extracted based on envelope threshold, and the initial high-value points are filtered in multiple rounds. The multiple rounds of filtering include: excluding initial abnormal channels, filtering significant points based on time windows, and removing abnormal interference channels with an excessive number of significant points. The time-channel coordinates of the significantly high-value points obtained after filtering are normalized and converted into pixel coordinates to construct a binary image; The binary image is subjected to Hough transform to detect potential straight lines, and points that are less than a preset threshold from the straight lines are selected as candidate trajectory point sets.

[0024] Clustering is performed on the points filtered by Hough transform. The similarity between points is calculated using Euclidean distance, dividing densely distributed points into different clusters and eliminating noise points. (See [link to relevant documentation]). Figure 3 As shown, the clustering results reflect the changing trends over time well, with high separation between clusters and good clustering performance. A few time points (such as around 1200 seconds) show inter-cluster overlap, which may represent transitional states or noise points.

[0025] By employing multiple filtering strategies, including energy threshold screening, time window constraints, and abnormal channel removal, environmental noise and sudden interference are effectively suppressed, improving the reliability of trajectory point extraction. Spatiotemporal coordinates are mapped to binary images for line detection, and potential trajectory patterns are identified using geometric characteristics, reducing the probability of false alarms in complex backgrounds.

[0026] Specifically, based on the preprocessed phase data, a Hilbert transform is performed on each channel signal to obtain an analytic signal. The signal envelope is then obtained by calculating the amplitude of the analytic signal, highlighting the signal's intensity characteristics. The mathematical expression of the Hilbert transform is: , in For the Hilbert transform of the phase signal, The result of the Hilbert transform. The signal is a phase signal source, and t is a time variable. It is an integral variable.

[0027] High-value points are initially extracted by setting an envelope threshold. The reliability of the point set is improved through a multi-round filtering mechanism to obtain candidate trajectory points. The multi-round filtering mechanism includes: (1) Eliminate initial abnormal channels: Eliminate initial abnormal channels to avoid interference from invalid channel data; (2) Speculative point selection based on time window: A 1-second time window is used for statistics, and significant points with ≥20 high-value points within the window are retained. The formula for judging high-value points is the envelope amplitude. ; (3) Interference channel elimination: Channels with more than 700 significant points are eliminated, and such channels are judged as abnormal interference sources.

[0028] The time-channel coordinates of significantly high-value points are normalized and converted to pixel coordinates (magnified 10 times for improved accuracy) to construct a binary image. Potential lines are detected using the Hough transform, and the fitted lines corresponding to the top 10 peaks are extracted. Points less than 2 pixels away from the fitted lines are selected as candidate points. The polar coordinate representation of the fitted line in the Hough transform is as follows: , in The distance from the origin to the fitted line is denoted as . for Axis coordinates for Axis coordinates The angle between the normal and the x-axis

[0029] In one embodiment, density-based spatial clustering is performed on the points in the candidate trajectory point set to divide the points that meet the density connectivity condition into multiple trajectory clusters, including: Each point in the candidate trajectory point set is used to construct a two-dimensional coordinate point with a timestamp and channel number as a clustering input sample; The neighborhood radius and minimum number of contained points are set for density clustering, wherein the neighborhood radius is calibrated based on the fiber channel spacing and signal sampling frequency, and the minimum number of contained points is determined based on the minimum signal duration of the traffic target passing through the fiber monitoring segment; Calculate the Euclidean distance between any two sample points, and determine whether the two points are neighbors based on the relationship between the Euclidean distance and the neighborhood radius; Traverse all input samples and label them as core points, boundary points, or noise points based on the number of points contained in the neighborhood of each sample point. Starting from the core point, all its density-reachable neighboring points are recursively merged to form an initial cluster, and noise points are removed to obtain multiple trajectory candidate point clusters corresponding to the movement of traffic targets.

[0030] In one example, each point in the core candidate point set is indexed by (timestamp) Channel number Constructing two-dimensional coordinate points , as input samples for density clustering (DBSCAN); set key parameters for DBSCAN clustering: neighborhood radius is calibrated based on fiber channel spacing and signal sampling frequency, and minimum number of included points is determined based on the minimum signal duration of traffic targets passing through the fiber monitoring segment; Calculate any two points ( and The Euclidean distance between them is given by the formula: , in, The distance is Euclidean; when the distance is less than the neighborhood radius, the two points are considered to be neighbors. Iterate through all input samples, marking core points (number of points in the neighborhood ≥ minimum number of points), boundary points (number of points in the neighborhood < minimum number of points, but belonging to the neighborhood of the core point), and noise points (neither core points nor boundary points). Starting from the core point, recursively merge all its reachable neighborhood points to form an initial cluster, remove noise points, and obtain the trajectory clusters corresponding to the movement of the traffic target. If there are overlapping point clusters in the same time interval, calculate the overlap between point clusters (number of overlapping points / total number of points in the smaller point cluster). When the overlap is greater than 30%, merge them into the same trajectory cluster to ensure the integrity of the trajectory of a single vehicle.

[0031] In step S104, for the coordinate points within each trajectory cluster A linear fitting model is established, and the LAD (Least Absolute Deviation) method is used to achieve accurate fitting against outliers. Specifically, this includes: Construct a linear fitting model: , In the formula, The slope of the line (directly reflects the speed-related characteristics of traffic targets). The intercept is... For timestamps, This is the channel number.

[0032] The LAD fitting loss function is constructed by minimizing the fitting error of all points in absolute value form. The fitting error can be understood as the absolute bias and reducing the impact of outliers on the fitting results. , In the formula, For the first in the trajectory cluster The coordinates of the points The total number of points within the trajectory cluster is denoted by . Compared to the traditional least squares loss function, this loss function is significantly less sensitive to outliers and has better fitting stability.

[0033] The minimum value of the LAD fitting loss function is found by using an iterative optimization algorithm: starting from the initial guess value (the initial guess of the slope is 0, and the initial guess of the intercept is the mean value of the channel number in the trajectory cluster), the optimization algorithm is used to iteratively solve the problem and quickly converge to obtain the optimal solution of the slope m and the intercept b, thus determining the initial fitting line.

[0034] Construct a secondary outlier removal mechanism: Calculate the perpendicular distance from each point within the trajectory cluster to the initial fitted line, using the following formula: , in, It is the vertical distance. For the first A timestamp at each point, For the first The channel number of each point.

[0035] In one example, a distance threshold of 2 is set (based on fiber optic signal noise level calibration) to remove outliers whose distance is greater than the distance threshold; the linear fitting of the LAD fitting loss function is re-executed based on the remaining valid points to obtain the final trajectory cluster fitting line, further improving the fitting accuracy.

[0036] For each valid cluster, a straight line is fitted using the Least Absolute Deviation (LAD) method, and the slope is calculated based on the loss function. and intercept After fitting, stable results were obtained through multiple conditional verifications. See [link / reference]. Figure 4 As shown, after fitting, most of the data points closely surround the fitted line.

[0037] In step S105, based on a preset similarity criterion, all final fitted lines are verified and merged, including: calculating the similarity index between any two fitted lines, wherein the similarity index includes slope difference, intercept difference, and the proportion of overlapping time axis intervals. A similarity threshold is set. When two fitted lines meet the similarity threshold, the two fitted lines are determined to belong to the same vehicle trajectory segment and are merged. The validity of the merged fitted lines is verified, and the fitted lines that meet the line coverage requirements are selected as the final valid trajectory lines.

[0038] Specifically, the fitted lines are verified and merged based on a preset similarity criterion to draw the complete vehicle trajectory: Calculate any two fitted lines and Similarity metrics The first fitted line, This is the second fitted line. Let be the slope of the first fitted line. The intercept of the first fitted line is given. The slope of the second fitted line is included. The intercept of the first fitted line is used. Similarity indices include: slope difference. Intercept difference: (within the same time interval) (Internal calculation), percentage of overlapping time axis intervals.

[0039] Set a similarity threshold. In one example, set the similarity threshold as follows: (Determined based on the range of fluctuations in the speed of the moving traffic target). (Based on fiber channel number span calibration), and the overlap interval of the trajectory clusters corresponding to the two fitted lines on the time axis is >50%; when all the above conditions are met, it is determined that the two fitted lines belong to the same vehicle trajectory segment, all fitted lines are calculated pairwise, and finally the fitted lines are merged using the mean fusion method: merged fitted line slope The intercept of the merged line:

[0040] , In the formula, The number of lines required to meet the merging conditions. For the first The slope of the fitted straight line, For the first The intercept of the fitted straight line.

[0041] The validity of the merged fitted lines is verified, and the fitted lines that meet the line coverage requirements are selected as the final valid trajectory lines: the line coverage is calculated (number of points in the point cluster covered by the line / total number of points in the point cluster). When the coverage is >80%, it is determined as the final valid trajectory line; all the final valid trajectory lines are connected in chronological order, and combined with the direction of traffic target movement (determined by the positive or negative slope: positive slope is along the direction of increasing channel number, negative slope is along the direction of decreasing channel number), the complete vehicle movement trajectory is drawn.

[0042] After fitting, verification, and merging, the final effective trajectory lines are obtained. All the final effective trajectory lines are then connected in chronological order to finally draw the complete vehicle trajectory. (See [link]). Figure 5 As shown, the fitted curve is smooth with no obvious abrupt changes; the fitted curve has a high degree of fit with the original data points, indicating that it can accurately describe the vehicle's trajectory.

[0043] By loading and preprocessing the measured data, using data from 1350-1450 channels, an automated process was built from file reading, phase dewinding, multi-stage filtering (a combination of 4th-order Butterworth bandpass filtering, 5×5 median filtering and 3×3 Wiener filtering) to downsampling, preserving the effective signals from 1300-1500 channels.

[0044] The final fitting result is significantly better than that of traditional methods, and the fitting error is reduced by 15% compared with the least squares method. This provides accurate quantitative basis for signal pattern analysis and effectively supports subsequent trajectory feature extraction and application.

[0045] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing, characterized in that, include: The raw phase signal output from the distributed optical fiber sensing system is acquired, and the raw phase signal is preprocessed to obtain the phase signal of the target channel. The phase signal is subjected to signal envelope extraction, and the signal envelope is filtered according to a preset energy threshold to obtain a set of candidate trajectory points; Density-based spatial clustering is performed on the points in the candidate trajectory point set, and the points that meet the density connectivity condition are divided into multiple trajectory clusters; For each trajectory cluster, the minimum absolute value deviation method is used to fit a straight line, and a trajectory fitting model against outliers is constructed to obtain the initial fitting line corresponding to each trajectory cluster. Outliers are then removed twice to obtain the final fitting line. Based on the preset similarity criteria, all the final fitted lines are verified and merged to obtain the final valid trajectory lines. All the final valid trajectory lines are then connected in chronological order to finally draw the complete vehicle trajectory.

2. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, The original phase signal is preprocessed, including: Select one or more channels from multiple channels that correspond to the lane area to be monitored as target channels; Based on the determined channel length and sampling rate, the phase signal of the target channel is extracted.

3. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, The phase signal is subjected to signal envelope extraction, and the signal envelope is filtered according to a preset energy threshold to obtain a candidate trajectory point set, including: Perform a Hilbert transform on the phase signal of each target channel to obtain the corresponding analytic signal, and calculate the signal envelope of each target channel by calculating the amplitude of the analytic signal; Initial high-value points are extracted based on envelope threshold, and the initial high-value points are filtered in multiple rounds. The multiple rounds of filtering include: excluding initial abnormal channels, filtering significant points based on time windows, and removing abnormal interference channels with an excessive number of significant points. The time-channel coordinates of the significantly high-value points obtained after filtering are normalized and converted into pixel coordinates to construct a binary image; The binary image is subjected to Hough transform to detect potential straight lines, and points that are less than a preset threshold from the straight lines are selected as candidate trajectory point sets.

4. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, Density-based spatial clustering is performed on the points in the candidate trajectory point set, dividing the points that meet the density connectivity condition into multiple trajectory clusters, including: Each point in the candidate trajectory point set is used to construct a two-dimensional coordinate point with a timestamp and channel number as a clustering input sample; The neighborhood radius and minimum number of contained points are set for density clustering, wherein the neighborhood radius is calibrated based on the fiber channel spacing and signal sampling frequency, and the minimum number of contained points is determined based on the minimum signal duration of the traffic target passing through the fiber monitoring segment; Calculate the Euclidean distance between any two sample points, and determine whether the two points are neighbors based on the relationship between the Euclidean distance and the neighborhood radius; Traverse all input samples and label them as core points, boundary points, or noise points based on the number of points contained in the neighborhood of each sample point. Starting from the core point, all its density-reachable neighboring points are recursively merged to form an initial cluster, and noise points are removed to obtain multiple trajectory candidate point clusters corresponding to the movement of traffic targets.

5. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 4, characterized in that, The labeling rules include: labeling sample points whose number of points in their neighborhood is greater than or equal to the minimum number of points as core points; labeling sample points whose number of points in their neighborhood is less than the minimum number of points but belong to the neighborhood of a core point as boundary points; and labeling sample points that are neither core points nor boundary points as noise points.

6. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 4, characterized in that, After obtaining multiple candidate trajectory point clusters corresponding to the movement of the traffic target, the method further includes: Detect whether there are overlapping clusters of candidate trajectory points within the same time interval; For candidate point clusters of trajectory that overlap, the overlap degree between point clusters is calculated, whereby the overlap degree is the ratio of the number of overlapping points to the total number of points in the smaller point cluster of the two point clusters. When the overlap is greater than a preset overlap threshold, the overlapping trajectory candidate point clusters are merged into the same trajectory candidate point cluster.

7. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, The minimum absolute value deviation method is used for line fitting to construct a trajectory fitting model that resists outliers, obtaining the initial fitted line corresponding to each trajectory cluster, including: For coordinate points within each trajectory cluster Establish a linear fitting model ,in The slope of the line. The intercept is... For timestamps, Number the channel; Construct the LAD fitting loss function by minimizing the fitting error of all points in absolute value form; The slope is obtained by solving the minimum value of the LAD fitting loss function using an iterative optimization algorithm. With intercept The optimal solution is used to determine the initial fitted line corresponding to each trajectory cluster.

8. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 7, characterized in that, The secondary removal of outliers includes: Calculate the vertical distance from each point within the trajectory cluster to the initial fitted line; Set a distance threshold, and identify points whose vertical distance is greater than the distance threshold as outliers and remove them; Based on the remaining trajectory points after removal, the straight line fitting of the LAD fitting loss function is re-executed to obtain the final trajectory cluster fitting straight line.

9. The vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 1, characterized in that, Based on the preset similarity criteria, all the final fitted lines are verified and merged, including: calculating the similarity index between any two fitted lines, the similarity index including slope difference, intercept difference and the proportion of overlapping time axis intervals; A similarity threshold is set. When two fitted lines meet the similarity threshold, the two fitted lines are determined to belong to the same vehicle trajectory segment and are merged. The validity of the merged fitted lines is verified, and the fitted lines that meet the line coverage requirements are selected as the final valid trajectory lines.

10. A vehicle trajectory monitoring method based on distributed fiber optic acoustic sensing according to claim 9, characterized in that, Connect all the final valid trajectory lines in chronological order to draw the complete vehicle trajectory, including: The direction of movement of the traffic target is determined by the slope of the final effective trajectory straight line. A positive slope indicates movement along the direction of increasing channel number, while a negative slope indicates movement along the direction of decreasing channel number.

Images