A rule grid-based roadside perception anomaly real-time detection method and system

Abstract

The application provides a rule grid-based roadside perception anomaly real-time detection method and system, which comprises the following steps: preprocessing given map data, monitoring area data and road test perception data to obtain a preprocessed road test perception data sequence; performing anomaly detection on the preprocessed road test perception data sequence by using a rule network composed of 18 types of meta-rules, and obtaining a detection result sequence; performing quality index calculation on the detection result sequence by using a K-means algorithm, and obtaining a quality level set; performing anomaly tracing on the anomaly detection sequence by using the quality level set, and obtaining an anomaly tracing result. The application performs anomaly detection on the preprocessed perception data sequence by using the rule network composed of 18 types of meta-rules, comprehensively identifies anomaly events in multiple scenes, improves the coverage rate of anomaly detection, and guarantees the real-time performance and efficiency of the detection process based on the design of the rule grid structure.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic management, and in particular to a method and system for real-time detection of roadside perception anomalies based on a rule grid. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, vehicle-road-cloud integrated systems have become an important foundation for promoting intelligent transportation. However, the complex real-time data interaction and high dependence on multi-dimensional fused data in this system make the dynamic monitoring of cloud data quality a critical issue that urgently needs to be addressed. It is necessary to ensure that the multi-source heterogeneous data interacting in real time meets the operational requirements of intelligent transportation systems in terms of key quality dimensions such as accuracy, completeness, consistency, timeliness, and reliability.

[0003] Currently, research on anomaly detection for time-series data can be mainly summarized into the following categories of methods:

[0004] 1. Traditional Anomaly Detection Methods: Traditional anomaly detection methods are mostly based on statistical analysis and classical machine learning models, and can be broadly classified into three categories: statistical thresholding (such as 3σ, GMM), classical machine learning (such as K-Means, Isolation Forest, SVM), and linear dimensionality reduction methods (such as PCA). Rajbahadur et al. pointed out that statistical thresholding methods rely on strict distribution assumptions and are prone to failure when the data is heavy-tailed or multimodal; classical machine learning requires manual feature engineering, and the detection rate drops rapidly in high-dimensional scenarios; linear dimensionality reduction can only capture linear relationships, and the false positive rate increases significantly after losing nonlinear temporal features. The common feature of traditional methods is that they are based on static and time-invariant models to detect anomalies. When faced with high-dimensional time-series data generated by intelligent connected vehicles, these methods often struggle to capture their temporal dependencies and complex dynamic evolution patterns, thus limiting their detection performance.

[0005] 2. Deep Learning-Based Anomaly Detection Methods: With the widespread application of deep learning in sequence modeling, recurrent neural networks (RNNs) and their variants, such as long short-term memory networks (LSTM), have demonstrated superior time series modeling capabilities. Qin proposed an LSTM-based CAN bus anomaly detection method, comparing detection performance in binary and hexadecimal formats using five loss functions. However, this method is only suitable for low-dimensional, fixed-period vehicle network message sequences, making it difficult to directly adapt to complex anomaly scenarios in roadside perception, characterized by high dimensions, irregular sampling, and spatial relationships. Lu proposed F-SE-LSTM, which combines FFT and SENet-LSTM in the frequency domain to improve the accuracy of temporal anomaly detection. However, this method requires a sliding window of uniform length for input, and the frequency matrix becomes incomplete when roadside data is dynamically missing, leading to detection failure. Furthermore, it lacks the ability to model multi-target interactions and spatial anomalies. TranAD proposes a Transformer-based self-conditional adversarial network that can capture long-range dependencies and locate anomalous dimensions in parallel on multivariate time-series data. However, this method requires the input dimension to be fixed and persistent, lacks modeling of spatial relationships, cannot handle dynamic dimension changes caused by factors such as target flickering in the data, and lacks modeling of spatial relationships, making it difficult to identify anomalies that depend on geometric spatial interactions.

[0006] 3. Anomaly Detection Methods for Intelligent Connected Vehicle Data: In recent years, deep learning algorithms have received widespread attention due to their significant success in image recognition, speech processing, and natural language processing. In particular, deep learning models such as Recurrent Neural Networks (RNNs) and Long Short-Term Memory Networks (LSTMs) have demonstrated their powerful capabilities in processing time-series data. Researchers have gradually explored the application of LSTMs to anomaly detection tasks in multi-dimensional sensor data generated by intelligent vehicles. Tang et al. introduced a deep learning model combining Long Short-Term Memory Networks (LSTMs) and Convolutional Neural Networks (CNNs) to improve the anomaly detection performance of time-series data from intelligent connected vehicles. By utilizing the long-term dependency learning capability of LSTMs and the local feature extraction capability of CNNs, this model can effectively identify abnormal behaviors in time-series data. However, it is mainly suitable for single anomaly detection and cannot handle multiple anomalies or anomalies involving multiple object interactions. Summary of the Invention

[0007] In view of the above situation, the main objective of this invention is to propose a real-time roadside sensing anomaly detection method based on regular grids to solve the above-mentioned technical problems.

[0008] This invention proposes a real-time anomaly detection method for roadside sensing based on a regular grid, the method comprising the following steps:

[0009] Step 1: Preprocess the given map data, monitoring area data, and road test sensing data in the road test sensing data to obtain the preprocessed road test sensing data sequence;

[0010] Step 2: Use a rule network consisting of 18 types of meta-rules to perform anomaly detection on the preprocessed road test perception data sequence to obtain the detection result sequence;

[0011] Step 3: Based on the detection result sequence, use the K-means algorithm to calculate the quality index and obtain the quality level set;

[0012] Step 4: Use the quality level set to trace the source of the anomaly detection sequence and obtain the anomaly tracing results.

[0013] This invention also proposes a real-time anomaly detection system for roadside sensing based on a regular grid, the system comprising:

[0014] The preprocessing module is used for:

[0015] The given map data, monitoring area data, and road test perception data in the road test perception data are preprocessed to obtain the preprocessed road test perception data sequence;

[0016] The rules network module is used for:

[0017] Anomaly detection is performed on the preprocessed road test perception data sequence using a rule network consisting of 18 types of meta-rules to obtain the detection result sequence;

[0018] The quality indicator calculation module is used for:

[0019] Based on the detection result sequence, the K-means algorithm is used to calculate the quality index and obtain a set of quality levels.

[0020] The anomaly tracing module is used for:

[0021] Anomaly tracing results are obtained by using a set of quality levels to trace the origin of anomalies in the detected sequences.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] 1. This invention uses a rule network composed of 18 types of meta-rules to perform anomaly detection on the preprocessed perception data sequence, thereby achieving comprehensive identification of abnormal events in multiple scenarios, improving the coverage of anomaly detection, and the design based on the rule grid structure ensures the real-time performance and efficiency of the detection process.

[0024] 2. This invention uses the K-means algorithm to calculate quality indicators based on the detection result sequence, thereby obtaining a set of quality levels. This enables objective clustering and grading of anomaly detection results, avoiding the uncertainty of manual judgment and improving the scientificity and objectivity of anomaly detection result evaluation. Attached Figure Description

[0025] Figure 1 This is a flowchart of the real-time anomaly detection method based on regular grids proposed in this invention;

[0026] Figure 2 This is the anomaly detection rule grid diagram of the roadside sensing anomaly real-time detection method based on rule grid proposed in this invention;

[0027] Figure 3 This is a schematic diagram of the overall framework of the roadside perception anomaly real-time detection system based on regular grids proposed in this invention. Detailed Implementation

[0028] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0029] These and other aspects of the embodiments of the present invention will become clear from the following description and accompanying drawings. In these descriptions and drawings, some specific embodiments of the present invention are specifically disclosed to illustrate some ways of implementing the principles of the embodiments of the present invention; however, it should be understood that the scope of the embodiments of the present invention is not limited thereto.

[0030] Please see Figure 1 This invention proposes a real-time anomaly detection method for roadside sensing based on a regular grid, which includes the following steps:

[0031] Step 1: Preprocess the given map data, monitoring area data, and road test perception data in the road test perception data to obtain the preprocessed road test perception data sequence.

[0032] Step 2: Use a rule network consisting of 18 types of meta-rules to perform anomaly detection on the preprocessed road test perception data sequence to obtain the detection result sequence;

[0033] In step 2, anomaly detection is performed on the preprocessed road test perception data sequence using a rule network composed of 18 types of meta-rules to obtain the detection result sequence. The specific steps are as follows:

[0034] S201. Using the target position, target speed and merged monitoring area obtained from the preprocessed road test perception data sequence, determine whether the target is within the perception area, so as to simultaneously exclude the possibility of driving out of the detection area or entering from outside the detection area in the next moment, so as to obtain the first type of meta rule.

[0035] S202. Obtain a set of candidate flashing target IDs from the set of target IDs in adjacent messages obtained from the preprocessed road test sensing data sequence. Use the set of candidate flashing target IDs to determine whether the target is flashing in order to obtain the second type of meta-rule.

[0036] S203. Based on the target position, target speed and target heading angle obtained from the preprocessed road test perception data sequence, the latitude and longitude coordinates of the predicted position are obtained. The latitude and longitude coordinates of the predicted position are used to determine whether the target before and after the flashing is the same target, so as to obtain the third type of meta rule.

[0037] S204. Using the latitude and longitude coordinates of the predicted location, the location coordinates of the suddenly appearing target obtained in the preprocessed road test perception data sequence are judged to determine whether the ID of the target has changed before and after the flashing, so as to obtain the fourth type of meta-rule.

[0038] S205. Using the target size parameters and target heading angle in the preprocessed road test perception data sequence, the actual latitude and longitude coordinates of the four corner points of the target are obtained. The actual latitude and longitude coordinates of the four corner points of the target are used to determine whether there is overlap or line-crossing anomaly, so as to obtain the fifth type of meta-rule.

[0039] S206. Based on the map parameters and target latitude and longitude coordinates in the preprocessed road test perception data sequence, obtain the set of targets that overlap the lane lines. Use the set of targets that overlap the lane lines to determine whether there is an overlap phenomenon in the target domain, so as to obtain the sixth type of meta-rule.

[0040] S207. Using the target set for crossing the line, the target set for instantaneous large-scale crossing is obtained by using the total number of targets appearing in the message and the preset threshold for instantaneous large-scale crossing judgment. The target set for instantaneous large-scale crossing is used to determine whether instantaneous large-scale crossing has occurred, so as to obtain the seventh type of meta rule.

[0041] S208. Based on the target line-over set and map parameters, obtain the grid of large-scale line overtaking in the area. Use the grid of large-scale line overtaking in the area to determine whether large-scale line overtaking occurs in the area within a time period, so as to obtain the eighth type of meta-rule.

[0042] S209. Using the target types in adjacent messages, obtain the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after flashing. Using the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after flashing, determine whether the target is repeatedly identified as a different type, so as to obtain the ninth meta-rule and the tenth meta-rule.

[0043] S210. Use the speeds of two adjacent targets, the heading angles of two adjacent targets, and the dimensions of two adjacent targets to obtain the target set where target splitting occurs. Use the target set where target splitting occurs to determine whether target splitting has occurred, so as to obtain the eleventh type meta-rule.

[0044] S211. Obtain the coordinates of the four corner points of the approximate rectangles of adjacent targets from the preprocessed road test perception data sequence to obtain the intersection area. Use the intersection area to determine whether the adjacent targets overlap to obtain the twelfth type meta-rule.

[0045] S212. Determine whether the target is turning based on the target heading angle and the distance the target has moved, in order to obtain the thirteenth type of meta-rule;

[0046] S213. Using the target heading angle and the preset heading angle anomaly judgment threshold, determine whether the target generates heading angle data anomalies when stationary, so as to obtain the fourteenth type meta-rule.

[0047] S214. Based on the road direction angle and target heading angle obtained from the preprocessed road test perception data sequence, determine whether the target has a heading angle deviation value in order to obtain the fifteenth type meta rule;

[0048] S215. Calculate the direction vector using the latitude and longitude of adjacent targets, and determine whether there is an unreasonable trajectory of the target based on the changes in the direction vector data, so as to obtain the sixteenth type of meta-rule;

[0049] S216. Based on the coordinate points of adjacent lanes obtained from the preprocessed road test perception data sequence, the lane direction is obtained. The lane direction is used to determine whether there is an anomaly in the target driving direction, so as to obtain the seventeenth meta-rule.

[0050] S217. Calculate the shortest distance between target coordinates using the latitude of adjacent targets and the Earth's radius. Determine whether the target is jumping by using the shortest distance between target coordinates to obtain the eighteenth meta-rule.

[0051] S218. Using the first to eighteenth categories of meta-rules obtained from S201 to S217, the data in the preprocessed road test perception data sequence are judged and labeled to obtain the detection result sequence;

[0052] In the process of determining whether a target is within the sensing area by using the target position, target speed, and merged monitoring area obtained from the preprocessed road test sensing data sequence, in order to simultaneously exclude the possibility of the target leaving the detection area or entering from outside the detection area in the next moment, and in order to obtain the first type of meta-rule, the following relationship exists:

[0053] ;

[0054] in, This refers to targets within the perception area. express Time of the first The location of the target Indicates the merged monitoring area. express arrive The shortest distance to the boundary, express Time of the first The speed of the target This represents the time difference between two consecutive messages. This represents a predefined distance threshold; Indicates that the target is in At the same time, arrive The shortest distance to the boundary is less than The sum of the theoretical distance moved;

[0055] Furthermore, this rule is used to determine whether the target is within the perception area, that is, to determine whether the target is within the perception area, while excluding the possibility that the target may leave the perception area or has just entered from outside the perception area in the next moment.

[0056] In the process of obtaining a set of candidate blinking target IDs from the set of target IDs within adjacent messages obtained from the preprocessed drive test sensing data sequence, and using the set of candidate blinking target IDs to determine whether a target is blinking in order to obtain the second type of meta-rule, the following relationship exists:

[0057] ;

[0058] in, express The set of target IDs within the time message. express The set of target IDs within the time message; This represents the set of target IDs where the previous message did not exist and the next message suddenly appeared. This represents the set of target IDs where the previous message existed and the next message suddenly disappeared.

[0059] Furthermore, this rule is used to determine whether a target suddenly appears or disappears, that is, to compare the occurrence or disappearance of a target in two consecutive messages in order to generate a flashing target;

[0060] Based on the target position, target velocity, and target heading angle obtained from the preprocessed road test sensing data sequence, the latitude and longitude coordinates of the predicted position are obtained. The latitude and longitude coordinates of the predicted position are then used to determine whether the target before and after the flash is the same target, thus obtaining the third-type meta-rule. The specific steps for obtaining the latitude and longitude coordinates of the predicted position are as follows:

[0061] Based on the target speed and the time difference between the two consecutive messages, the distance increment is calculated, and the corresponding relationship is as follows:

[0062] ;

[0063] in, Indicates the distance increment;

[0064] Using the distance increment, target heading angle, and Earth's average radius, the latitude increment and longitude increment are calculated respectively. The corresponding relationship is as follows:

[0065] ;

[0066] in, Indicates the latitude increment. Indicates the increment of longitude. express Time of the first The heading angle of the target, This represents the average Earth radius. Indicates the latitude of the current location;

[0067] Using the latitude increment, longitude increment, current latitude, and current longitude, the latitude and longitude of the predicted location in Germany and Austria are calculated. The corresponding relationship in this process is as follows:

[0068] ;

[0069] in, Indicates the latitude of the predicted location. Indicates the longitude of the predicted location. The latitude and longitude coordinates of the predicted location. Indicates the longitude of the current location;

[0070] Furthermore, this rule is used to match targets that reappear after disappearing from the predicted location and to determine whether they are likely the same target.

[0071] In the process of using the latitude and longitude coordinates of the predicted location to determine the location coordinates of suddenly appearing targets obtained from the preprocessed road test perception data sequence, in order to determine whether the target ID has changed before and after the flashing, and in order to obtain the fourth type of meta-rule, the following relationship exists:

[0072] ;

[0073] in, This represents the set of targets that can be matched by a given location. Indicates the location coordinates of the suddenly appearing target; This represents a predefined distance matching threshold, with a value ranging from 0 to 3 meters.

[0074] Furthermore, in cases of target blinking anomalies, there are situations where the ID changes after the target disappears and then reappears. This rule is used to match the predicted location of a suddenly disappearing target with the location of a suddenly appearing target to determine whether the two targets are the same.

[0075] The specific steps for obtaining the actual latitude and longitude coordinates of the four corner points of the target by using the target size parameters and target heading angle in the preprocessed road test sensing data sequence, and using the actual latitude and longitude coordinates of the four corner points to determine whether there is overlap or line-crossing anomalies in order to obtain the fifth type of meta-rule are as follows:

[0076] The local coordinates of the four corner points relative to the center point of the target are obtained from the target's size parameters. The corresponding relationship in this process is as follows:

[0077] ;

[0078] in, This represents the local coordinates of the four corner points relative to the center point of the target. Indicates the length of the target. Indicates the width of the target;

[0079] The local coordinates of the four corner points relative to the target center point are rotated based on the target heading angle to obtain the rotated coordinates. The corresponding relationship in this process is as follows:

[0080] ;

[0081] in, Represents the coordinates after rotation;

[0082] The actual latitude and longitude coordinates of the four corner points of the target are obtained by using the rotated coordinates and the average radius of the Earth. The corresponding relationship in this process is as follows:

[0083] ;

[0084] in, This represents the actual latitude and longitude coordinates of the four corner points of the target; This indicates the east-west offset of the corner point relative to the target center point after rotation. This indicates the north-south offset of the corner point relative to the target center point after rotation. Indicates the longitude conversion factor. Indicates the latitude conversion factor;

[0085] Furthermore, in this rule, the target is approximated as a rectangle in order to calculate the coordinates of the four corner points of the target;

[0086] and All of these represent conversion factors, which are calculated from the Earth's average radius and the arc length corresponding to latitude and longitude. These conversion factors can convert rotating coordinate units into latitude and longitude.

[0087] The specific steps for obtaining the target line-crossing set are as follows: Based on the map parameters and target latitude and longitude coordinates in the preprocessed road test perception data sequence, a set of line-crossing targets is obtained. Then, this set is used to determine whether lane lines in the target domain overlap, in order to derive the sixth type of meta-rule.

[0088] Using the target's latitude and longitude coordinates and the unit size of the grid, the grid index where the target is located is calculated. The corresponding relationship in this process is as follows:

[0089] ;

[0090] in, Indicates the grid index where the target is located. Indicates the unit size for dividing the grid;

[0091] Use the grid index of the target to find the set of lane lines within the target grid in the map data;

[0092] Intersection detection is performed between the set of lane lines within the target grid and the target to obtain the set of targets crossing the lane lines. The corresponding relationship in this process is as follows:

[0093] ;

[0094] in, Represents the set of targets that press the line. This indicates the calculation of the intersection length. express The first moment One goal, Represents the set of lane lines within the target grid. Indicates the intersection length threshold;

[0095] In the process of using the target set for instantaneous large-scale overpasses, obtaining the target set for instantaneous large-scale overpasses by combining the total number of targets appearing in the message with a preset threshold for instantaneous large-scale overpass judgment, and using the target set for instantaneous large-scale overpasses to determine whether an instantaneous large-scale overpass situation has occurred, in order to obtain the seventh type of meta-rule, the following relationship exists:

[0096] ;

[0097] in, This represents the set of targets where a large number of items simultaneously press against the line. express Total number of targets at any given time; This represents the preset threshold for judging a large number of instantaneous lines being pressed, with a value of 1 / 3.

[0098] Furthermore, large-volume overlay is divided into two cases: instantaneous large-volume overlay and regional large-volume overlay within a time period. This rule applies to instantaneous large-volume overlay.

[0099] The term "instantaneous large-scale line overlap" refers to the phenomenon where, at a certain point in time, more than a certain proportion of targets on the entire map are overlapping the line.

[0100] To determine if a large number of targets are crossing the line at a given moment, divide the number of targets crossing the line at that moment by the total number of targets to calculate the percentage of targets crossing the line at that moment. If it exceeds... If so, it is considered that a large number of lines were pressed at that moment;

[0101] In the process of obtaining a grid of large-scale line crossings in the area based on the target line crossing set and map parameters, and using this grid to determine whether large-scale line crossings occurred within a specific time period to obtain the eighth type of meta-rule, the following relationship exists:

[0102] ;

[0103] in, This indicates a grid representing a large number of lines being pressed into the area. Indicates the start time. Indicates the monitoring duration. Representing different grids, Represents the entire map grid; This indicates the preset threshold for the number of times the wires are pressed within the grid, with a value ranging from 3 to 5 times.

[0104] Furthermore, large-scale line pressing is divided into two cases: instantaneous large-scale line pressing and regional large-scale line pressing within a time period. This rule applies to regional large-scale line pressing within a time period.

[0105] statistics arrive The number of times each grid cell crosses the line within a given time period. If the number of times a certain grid cell crosses the line exceeds a certain threshold... If so, it is considered that a large number of people have crossed the line in the area during that period.

[0106] In the process of obtaining the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after blinking by utilizing the target types in adjacent messages, and then determining whether a target is repeatedly identified as a different type to obtain the ninth and tenth meta-rules, the specific steps for obtaining the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after blinking are as follows:

[0107] The relationship between obtaining the target types in the message and the types of all targets at the same time is as follows:

[0108] ;

[0109] in, Indicates time All target types, Indicates the first The type of target, Indicates the first One goal, Indicates time The message;

[0110] By utilizing the types of all targets at the same time, we obtain a set of targets repeatedly identified into different types. The corresponding relationship in this process is as follows:

[0111] ;

[0112] in, This indicates that the target is repeatedly identified as a set of different types. Indicates the first The goal is at any time The type of disappearance Indicates the first The goal is at any time The type of disappearance;

[0113] Furthermore, the repeated identification of targets into different types includes two cases: targets with the same ID are repeatedly identified into different types; and both the ID and type change after the target flashes.

[0114] This rule applies when a target with the same ID is repeatedly identified as a different type.

[0115] Using the types of all targets at the same time, the set of targets that are repeatedly identified as different types after flashing is obtained. The corresponding relationship in the process is as follows:

[0116] ;

[0117] in, This indicates that the target is repeatedly identified as a set of different types after it blinks. Indicates the first Target time interval The type when it reappears;

[0118] Furthermore, the repeated identification of targets into different types includes two cases: targets with the same ID are repeatedly identified into different types; and both the ID and type change after the target flashes.

[0119] This rule applies when both the ID and type change after the mark flashes;

[0120] In the process of obtaining the target set where target splitting occurs using the velocities, heading angles, and dimensions of two adjacent targets, and then using this target set to determine whether target splitting has occurred in order to derive the eleventh type of meta-rule, the following relationship exists:

[0121] ;

[0122] in, This represents the set of targets where target splitting occurs. express Time of the first One goal, express Time of the first The speed of the target; This represents the threshold for the velocity difference between two targets, with a value of 5 m / s. Indicates the first The size of the target Indicates the first The size of the target; This represents the threshold for the size difference between two targets, with a value ranging from 0 to 10 dm. express Time of the first The heading angle of the target; This represents the threshold for the difference in heading angle between two targets, with a value of 30°.

[0123] Furthermore, this rule is used to determine whether target splitting has occurred, that is, to determine whether the speed, heading angle and size of two close targets (one of which appears suddenly) are similar. If they are similar, it can be considered that target splitting has occurred.

[0124] The process of obtaining the intersection area by approximating the four corner coordinates of adjacent target rectangles from the preprocessed road test perception data sequence, and using the intersection area to determine whether adjacent targets overlap, in order to obtain the twelfth type meta-rule, also includes the following sub-steps:

[0125] The proximity of two targets is determined by using their latitude and longitude coordinates and a preset distance threshold. The relationship in this process is as follows:

[0126] ;

[0127] in, Indicates the first The longitude of the target Indicates the first The dimensions of each target This represents a preset distance threshold between targets;

[0128] When two targets are determined to be close in location, the intersection region is obtained by approximating the coordinates of the four corner points using the rectangles of adjacent targets. The corresponding relationship in this process is as follows:

[0129] ;

[0130] in, Indicates the intersection region. Indicates the first The rectangular approximation of the four corner points of the target. Indicates the first The rectangular approximation of the four corner points of the target;

[0131] By using the intersection region and a preset area threshold, it is determined whether adjacent targets overlap. The corresponding relationship in this process is as follows:

[0132] ;

[0133] in, This represents the set of targets where overlap occurs. This indicates whether the overlapping area of ​​the targets is greater than a preset area threshold. This indicates the preset area threshold, with a value of 0.05㎡;

[0134] Furthermore, this rule is used to determine whether polygons intersect, that is, to determine whether the polygons formed by approximations of two targets intersect. If they intersect, it can be considered that the two targets overlap.

[0135] In determining whether a target is turning based on its heading angle and the distance it has traveled, in order to derive the thirteenth meta-rule, the relevant relationship is as follows:

[0136] ;

[0137] in, This describes the rules for determining whether a target should turn. This represents the change in the target's heading angle between two time points. It represents the distance the target has moved between two points in time. Indicates the threshold for change in heading angle. Indicates the location distance threshold;

[0138] Furthermore, in this rule, if the output is "yes", the target is considered to be in the process of turning; otherwise, it is not.

[0139] In the process of determining whether a target generates abnormal heading angle data when stationary by using the target heading angle and a preset heading angle anomaly detection threshold, in order to obtain the fourteenth type of meta-rule, the following relationship exists:

[0140] ;

[0141] in, express Time of the first The heading angle of the target, This indicates the preset threshold for determining abnormal heading angles;

[0142] Furthermore, since the heading angle is periodic in the data message, when the heading angle changes by more than 180°, the heading angle change threshold is adjusted. The corresponding relationship is as follows:

[0143] ;

[0144] in, This indicates the adjusted threshold.

[0145] In the process of determining whether the target has a heading angle deviation value based on the road direction angle and target heading angle obtained from the preprocessed road test perception data sequence, in order to obtain the fifteenth type meta-rule, the following relationship exists:

[0146] ;

[0147] in, This represents the difference between the target heading angle and the road heading angle. Indicates the road direction angle;

[0148] Furthermore, since the heading angle in the data message is periodic, when the difference between the target heading angle and the road heading angle is greater than 180°, the angle difference is adjusted. The corresponding relationship is as follows:

[0149] ;

[0150] in, This represents the angular difference between the target direction and the road direction, within the range of 0° to 180° obtained after periodic conversion.

[0151] In the process of calculating the direction vector using the latitude and longitude of adjacent targets, and judging whether the target has an unreasonable trajectory based on the changes in the direction vector data, in order to obtain the sixteenth type meta-rule, the following relationship exists:

[0152] ;

[0153] in, Represents the direction vector. This represents the difference in longitude between two targets;

[0154] Furthermore, this rule is used to determine whether a target exhibits an unreasonable trajectory. It records the data changes of the direction vector between two targets over a period of time. If the data changes smoothly and gradually, it indicates normality; otherwise, it indicates that the target has an unreasonable trajectory.

[0155] In the process of obtaining lane directions from the coordinates of adjacent lanes in the preprocessed road test perception data sequence, and using the lane directions to determine whether there are anomalies in the target driving direction in order to obtain the seventeenth type of meta-rule, the following relationship exists:

[0156] ;

[0157] in, Indicates lane direction, Indicates the number of lane coordinate points. This means that the direction vector between two coordinate points is calculated using two coordinate points. Indicates the first Each lane coordinate point Indicates the first Each lane's coordinate points;

[0158] Furthermore, since lane data is composed of numerous coordinate points, this rule applies the average direction instead of the lane direction;

[0159] The lane direction derived from this rule is used to determine whether there is an anomaly between the current lane direction and the target driving direction;

[0160] The specific steps for obtaining the shortest distance between target coordinates by calculating the shortest distance using the latitude of adjacent targets and the Earth's radius, and then using this shortest distance to determine whether the target is moving in order to derive the eighteenth type of meta-rule, are as follows:

[0161] The first intermediate variable is obtained by calculating the latitude of adjacent targets, and the corresponding relationship in the process is as follows:

[0162] ;

[0163] in, Indicates the first intermediate variable. This represents the difference in longitude between two targets;

[0164] The second intermediate variable is obtained by using the first intermediate variable, and the corresponding relationship in the process is as follows:

[0165] ;

[0166] in, Indicates the second intermediate variable;

[0167] The shortest distance between the target coordinates is obtained using the second intermediate variable, and the corresponding relationship in the process is as follows:

[0168] ;

[0169] in, Indicates the shortest distance between target coordinates; This represents the Earth's radius, with a value of 63,710,000 m.

[0170] Furthermore, in this rule, a unified target with a velocity of 0 at adjacent time points is considered... The value of is calculated so that if the value is between 0 and 1, it can be determined that the target has not fluctuated; otherwise, it can be determined that the data of the stationary target has fluctuated significantly.

[0171] Step 3: Based on the detection result sequence, calculate the quality index using the K-means algorithm and obtain the quality level set;

[0172] In step 3, based on the detection result sequence, the K-means algorithm is used to calculate the quality index and obtain a set of quality levels. The specific steps are as follows:

[0173] The detection result sequence is divided into time windows to obtain partitions. The corresponding relationship in this process is as follows:

[0174] ;

[0175] in, Indicates partition, This indicates a slice operation. Represents the sequence of detection results. This indicates the size of each window;

[0176] By using the anomaly rate, missing rate, and latency rate obtained from the detection result sequence, a feature vector is constructed for each time period in the partition. The corresponding relationship in this process is as follows:

[0177] ;

[0178] in, This represents the feature vector for each time period. Indicates the anomaly rate. Indicates the missing rate. Indicates latency;

[0179] Using the number of outlier data points and the total number of data points, the outlier rate for each time period is calculated, and the outlier rate for each time period is used as the outlier score. The corresponding relationship in this process is as follows:

[0180] ;

[0181] in, Indicates abnormal scores. Indicates the number of outlier data points. Indicates the total number of data points;

[0182] The quality feature matrix is ​​constructed based on the feature vectors of each time period, and the corresponding relationship in the process is as follows:

[0183] ;

[0184] in, Represents the quality characteristic matrix, Each represents a feature vector for each time period;

[0185] Based on the quality feature matrix, four cluster centers are initialized using the K-means algorithm;

[0186] Clustering is performed based on the distance between data points and cluster centers to obtain clustering results. The corresponding relationship in this process is as follows:

[0187] ;

[0188] in, This means assigning each data point to the nearest cluster center. Represents data points, Indicates the first Cluster centers, This indicates selecting the smallest cluster between the data point and its assigned cluster center;

[0189] The cluster centers are updated using the clustering results to obtain the updated cluster centers. The corresponding relationship in this process is as follows:

[0190] ;

[0191] in, Indicates the updated cluster centers. Indicates assignment to the first The number of data points for each cluster center;

[0192] The midpoint between adjacent cluster centers is calculated using the updated cluster centers. The corresponding relationship in this process is as follows:

[0193] ;

[0194] in, Both represent the midpoint between adjacent cluster centers. Both represent the updated cluster centers;

[0195] The grade interval is defined based on the midpoint between the outlier score and the adjacent cluster centers. The corresponding relationship in the process is as follows:

[0196] ;

[0197] in, This indicates that the scores are categorized based on the anomalies. All are quality grades;

[0198] A set of quality levels is generated for each time period, and the corresponding relationship in the process is as follows:

[0199] ;

[0200] in, This represents the set of quality levels for each time period, such as the set of quality levels for the first hour. This indicates that the process has been handled by the allocation level function.

[0201] Step 4: Use the quality level set to trace the source of the anomaly detection sequence and obtain the anomaly tracing results.

[0202] For further details, please refer to Figure 2 , Figure 2 The anomaly detection rule grid diagram of this invention can serve as a "fault diagnosis map" to help technicians locate the corresponding meta-rules based on the anomaly type, and further trace the root cause of problems in upstream data input or calculation logic, thereby significantly improving the efficiency and accuracy of anomaly tracing.

[0203] The construction logic of the visualization carrier of the anomaly detection rule system closely revolves around the anomaly detection needs of roadside perception data. Specifically, it uses 18 types of meta-rules as the core basic unit. Each type of meta-rule is designed for specific anomaly scenarios in roadside perception data (such as target flashing, overlapping line crossing, trajectory anomalies, etc.) and clearly includes input parameters (such as target latitude and longitude, speed, and heading angle), mathematical calculation logic (such as predicting the location and calculating the intersection area), and anomaly judgment output. On this basis, the 18 types of meta-rules are divided into levels such as target existence detection, line crossing anomaly detection, heading angle and trajectory detection according to the anomaly detection dimension. Horizontal associations are established based on the data flow dependencies between meta-rules (for example, "the set of line crossing targets" is used as the input data for "instantaneous large-scale line crossing judgment"), ultimately forming a structured rule grid.

[0204] The anomaly detection rule grid diagram plays a crucial role in the technical implementation and application of this invention. On the one hand, the anomaly detection rule grid diagram clearly presents the full coverage of the 18 types of meta-rules for the main anomaly scenarios of roadside perception, as well as the logical dependencies and data flow between meta-rules, facilitating technicians to quickly understand the complete anomaly detection process and providing an intuitive reference for algorithm deployment, debugging, and daily maintenance. On the other hand, the modular structure of this grid facilitates subsequent functional expansion. If a new anomaly detection scenario (such as target speed change detection) needs to be added, only the meta-rule node needs to be added at the corresponding level and associated with the existing data input, without reconstructing the overall system. Furthermore, during the anomaly tracing process...

[0205] It should be noted that each type of meta-rule can make judgments on multiple results based on commonalities. Taking the first type of meta-rule as an example, the first type of meta-rule is used to determine whether the target is within the perception area. When judging whether the target is abnormally outside the perception area for a long time, the first type of meta-rule is first used to determine whether the target is within the perception area. In addition, for the discontinuous target trajectory, whether the target is flickering, target splitting and target duplication identification, which are further judgments for suddenly added or disappeared targets, the first type of meta-rule is also used for preliminary judgment.

[0206] Please see Figure 3 This invention also provides a real-time roadside sensing anomaly detection system based on a regular grid, the system comprising:

[0207] The preprocessing module is used for:

[0208] The given map data, monitoring area data, and road test perception data in the road test perception data are preprocessed to obtain the preprocessed road test perception data sequence;

[0209] The rules network module is used for:

[0210] Anomaly detection is performed on the preprocessed road test perception data sequence using a rule network consisting of 18 types of meta-rules, and the detection result sequence is obtained.

[0211] The quality indicator calculation module is used for:

[0212] Based on the detection result sequence, the K-means algorithm is used to calculate the quality index and obtain a set of quality levels;

[0213] The anomaly tracing module is used for:

[0214] Anomaly tracing results are obtained by using a set of quality levels to trace the origin of anomalies in the detected sequences.

[0215] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0216] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0217] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A real-time anomaly detection method based on a regular grid for roadside sensing, characterized in that, The method includes the following steps: Step 1: Preprocess the given map data, monitoring area data, and road test sensing data in the road test sensing data to obtain the preprocessed road test sensing data sequence; Step 2: Use a rule network composed of 18 types of meta-rules to perform anomaly detection on the preprocessed road test perception data sequence to obtain the detection result sequence. The specific steps are as follows: S201. Using the target position, target speed and merged monitoring area obtained from the preprocessed road test perception data sequence, determine whether the target is within the perception area, so as to simultaneously exclude the possibility of driving out of the detection area or entering from outside the detection area in the next moment, so as to obtain the first type of meta rule. S202. Obtain a set of candidate flashing target IDs from the set of target IDs in adjacent messages obtained from the preprocessed road test sensing data sequence. Use the set of candidate flashing target IDs to determine whether the target is flashing in order to obtain the second type of meta-rule. S203. Based on the target position, target speed and target heading angle obtained from the preprocessed road test perception data sequence, the latitude and longitude coordinates of the predicted position are obtained. The latitude and longitude coordinates of the predicted position are used to determine whether the target before and after the flashing is the same target, so as to obtain the third type of meta rule. S204. Using the latitude and longitude coordinates of the predicted location, the location coordinates of the suddenly appearing target obtained in the preprocessed road test perception data sequence are judged to determine whether the ID of the target has changed before and after the flashing, so as to obtain the fourth type of meta-rule. S205. Using the target size parameters and target heading angle in the preprocessed road test perception data sequence, the actual latitude and longitude coordinates of the four corner points of the target are obtained. The actual latitude and longitude coordinates of the four corner points of the target are used to determine whether there is overlap or line-crossing anomaly, so as to obtain the fifth type of meta-rule. S206. Based on the map parameters and target latitude and longitude coordinates in the preprocessed road test perception data sequence, obtain the set of targets that overlap the lane lines. Use the set of targets that overlap the lane lines to determine whether there is an overlap phenomenon in the target domain, so as to obtain the sixth type of meta-rule. S207. Using the target set for crossing the line, the target set for instantaneous large-scale crossing is obtained by using the total number of targets appearing in the message and the preset threshold for instantaneous large-scale crossing judgment. The target set for instantaneous large-scale crossing is used to determine whether instantaneous large-scale crossing has occurred, so as to obtain the seventh type of meta rule. S208. Based on the target line-over set and map parameters, obtain the grid of large-scale line overtaking in the area. Use the grid of large-scale line overtaking in the area to determine whether large-scale line overtaking occurs in the area within a time period, so as to obtain the eighth type of meta-rule. S209. Using the target types in adjacent messages, obtain the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after flashing. Using the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after flashing, determine whether the target is repeatedly identified as a different type, so as to obtain the ninth meta-rule and the tenth meta-rule. S210. Use the speeds of two adjacent targets, the heading angles of two adjacent targets, and the dimensions of two adjacent targets to obtain the target set where target splitting occurs. Use the target set where target splitting occurs to determine whether target splitting has occurred, so as to obtain the eleventh type meta-rule. S211. Obtain the coordinates of the four corner points of the approximate rectangles of adjacent targets from the preprocessed road test perception data sequence to obtain the intersection area. Use the intersection area to determine whether the adjacent targets overlap to obtain the twelfth type meta-rule. S212. Determine whether the target is turning based on the target heading angle and the distance the target has moved, in order to obtain the thirteenth type of meta-rule; S213. Using the target heading angle and the preset heading angle anomaly judgment threshold, determine whether the target generates heading angle data anomalies when stationary, so as to obtain the fourteenth type meta-rule. S214. Based on the road direction angle and target heading angle obtained from the preprocessed road test perception data sequence, determine whether the target has a heading angle deviation value in order to obtain the fifteenth type meta rule; S215. Calculate the direction vector using the latitude and longitude of adjacent targets, and determine whether there is an unreasonable trajectory of the target based on the changes in the direction vector data, so as to obtain the sixteenth type of meta-rule; S216. Based on the coordinate points of adjacent lanes obtained from the preprocessed road test perception data sequence, the lane direction is obtained. The lane direction is used to determine whether there is an anomaly in the target driving direction, so as to obtain the seventeenth meta-rule. S217. Calculate the shortest distance between target coordinates using the latitude of adjacent targets and the Earth's radius. Determine whether the target is jumping by using the shortest distance between target coordinates to obtain the eighteenth meta-rule. S218. Using the first to eighteenth categories of meta-rules obtained from S201 to S217, the data in the preprocessed road test perception data sequence are judged and labeled to obtain the detection result sequence; Step 3: Based on the detection result sequence, use the K-means algorithm to calculate the quality index and obtain the quality level set; Step 4: Use the quality level set to trace the source of the anomaly detection sequence and obtain the anomaly tracing results.

2. The real-time anomaly detection method based on regular grids for roadside sensing according to claim 1, characterized in that, In the process of determining whether a target is within the sensing area by using the target position, target speed, and merged monitoring area obtained from the preprocessed road test sensing data sequence, in order to simultaneously exclude the possibility of the target leaving the detection area or entering from outside the detection area in the next moment, and in order to obtain the first type of meta-rule, the following relationship exists: ； in, This refers to targets within the perception area. express Time of the first The location of the target Indicates the merged monitoring area. express arrive The shortest distance to the boundary, express Time of the first The speed of the target This represents the time difference between two consecutive messages. This represents a predefined distance threshold; Indicates that the target is in At the same time, arrive The shortest distance to the boundary is less than The sum of the theoretical distance moved; In the process of obtaining a set of candidate blinking target IDs from the set of target IDs within adjacent messages obtained from the preprocessed drive test sensing data sequence, and using the set of candidate blinking target IDs to determine whether a target is blinking in order to obtain the second type of meta-rule, the following relationship exists: ； in, express The set of target IDs within the time message. express The set of target IDs within the time message; This represents the set of target IDs where the previous message did not exist and the next message suddenly appeared. This represents the set of target IDs where the previous message existed and the next message suddenly disappeared.

3. The real-time anomaly detection method based on regular grids for roadside sensing as described in claim 2, characterized in that, Based on the target position, target velocity, and target heading angle obtained from the preprocessed road test sensing data sequence, the latitude and longitude coordinates of the predicted position are obtained. The latitude and longitude coordinates of the predicted position are then used to determine whether the target before and after the flash is the same target, thus obtaining the third-type meta-rule. The specific steps for obtaining the latitude and longitude coordinates of the predicted position are as follows: Based on the target speed and the time difference between the two consecutive messages, the distance increment is calculated, and the corresponding relationship is as follows: ； in, Indicates the distance increment; Using the distance increment, target heading angle, and Earth's average radius, the latitude increment and longitude increment are calculated respectively. The corresponding relationship is as follows: ； in, Indicates the latitude increment. Indicates the increment of longitude. express Time of the first The heading angle of the target, This represents the average Earth radius. Indicates the latitude of the current location; Using the latitude increment, longitude increment, current latitude, and current longitude, the predicted latitude and longitude coordinates are calculated. The corresponding relationship in this process is as follows: ； in, Indicates the latitude of the predicted location. Indicates the longitude of the predicted location. The latitude and longitude coordinates of the predicted location. Indicates the longitude of the current location; In the process of using the latitude and longitude coordinates of the predicted location to determine the location coordinates of suddenly appearing targets obtained from the preprocessed road test perception data sequence, in order to determine whether the target ID has changed before and after the flashing, and in order to obtain the fourth type of meta-rule, the following relationship exists: ； in, This represents the set of targets that can be matched by a given location. This indicates the location coordinates of the suddenly appearing target. This indicates a predefined distance matching threshold.

4. The real-time anomaly detection method based on regular grids for roadside sensing according to claim 3, characterized in that, The specific steps for obtaining the actual latitude and longitude coordinates of the four corner points of the target by using the target size parameters and target heading angle in the preprocessed road test sensing data sequence, and using the actual latitude and longitude coordinates of the four corner points to determine whether there is overlap or line-crossing anomalies in order to obtain the fifth type of meta-rule are as follows: The local coordinates of the four corner points relative to the center point of the target are obtained from the target's size parameters. The corresponding relationship in this process is as follows: ； in, This represents the local coordinates of the four corner points relative to the center point of the target. Indicates the length of the target. Indicates the width of the target; The local coordinates of the four corner points relative to the target center point are rotated based on the target heading angle to obtain the rotated coordinates. The corresponding relationship in this process is as follows: ； in, Represents the coordinates after rotation; The actual latitude and longitude coordinates of the four corner points of the target are obtained by using the rotated coordinates and the average radius of the Earth. The corresponding relationship in this process is as follows: ； in, This represents the actual latitude and longitude coordinates of the four corner points of the target; This indicates the east-west offset of the corner point relative to the target center point after rotation. This indicates the north-south offset of the corner point relative to the target center point after rotation. Indicates the longitude conversion factor. Indicates the latitude conversion factor; The specific steps for obtaining the target line-crossing set are as follows: Based on the map parameters and target latitude and longitude coordinates in the preprocessed road test perception data sequence, a set of line-crossing targets is obtained. Then, this set is used to determine whether lane lines in the target domain overlap, in order to derive the sixth type of meta-rule. Using the target's latitude and longitude coordinates and the unit size of the grid, the grid index where the target is located is calculated. The corresponding relationship in this process is as follows: ； in, Indicates the grid index where the target is located. Indicates the unit size for dividing the grid; Use the grid index of the target to find the set of lane lines within the target grid in the map data; Intersection detection is performed between the set of lane lines within the target grid and the target to obtain the set of targets crossing the lane lines. The corresponding relationship in this process is as follows: ； in, Represents the set of targets that press the line. This indicates the calculation of the intersection length. express The first moment One goal, Represents the set of lane lines within the target grid. This represents the intersection length threshold.

5. The real-time anomaly detection method based on regular grids for roadside sensing according to claim 4, characterized in that, In the process of using the target set for instantaneous large-scale overpasses, obtaining the target set for instantaneous large-scale overpasses by combining the total number of targets appearing in the message with a preset threshold for instantaneous large-scale overpass judgment, and using the target set for instantaneous large-scale overpasses to determine whether an instantaneous large-scale overpass situation has occurred, in order to obtain the seventh type of meta-rule, the following relationship exists: ； in, This represents the set of targets where a large number of items simultaneously press against the line. express The total number of targets at any given time This indicates the preset threshold for judging a large number of lines being pressed down in an instant; In the process of obtaining a grid of large-scale line crossings in the area based on the target line crossing set and map parameters, and using this grid to determine whether large-scale line crossings occurred within a specific time period to obtain the eighth type of meta-rule, the following relationship exists: ； in, This indicates a grid representing a large number of lines being pressed into the area. Indicates the start time. Indicates the monitoring duration. Representing different grids, Represents the entire map grid. This indicates the preset threshold for the number of times the wires are pressed within the grid. In the process of obtaining the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after blinking by utilizing the target types in adjacent messages, and then determining whether a target is repeatedly identified as a different type to obtain the ninth and tenth meta-rules, the specific steps for obtaining the sets of targets repeatedly identified as different types and the sets of targets repeatedly identified as different types after blinking are as follows: The relationship between obtaining the target types in the message and the types of all targets at the same time is as follows: ； in, Indicates time All target types, Indicates the first The type of target, Indicates the first One goal, Indicates time The message; By utilizing the types of all targets at the same time, we obtain a set of targets repeatedly identified into different types. The corresponding relationship in this process is as follows: ； in, This indicates that the target is repeatedly identified as a set of different types. Indicates the first The goal is at a certain moment The type of disappearance Indicates the first The goal is at a certain moment The type of disappearance; Using the types of all targets at the same time, the set of targets that are repeatedly identified as different types after flashing is obtained. The corresponding relationship in the process is as follows: ； in, This indicates that the target is repeatedly identified as a set of different types after it blinks. Indicates the first Target time interval The type when it reappears.

6. The real-time anomaly detection method based on regular grids for roadside sensing according to claim 5, characterized in that, In the process of obtaining the target set where target splitting occurs using the velocities, heading angles, and dimensions of two adjacent targets, and then using this target set to determine whether target splitting has occurred in order to derive the eleventh type of meta-rule, the following relationship exists: ； in, This represents the set of targets where target splitting occurs. express Time of the first One goal, express Time of the first The speed of the target This represents the threshold value representing the velocity difference between two targets. Indicates the first The size of the target Indicates the first The size of the target This represents the threshold representing the size difference between two targets. express Time of the first The heading angle of the target, This represents the threshold for the difference in heading angle between two targets; The process of obtaining the intersection area by approximating the four corner coordinates of adjacent target rectangles from the preprocessed road test perception data sequence, and using the intersection area to determine whether adjacent targets overlap, in order to obtain the twelfth type meta-rule, also includes the following sub-steps: The proximity of two targets is determined by using their latitude and longitude coordinates and a preset distance threshold. The relationship in this process is as follows: ； in, Indicates the first The longitude of the target Indicates the first The dimensions of each target This represents a preset distance threshold between targets; When two targets are determined to be close in location, the intersection region is obtained by approximating the coordinates of the four corner points using the rectangles of adjacent targets. The corresponding relationship in this process is as follows: ； in, Indicates the intersection region. Indicates the first The rectangular approximation of the four corner points of the target. Indicates the first The rectangular approximation of the four corner points of the target; By using the intersection region and a preset area threshold, it is determined whether adjacent targets overlap. The corresponding relationship in this process is as follows: ； in, This represents the set of targets where overlap occurs. This indicates whether the overlapping area of ​​the targets is greater than a preset area threshold. This indicates a preset area threshold. In determining whether a target is turning based on its heading angle and the distance it has traveled, in order to derive the thirteenth meta-rule, the relevant relationship is as follows: ； in, This describes the rules for determining whether a target should turn. This represents the change in the target's heading angle between two time points. It represents the distance the target has moved between two points in time. Indicates the threshold for change in heading angle. This indicates the location distance threshold.

7. The real-time anomaly detection method based on regular grids for roadside sensing according to claim 6, characterized in that, In the process of determining whether a target generates abnormal heading angle data when stationary by using the target heading angle and a preset heading angle anomaly detection threshold, in order to obtain the fourteenth type of meta-rule, the following relationship exists: ； in, express Time of the first The heading angle of the target, This indicates the preset threshold for determining abnormal heading angles; In the process of determining whether the target has a heading angle deviation value based on the road direction angle and target heading angle obtained from the preprocessed road test perception data sequence, in order to obtain the fifteenth type meta-rule, the following relationship exists: ； in, This represents the difference between the target heading angle and the road heading angle. Indicates the road direction angle; In the process of calculating the direction vector using the latitude and longitude of adjacent targets, and judging whether the target has an unreasonable trajectory based on the changes in the direction vector data, in order to obtain the sixteenth type meta-rule, the following relationship exists: ； in, Represents the direction vector. This represents the difference in longitude between two targets; In the process of obtaining lane directions from the coordinates of adjacent lanes in the preprocessed road test perception data sequence, and using the lane directions to determine whether there are anomalies in the target driving direction in order to obtain the seventeenth type of meta-rule, the following relationship exists: ； in, Indicates lane direction, Indicates the number of lane coordinate points. This means that the direction vector between two coordinate points is calculated using two coordinate points. Indicates the first Each lane coordinate point Indicates the first Each lane's coordinate points; The specific steps for obtaining the shortest distance between target coordinates by calculating the shortest distance using the latitude of adjacent targets and the Earth's radius, and then using this shortest distance to determine whether the target is moving in order to derive the eighteenth type of meta-rule, are as follows: The first intermediate variable is obtained by calculating the latitude of adjacent targets, and the corresponding relationship in the process is as follows: ； in, Indicates the first intermediate variable. This represents the difference in longitude between two targets; The second intermediate variable is obtained by using the first intermediate variable, and the corresponding relationship in the process is as follows: ； in, Indicates the second intermediate variable; The shortest distance between the target coordinates is obtained using the second intermediate variable, and the corresponding relationship in the process is as follows: ； in, This represents the shortest distance between target coordinates. This represents the Earth's radius.

8. The real-time anomaly detection method based on regular grids for roadside sensing according to claim 7, characterized in that, In step 3, the quality index is calculated using the K-means algorithm based on the detection result sequence, and a set of quality levels is obtained. The specific steps are as follows: The detection result sequence is divided into time windows to obtain partitions. The corresponding relationship in this process is as follows: ； in, Indicates partition, This indicates a slice operation. Represents the sequence of detection results. This indicates the size of each window; By using the anomaly rate, missing rate, and latency rate obtained from the detection result sequence, a feature vector is constructed for each time period in the partition. The corresponding relationship in this process is as follows: ； in, This represents the feature vector for each time period. Indicates the anomaly rate. Indicates the missing rate. Indicates latency; Using the number of outlier data points and the total number of data points, the outlier rate for each time period is calculated, and the outlier rate for each time period is used as the outlier score. The corresponding relationship in this process is as follows: ； in, Indicates abnormal scores. Indicates the number of outlier data points. Indicates the total number of data points; The quality feature matrix is ​​constructed based on the feature vectors of each time period, and the corresponding relationship in the process is as follows: ； in, Represents the quality characteristic matrix, Each represents a feature vector for each time period; Based on the quality feature matrix, four cluster centers are initialized using the K-means algorithm; Clustering is performed based on the distance between data points and cluster centers to obtain clustering results. The corresponding relationship in this process is as follows: ； in, This means assigning each data point to the nearest cluster center. Represents data points, Indicates the first Cluster centers, This indicates selecting the smallest cluster between the data points and their assigned cluster centers; The cluster centers are updated using the clustering results to obtain the updated cluster centers. The corresponding relationship in this process is as follows: ； in, Indicates the updated cluster centers. Indicates assignment to the first The number of data points for each cluster center; The midpoint between adjacent cluster centers is calculated using the updated cluster centers. The corresponding relationship in this process is as follows: ； in, Both represent the midpoint between adjacent cluster centers. Both represent the updated cluster centers; The grade interval is defined based on the midpoint between the outlier score and the adjacent cluster centers. The corresponding relationship in the process is as follows: ； in, This indicates that the scores are categorized based on the anomalies. All are quality grades; A set of quality levels is generated for each time period, and the corresponding relationship in the process is as follows: ； in, This represents the set of quality levels for each time period. This indicates that the process has been handled by the allocation level function.

9. A real-time anomaly detection system based on a regular grid for roadside sensing, characterized in that, The system employs the roadside sensing anomaly real-time detection method based on a regular grid as described in any one of claims 1 to 8, and the system comprises: The preprocessing module is used for: The given map data, monitoring area data, and road test perception data in the road test perception data are preprocessed to obtain the preprocessed road test perception data sequence; The rules network module is used for: Anomaly detection is performed on the preprocessed road test perception data sequence using a rule network consisting of 18 types of meta-rules to obtain the detection result sequence; The quality indicator calculation module is used for: Based on the detection result sequence, the K-means algorithm is used to calculate the quality index and obtain a set of quality levels. The anomaly tracing module is used for: Anomaly tracing results are obtained by using a set of quality levels to trace the origin of anomalies in the detected sequences.

Images