A method of motorway accident monitoring

Abstract

The application provides a highway accident monitoring method, relates to the technical field of optical fiber vibration sensing, and acquires a vibration image according to each time of emitting a detection signal, constructs an entropy matrix through setting a distance window and a time window, takes the mean value of vibration peak points in each distance window as chaotic entropy, and represents the vehicle passing state of any road section at any time by the chaotic entropy; when an accident signal is detected, the chaotic entropy of the accident position is compared in time and space through a time component and a distance component to form a variation score, a variation coefficient threshold is generated according to the variation of a historical entropy matrix at the position, whether the position is misjudged is judged according to the variation amount, and whether to alarm is decided, the application comprehensively judges whether the position truly has an accident through the passing state of the accident signal position in time and space, and the accident judgment accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of fiber optic vibration sensing technology, specifically a method for monitoring highway accidents. Background Technology

[0002] Burying optical fibers in highways and detecting vibrations within these fibers using an OTDR system to obtain vehicle speed and accident information is a relatively advanced solution for highway safety monitoring. Specifically, Rayleigh scattering and Fresnel reflection are used to identify abnormal vibration points and breakage points in the fiber to determine accidents. Analysis of reflected signals reveals vibration information at various points along the fiber optic path. Vibration originates from tire pressure on the ground; the faster the speed, the greater the vibration intensity, and the greater the speed change, the stronger the vibration. However, with increasing monitoring accuracy, false alarms are increasingly prone to occur due to unforeseen circumstances. For example, in windy weather, a broken branch hitting a highway guardrail can transmit vibrations to the fiber optic cable. The OTDR system might detect abnormal vibrations and interpret this as an accident, resulting in a false alarm. Therefore, a monitoring method is needed to avoid detecting unforeseen events that do not affect traffic flow during accident monitoring.

[0003] In the prior art, document CN115580347A discloses a method, system, and device for early warning of optical cable damage in interference environments based on optical fiber sensing. This method utilizes redundant fiber cores in communication optical cables, connecting one core to an optical fiber sensing system as a vibration detector to collect vibration signals along the cable. The continuously collected vibration signals are converted into an image matrix. By analyzing background clutter in the images, it determines whether each detection section is in an interference environment or a static environment. Then, based on the environmental conditions, it automatically selects a signal filtering method to filter out truly threatening abnormal signals before making an alarm judgment, thereby optimizing the alarm effect.

[0004] While the publicly available technical documents enable the re-evaluation of abnormal signals along the fiber optic path, the complex scenarios of highways present a challenge. With diverse variables such as vehicle type and speed, and numerous temporal and spatial variations at any given location, analysis from a purely spatial perspective is insufficient for accurate judgment and hinders the improvement of judgment accuracy.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for monitoring highway accidents in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring highway accidents, comprising the following steps: Step 1: Obtain vibration images at each location on the optical fiber by detecting signals multiple times. Set a distance window to slide on the vibration images and use the average of all vibration peak points within the distance window as the chaos entropy. Construct an entropy matrix using the chaos entropy generated by a single vibration image as a row vector. Set a time window, and the number of times the detection signal is emitted within the time window is the dimension of the entropy matrix. Step 2: When an accident signal is detected, obtain the accident location. Based on the distance between the accident location and the point where the detection signal was emitted, determine the element corresponding to the accident location in the entropy matrix and mark it as an accident element. Step 3: Set the sub-vectors of the column vectors containing accident elements as time components, and the sub-vectors of the row vectors containing accident elements as distance components. Label the corresponding time components and distance components as corresponding vector groups. The correspondence means that the first element of the time component and the distance component are the same. Use the mean squared error in the corresponding vector group as the coefficient of variation, obtain the coefficient of variation of all corresponding vector groups in the historical entropy matrix, and select the coefficient of variation threshold. Step 4: Obtain multiple corresponding vector groups containing accident elements when the accident signal occurs, and obtain the coefficient of variation for each. Generate a variation score based on the rate of change of the coefficient of variation, generate an over-score based on the comparison between the coefficient of variation and the coefficient of variation threshold, construct a variation quantity by combining the variation score and the over-score, set a variation quantity threshold, and compare the variation quantity with the variation quantity threshold to determine whether the accident signal is a misjudgment.

[0008] Furthermore, the vibration images at various positions on the optical fiber are acquired through the OTDR system each time a detection signal is transmitted. The vibration images are two-dimensional rectangular coordinate system images, with the horizontal axis representing distance and the vertical axis representing amplitude. The origin of the distance is the detection signal transmission point of the OTDR system, and the distance value represents the distance to the detection signal transmission point of the OTDR system. Each wave peak is marked as a vehicle point, and the amplitude and distance values ​​of each vehicle point are recorded. All vibration images are sorted according to the transmission time of the detection signal to form a vibration image sequence.

[0009] Furthermore, a distance window and a time window are set respectively. The distance window slides on the vibration image of a single transmission detection signal. A disorder entropy is constructed based on the number of vehicle points within the sliding window and the amplitude of each vehicle point. The disorder entropy represents the average amplitude of all vehicle points within the distance window. The entropy matrix is ​​constructed using the following logic: As the distance window slides across the vibration image of a single detection signal, the average amplitude within each distance window is an element of the row vector of the entropy matrix, and the element number corresponds to the number of times the distance window slides. The chaotic entropy sequence formed by each single transmission detection signal is a row vector, and the number of all vibration images within the time window is the dimension of the entropy matrix.

[0010] Furthermore, when the OTDR system detects an accident signal to determine that an accident may have occurred at a certain location, it obtains the accident location and performs element-wise mapping of the accident location in the entropy matrix, with the mapping logic as follows: In a single row vector, obtain all distance windows including the accident location and all corresponding elements of the distance windows in the entropy matrix. When the number of all corresponding elements is odd, take the element corresponding to the middle position of all corresponding elements as the element corresponding to the accident location in the entropy matrix. When the number of all corresponding elements is even, take any one of the two elements at the middle position of all corresponding elements as the element corresponding to the accident location in the entropy matrix. The element corresponding to the accident location in the entropy matrix is ​​designated as the accident element.

[0011] Further, a historical entropy matrix is ​​obtained and time and distance components are defined. The historical entropy matrix is ​​a matrix formed during the normal traffic period of the highway without accidents. The distance component is a subvector of a single row vector containing accident elements in the entropy matrix. The time component represents a subvector of the column vector of the column containing the accident element in the entropy matrix. The first element in the time component and the first element in the distance component are the same element in the entropy matrix. The direction of increase of the element number in the distance component is opposite to the direction of travel on the road segment. The direction of increase of the element number in the time component is the same as the direction of increase of the detection signal transmission number. The number of elements in the time component and the distance component are the same. The time component and the distance component are then labeled as corresponding vector groups.

[0012] Furthermore, the coefficient of variation of the corresponding vector group in the historical entropy matrix is ​​obtained, as follows: Get the difference between the second element in the time component and the second element in the distance component, then get the difference between the three elements in the time component and the third element in the distance component, until the difference between all corresponding elements of the two sub-vectors that do not contain the first element is obtained, and the mean squared error of these differences is obtained and labeled as the coefficient of variation. Obtain the mean squared error of all corresponding vector groups before the accident signal, and select a threshold such that 90% of the coefficients of variation are less than the threshold. This threshold is then calibrated as the coefficient of variation threshold.

[0013] Furthermore, the accident element at the time of the accident signal is obtained, and this element is used as the penultimate element, the second-to-last element, and so on, up to the first element of the time component to construct multiple corresponding vector groups. The coefficient of variation of each vector group is obtained and a sequence of coefficients of variation is formed. To obtain the rate of change for each element in the coefficient of variation sequence, calculate the variation score for multiple vector groups, the logic is as follows: The weighted sum of all the rate of change values ​​is calculated. When a rate of change is positive, its weight is greater than 1, and when the rate of change is negative, its weight is 0.

[0014] Furthermore, the coefficient of variation sequence is compared with the coefficient of variation threshold to obtain an excess score, as follows: Subtract the coefficient of variation threshold from each value in the coefficient of variation sequence to obtain the difference. Sum all the results with positive differences. Determine the weight of the first element with a positive difference in the coefficient of variation by its order. The earlier the element is in the order, the greater its weight. The later the element is in the order, the smaller its weight. The weight is always greater than 0. Multiply the sum by the weight to obtain the overweight score.

[0015] Furthermore, the mutation score and the excess score are summed to obtain the mutation amount. A mutation amount threshold is set. When the mutation amount exceeds the mutation amount threshold, it is determined that an accident has indeed occurred at that location and an alarm is triggered. Otherwise, the location is determined to be a false alarm and no alarm is triggered.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention acquires vibration images based on each transmitted detection signal, constructs an entropy matrix by setting distance and time windows, and uses the mean of vibration peak points within each distance window as the chaos entropy. It then constructs a temporal and spatial variation matrix on the optical fiber channel, using chaos entropy to represent the vehicle traffic conditions of any road segment at any time. When an accident signal is detected, the chaos entropy at the accident location is compared temporally and spatially using both time and distance components to generate a variation score, determining the change in traffic conditions at the accident signal location. A variation coefficient threshold is generated based on the variation of the historical entropy matrix at that location to provide a variation coefficient threshold adapted to the current road segment. The overall variation is constructed using the variation coefficient threshold as a standard. The variation is used to determine whether a misjudgment has occurred at that location and whether to issue an alarm. This application improves the accuracy of accident judgment by comprehensively judging whether an accident has actually occurred at the accident signal location based on the temporal and spatial traffic conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Please see Figure 1 The present invention provides a technical solution: A method for monitoring highway accidents, comprising the following steps: Step 1: Obtain vibration images at each location on the optical fiber by detecting signals multiple times. Set a distance window to slide on the vibration images and use the average of all vibration peak points within the distance window as the chaos entropy. Construct an entropy matrix using the chaos entropy generated by a single vibration image as a row vector. Set a time window, and the number of times the detection signal is emitted within the time window is the dimension of the entropy matrix. Step 1 includes the following: Step 101: When the OTDR system transmits a detection signal, it acquires vibration images at various positions on the optical fiber. The vibration image is a two-dimensional rectangular coordinate system image, with the horizontal axis representing distance and the vertical axis representing amplitude. The origin of the distance is the detection signal transmission point of the OTDR system, and the distance value represents the distance to the detection signal transmission point of the OTDR system. Each wave peak is marked as a vehicle point, and the amplitude and distance values ​​of each vehicle point are recorded. All vibration images are sorted according to the transmission time of the detection signal to form a vibration image sequence.

[0021] This step is the first crucial step in transforming physical signals into structured data, establishing a standardized spatiotemporal data coordinate system for the entire monitoring system. The raw, simulated backscattered light signals from the OTDR system are transformed into a discrete two-dimensional data field with "distance" and "amplitude" as axes, using the concept of "vibration images." This gives random vibration events on the road "digital coordinates" that can be precisely measured and tracked by a computer. Defining "vehicle points" is the first dimensionality reduction and feature abstraction of complex vibration patterns, filtering out subtle noise caused by non-vehicles and focusing on potential traffic events with significant amplitudes. The "vibration image sequence," formed by chronological sorting, establishes the time axis for analysis. The sequence generated in this step is a "data raw material library," upon which all subsequent advanced analyses (such as entropy calculation and matrix construction) depend entirely on the quality and structure of this raw material library. It provides the underlying data framework for the entire method, ensuring the alignment and preservation of information in both time and space dimensions, and is the cornerstone of all subsequent spatiotemporal correlation analyses.

[0022] Step 102: Set a distance window and a time window respectively. The distance window slides on the vibration image of a single transmission detection signal. Construct a chaos entropy based on the number of vehicle points within the sliding window and the amplitude of each vehicle point. The chaos entropy represents the average amplitude of all vehicle points within the distance window. The entropy matrix is ​​constructed using the following logic: As the distance window slides across the vibration image of a single detection signal, the average amplitude within each distance window is an element of the row vector of the entropy matrix, and the element number corresponds to the number of times the distance window slides. The chaotic entropy sequence formed by each single transmission detection signal is a row vector, and the number of all vibration images within the time window is the dimension of the entropy matrix.

[0023] This step involves a dimensionality upgrade and fusion process from "point features" to "field features" and then to the "state matrix," creating a comprehensive state index—the "entropy matrix"—that can simultaneously characterize the spatiotemporal density and intensity of traffic flow. The calculation of "chaotic entropy" (local amplitude mean) using a "distance window" is essentially a spatial smoothing and aggregation operation. Its deeper purpose is to replace the random amplitude of a single vehicle point with the overall vibration energy level of a local area. This significantly improves the robustness of the data in representing the overall state of traffic flow and reduces the interference of single-point abrupt changes. The entropy matrix is ​​not a simple data list, but a state plane that grids time (rows: different detection times) and space (columns: different road locations). Each matrix element carries comprehensive vibration intensity information for a specific time and a specific road segment. This data structure greatly facilitates subsequent analysis: in the spatial dimension, any column represents the vibration history of a fixed location over time; in the temporal dimension, any row represents the spatial vibration distribution of the entire road at a certain time. It enables precise positioning in step 2, spatiotemporal component extraction in step 3, and sequence analysis in step 4 to be completed efficiently through direct matrix row and column operations, making it the core carrier for the entire scheme to achieve high-efficiency, high-dimensional data analysis.

[0024] Step 2: When an accident signal is detected, obtain the accident location. Based on the distance between the accident location and the point where the detection signal was emitted, determine the element corresponding to the accident location in the entropy matrix and mark it as an accident element. Step 2 includes the following: When the OTDR system detects an accident signal to determine that an accident may have occurred at a certain location, it obtains the accident location and performs element-wise mapping of the accident location in the entropy matrix. The mapping logic is as follows: In a single row vector, obtain all distance windows including the accident location and all corresponding elements of the distance windows in the entropy matrix. When the number of all corresponding elements is odd, take the element corresponding to the middle position of all corresponding elements as the element corresponding to the accident location in the entropy matrix. When the number of all corresponding elements is even, take any one of the two elements at the middle position of all corresponding elements as the element corresponding to the accident location in the entropy matrix. The element corresponding to the accident location in the entropy matrix is ​​designated as the accident element.

[0025] This step resolves the fuzzy mapping problem from "physical coordinates" to "data coordinates," achieving intelligent locking from a coarse accident alarm area to a precise data analysis target. When the OTDR initially detects an abnormal signal, the approximate location of the accident can be determined. The mapping rule in this step (selecting the central element based on the parity of the number of coverage windows) is a strategy for achieving "centralized representation" at the data level; the core impact point of the accident is most likely located near the geometric center of its impact range. Through this rule, regardless of how many data windows the accident impact covers, the system can automatically and unambiguously determine a unique "accident element" as the core anchor point for analysis. This step guides the initial, low-confidence area alarm to the "navigation point" of the subsequent high-precision data analysis process. Only by accurately anchoring the "accident element" can steps 3 and 4, like radar locking onto a target, conduct in-depth scanning and pattern recognition around that element in the spatiotemporal dimensions, thereby avoiding the spread or defocusing of the analysis range due to target ambiguity, ensuring the concentration of subsequent analysis resources and the accuracy of conclusions.

[0026] Step 3: Set the sub-vectors of the column vectors containing accident elements as time components, and the sub-vectors of the row vectors containing accident elements as distance components. Label the corresponding time components and distance components as corresponding vector groups. The correspondence means that the first element of the time component and the distance component are the same. Use the mean squared error in the corresponding vector group as the coefficient of variation, obtain the coefficient of variation of all corresponding vector groups in the historical entropy matrix, and select the coefficient of variation threshold. Step 3 includes the following: Obtain the historical entropy matrix and set time and distance components. The historical entropy matrix is ​​a matrix formed during the normal traffic period of the highway without accidents. The distance component is a subvector of a single row vector containing accident elements in the entropy matrix. The time component is a subvector of the column vector of the column containing the accident element in the entropy matrix. The first element in the time component and the first element in the distance component are the same element in the entropy matrix. The direction of increase of the element number in the distance component is opposite to the direction of travel on the road segment. The direction of increase of the element number in the time component is the same as the direction of increase of the detection signal transmission number. The number of elements in the time component and the distance component are the same. Then, the time component and the distance component are labeled as corresponding vector groups.

[0027] The coefficient of variation for the corresponding vector group in the historical entropy matrix is ​​obtained using the following logic: Get the difference between the second element in the time component and the second element in the distance component, then get the difference between the three elements in the time component and the third element in the distance component, until the difference between all corresponding elements of the two sub-vectors that do not contain the first element is obtained, and the mean squared error of these differences is obtained and labeled as the coefficient of variation. In the logic for obtaining the coefficient of variation, the dependent variable is the coefficient of variation, which is obtained by calculating the mean squared error of the difference between the corresponding elements of the time component and the distance component. The coefficient of variation reflects the degree of coordination between the changes in vibration entropy at the accident location in the time and spatial dimensions: the smaller the value, the more synchronized the temporal evolution and spatial propagation, and the smoother the traffic flow; the larger the value, the more discrepancy between the spatiotemporal patterns, which may indicate the disorder caused by the accident. The independent variables are the elements in the time and distance components. The time component elements represent the vibration intensity of the accident point at different times, reflecting the dynamic changes over time; the distance component elements represent the vibration intensity of adjacent locations of the accident point at the same time, reflecting the spatial range of influence. These correspond to the state of traffic flow over time and the immediate impact of the accident on the surrounding area, respectively. The time component captures the state evolution of the accident point over time (such as the process of vehicles passing by or congestion forming), while the distance component captures the vibration diffusion of the accident point on the surrounding road sections (such as the expansion of the range of influence), such as vehicles behind taking emergency swerves when they see an accident ahead. Together, they characterize the spatiotemporal correlation of traffic events. The coefficient of variation is related to these independent variables because it is calculated based on the differences between them: when temporal and spatial changes match (e.g., smooth traffic flow with coordinated vibration patterns in time and space), the difference is small, and the coefficient of variation is small; when an accident causes a sudden change in time (e.g., an instantaneous collision) and a spatial response (e.g., abnormal surrounding vibrations) that is inconsistent, the difference increases, and the coefficient of variation increases. The influence of independent variables on dependent variables is direct: the closer the corresponding element values ​​of the time and distance components are, the smaller the difference and the smaller the coefficient of variation; conversely, the larger the difference, the larger the coefficient of variation. Specifically, if traffic flow is normal, the independent variables (component elements) show regular changes with limited differences, and the dependent variable (coefficient of variation) remains low; if an accident occurs, the independent variables may fluctuate significantly or become mismatched, the difference widens, and the dependent variable increases significantly. This relationship of magnitude change allows the coefficient of variation to sensitively quantify the degree of disruption of spatiotemporal coordination, thus providing a key indicator for subsequent judgment of the authenticity of an accident.

[0028] In a preferred embodiment, highways have many vehicles traveling at high speeds, which requires vehicles to have a certain degree of stability. The distance component represents the vehicle traffic status before a certain road segment, while the time component represents the vehicle traffic status of vehicles passing through that segment within a certain period of time. Therefore, the performance of a location in the time component is based on the performance in the distance component. Although there may be some changes, due to the stability of the driving state, these changes will not be too large, and these changes can be quantified by constructing a coefficient of variation to form a basis for determining whether an accident has actually occurred.

[0029] As a preferred embodiment, when an accident occurs at a certain location, vehicles behind will definitely slow down and avoid it. Both slowing down and avoiding it will increase the vibration detected by the OTDR system. In the time dimension, the degree of chaos on that road segment will definitely increase because slowing down will increase the number of vehicles and the vibration amplitude. In the spatial dimension, vehicles on that road segment will pass through more orderly and slower or be stuck in traffic. Therefore, the chaos entropy will decrease in time, thus reflecting the variation effect of speed component and time component.

[0030] As a preferred embodiment, the coefficient of variation can be expressed as follows: in, Represents the coefficient of variation. Indicates the first time component within the time component The element and the distance component within the i-th element The difference between the elements, This represents the mean of the differences among all elements. Indicates the variable for element retrieval. , , For the time component, the first time component within the time component Each element and the distance component, and the number of elements within the distance component; Obtain the mean squared error of all corresponding vector groups before the accident signal, and select a threshold such that 90% of the coefficients of variation are less than the threshold. This threshold is then calibrated as the coefficient of variation threshold.

[0031] This step establishes a quantitative statistical baseline model for the "normal traffic flow spatiotemporal coordination pattern." By constructing "time components" and "distance components," it actively probes the inherent correlation between temporal evolution and spatial propagation within the traffic system. Calculating the "coefficient of variation" of these two components essentially measures the deviation between the temporal trajectory of any point on the road and its spatial trajectory at its adjacent points. In smooth, stable traffic flow, vehicle motion is continuous and consistent; therefore, the vibration change at a point should be highly coordinated with the vibration changes at its immediate vicinity (e.g., smooth transmission of traffic waves), resulting in a low coefficient of variation. Determining the "coefficient of variation threshold" by analyzing a large amount of historical data is essentially learning and defining the "upper limit of fluctuation in normal coordination." This threshold becomes an objective benchmark for distinguishing between "natural fluctuations in background traffic flow" and "abnormal discrepancies that may be caused by accidents." This step provides a crucial judgment benchmark for the final decision. It allows the system to no longer view the instantaneous vibration value of the accident point in isolation, but to examine it in a historically formed normal context of "spatiotemporal synergy", providing a statistically significant reference system for step 4 to determine whether the current anomaly is "truly abnormal".

[0032] Step 4: Obtain multiple corresponding vector groups containing accident elements when the accident signal occurs, and obtain the coefficient of variation for each. Generate a variation score based on the rate of change of the coefficient of variation, generate an over-score based on the comparison between the coefficient of variation and the coefficient of variation threshold, construct a variation quantity by combining the variation score and the over-score, set a variation quantity threshold, and compare the variation quantity with the variation quantity threshold to determine whether the accident signal is a misjudgment.

[0033] Step 4 includes the following: Step 401: Obtain the accident element when the accident signal occurs, and construct multiple corresponding vector groups by taking this element as the last element, the second to last element, and so on up to the first element of the time component. Obtain the coefficient of variation for each vector group and form a coefficient of variation sequence. To obtain the rate of change for each element in the coefficient of variation sequence, calculate the variation score for multiple vector groups, the logic is as follows: The weighted sum of all the rate of change values ​​is calculated. When a rate of change is positive, its weight is greater than 1, and when the rate of change is negative, its weight is 0.

[0034] In the logic of obtaining the variation score, the dependent variable is the variation score, which is obtained by weighted summation of the rates of change of an ordered sequence of multiple variation coefficients. The core meaning of the variation score is to quantify whether the degree of disruption to the spatiotemporal coordination around the accident point shows a dynamic trend of continuous deterioration. The higher the value, the more the abnormal pattern not only exists but also intensifies over time, which is a typical characteristic of real accidents (such as continuous congestion, secondary collisions, and more vehicles swerving or slowing down). The independent variable is the rate of change between adjacent elements in the variation coefficient sequence. Each rate of change specifically reflects the rate of increase or decrease of the degree of spatiotemporal disorder at the accident point within a specific time step. In a real environment, a positive rate of change represents an increase in disorder (such as the spread of accident impact), while a negative value represents a decrease in disorder (such as the dissipation of transient disturbances), which directly corresponds to the question of whether the traffic event is continuous or temporary.

[0035] The variance score is directly related to these independent variables (rates of change) because the score itself is a weighted sum of all rates of change. Positive rates of change greater than 1 are weighted, while negative rates of change are weighted at zero. In real traffic accidents, the resulting spatiotemporal disruptions to traffic flow are usually irreversible and cumulatively intensify in the initial stages (manifested as a monotonically increasing or fluctuating increase in the coefficient of variation sequence). Therefore, positive changes are strong evidence of the persistence of the accident; while negative changes often indicate that the system is self-recovering and are more likely characteristics of transient disturbances, thus being ignored in the score. Therefore, the influence of independent variables on the dependent variable is asymmetric and direction-sensitive: the system only "rewards" worsening trends and does not "punish" or ignores mitigating trends.

[0036] The larger the positive rate of change and the more frequently it occurs, the higher the weighted variance score. Conversely, if the rate of change in the sequence is mostly negative or has very small positive values, the score will be low. This means that even if the coefficient of variation is high at a certain moment (single-point anomaly), if its subsequent rate of change quickly turns negative (the anomaly dissipates rapidly), it cannot accumulate into a high variance score. This mechanism ensures that only anomalous signals that show a continuous expansion and deterioration trend over time can obtain high scores and thus be judged as real accidents in decision-making. This effectively distinguishes between transient impulse disturbances and the evolving impact of accidents.

[0037] This sub-step introduces "dynamic evolution process analysis" to distinguish between genuine and false accidents. It goes beyond static snapshot analysis of a single moment at the accident site, constructing "variance coefficient sequences" of different time lengths to recreate the entire process of anomaly patterns from their inception and emergence to their potential development. The "rate of change" of this sequence is calculated, and weighting rules are designed (positive change weight is greater than 1, negative change weight is 0). The traffic disturbances caused by real traffic accidents (such as collisions and rollovers) typically accumulate and spread over time, manifested as a continuously increasing coefficient of variation (positive change). Conversely, many false alarms caused by transient disturbances (such as roadside impacts or large vehicles passing by) often exhibit sudden and short-lived anomalies, with the coefficient of variation potentially rising briefly before rapidly declining (negative change). By focusing on the positive trend through weighted summation, the "variance score" effectively captures and amplifies the "continuous deterioration" signal unique to real accidents, while suppressing the "pulsating" signal of transient disturbances. This step of analysis deepens the judgment criteria from "how strong the anomaly is" to "how the anomaly evolves", greatly enhancing the ability to identify false alarms.

[0038] Step 402: Compare the coefficient of variation sequence with the coefficient of variation threshold to obtain the excess score, as follows: Subtract the coefficient of variation threshold from each value in the coefficient of variation sequence to obtain the difference. Sum all the results with positive differences. Determine the order of the first element with a positive difference in the coefficient of variation to generate a weight. The earlier the element is in the order, the greater its weight. The later the element is in the order, the smaller its weight. The weight is always greater than 0. Multiply the sum by the weight to obtain the overweight score. The excess score, as the dependent variable, quantifies the comprehensive anomalous intensity of the spatiotemporal disturbances caused by the accident signal in terms of both static amplitude and timing. It is calculated by summing the differences between each value in the coefficient of variation sequence and the historical coefficient of variation threshold, summing only the positive differences, and then multiplying by a weight generated based on the ranking of the first positive difference element. Its meaning lies in assessing whether the anomaly significantly exceeds the historical normal fluctuation range and emphasizing the severity of early excesses—in a real-world environment, this directly corresponds to whether the accident's impact is sufficiently strong and rapidly manifested, rather than merely a matter of instantaneous fluctuations. The independent variables include the coefficient of variation sequence (reflecting the change in the degree of disturbance over time), the coefficient of variation threshold (the upper limit of historical normal synergy), and the weights (determined by the ranking of the first excess element). The excess score is related to these independent variables because the sum of the positive differences captures the overall excess, while the weights amplify the importance of early excesses: the earlier the ranking, the greater the weight, reflecting the characteristic that the accident's impact often exceeds the normal limit early on. The influence of the independent variable on the dependent variable is synergistic: the larger the positive difference and the earlier it appears, the higher the excess score; conversely, the smaller the difference or the later the excess occurs, the lower the score. Specifically, the more the coefficient of variation sequence values ​​exceed the threshold and the larger the difference, the greater the sum of the values. At the same time, the earlier the first positive difference is in the sequence (such as the first or second element), the greater its weight. The product of these two factors drives the excess score to rise, which makes the system more sensitive to early and strong anomalous signals.

[0039] The mutation score and the excess score are summed to obtain the mutation amount. A mutation amount threshold is set. When the mutation amount exceeds the mutation amount threshold, it is determined that an accident has indeed occurred at that location and an alarm is triggered. Otherwise, the location is determined to be a false alarm and no alarm is triggered.

[0040] The variability, as the dependent variable, is the final decision indicator that integrates dynamic trends and static exceedance assessments. It is obtained by summing the variability score and the exceedance score. Its significance lies in providing a comprehensive criterion to simultaneously consider the worsening trend of the anomalous signal (reflected by the variability score) and the intensity and timing of the exceedance (reflected by the exceedance score), thereby more reliably distinguishing between real accidents and misjudgments in real-world environments. The independent variables are the variability score and the exceedance score, which capture different dimensions of the anomaly: the variability score emphasizes the continuous escalation of the disorder, while the exceedance score emphasizes the absolute level and early nature of the disorder. The variability is directly related to these two independent variables because its value is their sum, reflecting an additive relationship—when both are high, the variability increases significantly, indicating a high probability of an accident; if only one is high, the variability is moderate; when both are low, the variability is low, potentially indicating a misjudgment. Regarding the relationship between magnitude and variation, the larger the variation score and the excess score, the greater the variation. This means that the system will only generate a high variation when the abnormal signal simultaneously shows a continuously deteriorating dynamic trend and an early significant static excess, thereby exceeding the preset threshold to trigger an alarm, thus improving the accuracy and robustness of decision-making.

[0041] This section constructs a weighted voting mechanism that integrates "trend strength," "excess magnitude," and "anomaly timing." When calculating the "excess score," it not only summarizes the total amount of coefficients of variation exceeding historical benchmarks, but more importantly, introduces weights based on the "ranking of the first excess point." This gives decision-making value to the "timing of anomalies": if an incident occurs immediately or is still in its early stages, its impact can exceed the historical normal fluctuation range, which is itself a strong risk indicator signal. Therefore, the earlier the sustained excess occurs, the greater the weight, which aligns with the safety management principle of "early warning, early judgment." Finally, the "variance score" (representing a worsening trend) and the "excess score" (representing static excess and early occurrence) are added to obtain the "variance quantity," achieving a fusion assessment of the multi-dimensional characteristics of the anomaly signal. Setting a "variance quantity threshold" and comparing it completes the transformation from continuous quantitative assessment to binary classification decision-making. This comprehensive criterion has much higher robustness and accuracy compared to single threshold judgments (such as looking only at amplitude or only at a single coefficient of variation). It ensures that only those signals that simultaneously worsen in trend, significantly exceed the limit in intensity, and appear early in time will be ultimately determined as real incidents, thereby maximizing the ultimate goal of "reducing false alarms and not missing real alarms".

[0042] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0043] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0044] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0045] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring highway accidents, characterized in that, The specific steps include: Step 1: Obtain vibration images at each location on the optical fiber by detecting signals multiple times. Set a distance window to slide on the vibration images and use the average of all vibration peak points within the distance window as the chaos entropy. Construct an entropy matrix using the chaos entropy generated by a single vibration image as a row vector. Set a time window, and the number of times the detection signal is emitted within the time window is the dimension of the entropy matrix. Step 2: When an accident signal is detected, obtain the accident location. Based on the distance between the accident location and the point where the detection signal was emitted, determine the element corresponding to the accident location in the entropy matrix and mark it as an accident element. Step 3: Set the sub-vectors of the column vectors containing accident elements as time components, and the sub-vectors of the row vectors containing accident elements as distance components. Label the corresponding time components and distance components as corresponding vector groups. The correspondence means that the first element of the time component and the distance component are the same. Use the mean squared error in the corresponding vector group as the coefficient of variation, obtain the coefficient of variation of all corresponding vector groups in the historical entropy matrix, and select the coefficient of variation threshold. Step 4: Obtain multiple corresponding vector groups containing accident elements when the accident signal occurs, and obtain the coefficient of variation for each. Generate a variation score based on the rate of change of the coefficient of variation, generate an over-score based on the comparison between the coefficient of variation and the coefficient of variation threshold, construct a variation quantity by combining the variation score and the over-score, set a variation quantity threshold, and compare the variation quantity with the variation quantity threshold to determine whether the accident signal is a misjudgment.

2. The method for monitoring highway accidents according to claim 1, characterized in that: When the OTDR system is used, vibration images of various positions on the optical fiber are acquired each time a detection signal is transmitted. The vibration image is a two-dimensional rectangular coordinate system image, with the horizontal axis representing distance and the vertical axis representing amplitude. The origin of the distance is the detection signal transmission point of the OTDR system, and the distance value represents the distance to the detection signal transmission point of the OTDR system. Each wave peak is marked as a vehicle point, and the amplitude and distance values ​​of each vehicle point are recorded. All vibration images are sorted according to the transmission time of the detection signal to form a vibration image sequence.

3. The method for monitoring highway accidents according to claim 2, characterized in that: A distance window and a time window are set respectively. The distance window slides on the vibration image of a single transmission detection signal. A chaos entropy is constructed based on the number of vehicle points within the sliding window and the amplitude of each vehicle point. The chaos entropy represents the average amplitude of all vehicle points within the distance window. The entropy matrix is ​​constructed using the following logic: As the distance window slides across the vibration image of a single detection signal, the average amplitude within each distance window is an element of the row vector of the entropy matrix, and the element number corresponds to the number of times the distance window slides. The chaotic entropy sequence formed by each single transmission detection signal is a row vector, and the number of all vibration images within the time window is the dimension of the entropy matrix.

4. The method for monitoring highway accidents according to claim 4, characterized in that: When the OTDR system detects an accident signal to determine that an accident may have occurred at a certain location, it obtains the accident location and performs element-wise mapping of the accident location in the entropy matrix. The mapping logic is as follows: In a single row vector, obtain all distance windows including the accident location and all corresponding elements of the distance windows in the entropy matrix. When the number of all corresponding elements is odd, take the element corresponding to the middle position of all corresponding elements as the element corresponding to the accident location in the entropy matrix. When the number of all corresponding elements is even, take any one of the two elements at the middle position of all corresponding elements as the element corresponding to the accident location in the entropy matrix. The element corresponding to the accident location in the entropy matrix is ​​designated as the accident element.

5. The method for monitoring highway accidents according to claim 4, characterized in that: Obtain the historical entropy matrix and set time and distance components. The historical entropy matrix is ​​a matrix formed during the normal traffic period of the highway without accidents. The distance component is a subvector of a single row vector containing accident elements in the entropy matrix. The time component is a subvector of the column vector of the column containing the accident element in the entropy matrix. The first element in the time component and the first element in the distance component are the same element in the entropy matrix. The direction of increase of the element number in the distance component is opposite to the direction of travel on the road segment. The direction of increase of the element number in the time component is the same as the direction of increase of the detection signal transmission number. The number of elements in the time component and the distance component are the same. Then, the time component and the distance component are labeled as corresponding vector groups.

6. The method for monitoring highway accidents according to claim 5, characterized in that: The coefficient of variation for the corresponding vector group in the historical entropy matrix is ​​obtained using the following logic: Get the difference between the second element in the time component and the second element in the distance component, then get the difference between the three elements in the time component and the third element in the distance component, until the difference between all corresponding elements of the two sub-vectors that do not contain the first element is obtained, and the mean squared error of these differences is obtained and labeled as the coefficient of variation. Obtain the mean squared error of all corresponding vector groups before the accident signal, and select a threshold such that 90% of the coefficients of variation are less than the threshold. This threshold is then calibrated as the coefficient of variation threshold.

7. A method for monitoring highway accidents according to claim 6, characterized in that: Obtain the accident element at the time of the accident signal, and construct multiple corresponding vector groups by taking this element as the last element, the second to last element, and so on up to the first element of the time component. Obtain the coefficient of variation for each vector group and form a coefficient of variation sequence. To obtain the rate of change for each element in the coefficient of variation sequence, calculate the variation score for multiple vector groups, the logic is as follows: The weighted sum of all the rate of change values ​​is calculated. When a rate of change is positive, its weight is greater than 1, and when the rate of change is negative, its weight is 0.

8. The method for monitoring highway accidents according to claim 7, characterized in that: The excess score is obtained by comparing the coefficient of variation sequence with the coefficient of variation threshold, as follows: Subtract the coefficient of variation threshold from each value in the coefficient of variation sequence to obtain the difference. Sum all the results with positive differences. Determine the weight of the first element with a positive difference in the coefficient of variation by its order. The earlier the element is in the order, the greater its weight. The later the element is in the order, the smaller its weight. The weight is always greater than 0. Multiply the sum by the weight to obtain the overweight score.

9. A method for monitoring highway accidents according to claim 8, characterized in that: The mutation score and the excess score are summed to obtain the mutation amount. A mutation amount threshold is set. When the mutation amount exceeds the mutation amount threshold, it is determined that an accident has indeed occurred at that location and an alarm is triggered. Otherwise, the location is determined to be a false alarm and no alarm is triggered.

Images