A method, medium and device for detecting illegally operated vehicles

Abstract

The application relates to the technical field of intelligent traffic control, and provides a detection method, a medium and equipment for illegal operation vehicles, the method comprising the following steps: acquiring bidirectional driving time sequence data of a target vehicle on a target road section and extracting vehicle behavior features representing vehicle operation tendencies from the bidirectional driving time sequence data; performing suspected operation preliminary screening according to the vehicle behavior features to obtain a preliminary screening result; when the preliminary screening result indicates that the target vehicle meets a preliminary operation behavior, performing vehicle type identification according to the bidirectional driving time sequence data to obtain a vehicle category of the target vehicle; when the vehicle category belongs to a preset vehicle type, performing a copilot feature analysis according to the bidirectional driving time sequence data to obtain copilot features; and determining whether the target vehicle belongs to a suspected illegal operation vehicle based on the vehicle behavior features, the vehicle category and the copilot features. The application improves detection accuracy and efficiency, reduces detection errors, and improves the intelligentization and precision level of vehicle illegal operation detection.

Description

Technical Field

[0001] This application relates to the field of intelligent traffic control technology, and in particular to a method, medium, and device for detecting illegally operating vehicles. Background Technology

[0002] In recent years, with the popularization of mobile internet travel, illegal carpooling, chartering and disguised online ride-hailing operations have continued to exist in urban clusters, urban-rural fringe areas and highway networks. These behaviors are covert, difficult to detect and mobile and scattered.

[0003] Current technologies, such as roadside checks at toll stations and service areas, setting up fixed or temporary enforcement checkpoints, relying on public reports and complaints for manual surveillance, and combining this with post-incident investigations of abnormal behavior at traffic police checkpoints (overloading, illegal parking for picking up or dropping off passengers), suffer from several drawbacks. These include low detection efficiency, limited coverage, inability to effectively cover all road sections and time periods, disruption of normal traffic flow due to prolonged checkpoint operations, and high manpower and time costs. Therefore, a new method for detecting illegally operating vehicles is urgently needed to address these technical issues. Summary of the Invention The purpose of this application is to provide a method, medium, and device for detecting illegally operating vehicles, in order to improve the technical problems raised in the background section.

[0004] For the purposes mentioned above, this application provides the following technical solution: The first aspect of this application provides a method for detecting illegally operating vehicles, including: acquiring bidirectional driving time sequence data of the target vehicle on a target road segment; Vehicle behavior features that characterize the vehicle's operational tendency are extracted from the driving time series data. Suspected operations are initially screened based on the vehicle behavior features to obtain the initial screening results. When the initial screening results indicate that the target vehicle meets the initial operational behavior, vehicle type identification is performed based on the two-way driving time sequence data to obtain the vehicle category of the target vehicle; When the target vehicle belongs to a preset vehicle type, the characteristics of the front passenger are obtained by analyzing the characteristics of the front passenger based on the bidirectional driving time sequence data. Based on the vehicle's behavioral characteristics, vehicle category, and the characteristics of the front passenger, it is determined whether the target vehicle is a suspected illegally operating vehicle.

[0005] Preferably, the vehicle behavior characteristics include one or more of the following: number of trips, trip frequency, trip time period, two-way round-trip pairing rate, and number of two-way round-trip pairings; The preliminary screening of suspected operations based on the vehicle's behavioral characteristics, to obtain the preliminary screening results, includes: Based on the vehicle behavior characteristics, the initial screening score of the target vehicle is calculated according to the preset initial screening calculation model. The initial screening result is obtained based on the initial screening score. When the initial screening score exceeds the preset first score threshold, it is determined that the initial screening result indicates that the target vehicle meets the preliminary operational behavior.

[0006] Preferably, the characteristics of the front passenger include identity features, and the step of analyzing the front passenger characteristics based on the bidirectional driving time series data to obtain the front passenger characteristics includes: Face recognition is performed on the passenger side area of ​​each frame of the vehicle passing image in the bidirectional driving time sequence data to obtain the face features at each vehicle passing time; Based on the facial feature analysis, the facial similarity between different vehicle images is obtained, and based on the facial similarity, identity features that reflect the changes in the identity of the front passenger are obtained.

[0007] Preferably, the characteristics of the front passenger also include quantitative characteristics. The step of analyzing the front passenger characteristics based on the bidirectional driving time series data to obtain the front passenger characteristics includes: Face recognition is performed on the passenger side area of ​​each frame of the vehicle passing image in the bidirectional driving time sequence data to obtain the number of passengers in the passenger side each time the vehicle passes. Determine whether the target vehicle has an overload of passengers in the front passenger seat based on the number of passengers in the front passenger seat, and count the number of times the vehicle has an overload of passengers; By comparing whether the number of front-seat passengers is the same in adjacent round-trip vehicle images, quantitative features reflecting changes in the number of front-seat passengers are obtained based on the comparison results and the number of times the number of passengers is overloaded.

[0008] Preferably, the front passenger characteristics also include interaction characteristics between the front passenger and the driver, and the front passenger characteristics are obtained by performing front passenger characteristic analysis based on the two-way driving time series data, including: For each frame of the vehicle passing image in the bidirectional driving time sequence data, one or more interactive actions of the driver and passenger are identified, including mouth shape, body movements, interaction distance, and head angle. The interactive features are obtained based on the identified interactive actions.

[0009] Preferably, determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle's behavioral characteristics, vehicle category, and the characteristics of the front passenger includes: Determine the initial screening score that matches the vehicle's behavioral characteristics; Determine the category score that matches the vehicle category; Determine the co-pilot score that matches the characteristics of the co-pilot; The illegal operation risk score of the target vehicle is calculated based on the initial screening score, the category score, and the co-driver's score. The illegal operation risk score is then used to determine whether the target vehicle is a suspected illegal operation vehicle.

[0010] Preferably, before determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle behavior characteristics, vehicle category, and front passenger characteristics, the method further includes: Analyze the passenger's riding situation in vehicles other than the target vehicle to obtain the associated occupant characteristics of the target vehicle; The determination of whether the target vehicle is a suspected illegal operating vehicle based on the vehicle's behavioral characteristics, vehicle category, and passenger characteristics includes: Based on the vehicle's behavioral characteristics, vehicle category, front passenger characteristics, and associated passenger characteristics, it is determined whether the target vehicle is a suspected illegally operating vehicle.

[0011] Preferably, the passenger in the front seat is analyzed in vehicles other than the target vehicle to obtain the associated occupant characteristics of the target vehicle, including: Obtain the identifiers of co-drivers who passed through the target road segment within the statistical time window from the co-driver feature database; Obtain the number of vehicle identifiers associated with the personnel identifier, and determine the associated occupant characteristics based on the number of vehicle identifiers.

[0012] Preferably, determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle behavior characteristics, vehicle category, front passenger characteristics, and associated passenger characteristics includes: Obtain the associated occupant score that matches the number of vehicle identifiers; Determine the initial screening score that matches the vehicle's behavioral characteristics; Determine the category score that matches the vehicle category; Determine the co-pilot score that matches the characteristics of the co-pilot; The illegal operation risk score of the target vehicle is calculated based on the initial screening score, category score, associated passenger score, and co-driver score. The illegal operation risk score is used to determine whether the target vehicle is suspected of being an illegal operation vehicle.

[0013] Preferably, before acquiring the bidirectional driving time sequence data of the target vehicle on the target road segment, the method further includes: Acquire vehicle images captured by monitoring points under the target road segment to form a two-way vehicle image set; Identify the vehicle identifiers in the passing images in the bidirectional vehicle passing image set, and construct the bidirectional driving time sequence data based on the vehicle identifiers and the shooting time.

[0014] In a second aspect, this application provides a computer-readable storage medium storing executable instructions that, when executed by a processor, cause the processor to perform the method described in any embodiment of this application.

[0015] A third aspect of this application provides an electronic device, comprising: one or more processors; A memory for storing one or more programs that, when executed by one or more processors, cause the one or more processors to perform the method as described in any embodiment of this application.

[0016] This application provides a method, medium, and equipment for detecting illegally operating vehicles, centered on a single two-way checkpoint. It constructs two-way driving time-series data of vehicles from captured images, extracts vehicle behavior characteristics representing operational tendencies from this time-series data, and filters out a first set of suspected vehicles that conform to operational time patterns. Based on this, vehicle categories are identified, and key vehicles meeting these criteria are used to generate a second set of suspected vehicles and vehicle category scores. Finally, the characteristics of the passenger in the front seat are combined to determine whether the vehicle is suspected of being illegally operating. This improves detection accuracy and efficiency, reduces detection errors, and enhances the intelligence and precision of illegal vehicle operation detection. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the method for detecting illegally operating vehicles provided in this application embodiment; Figure 2 A flowchart for calculating the characteristics of the co-driver / passenger provided in this application embodiment; Figure 3 Another flowchart for calculating the characteristics of the co-driver provided in the embodiments of this application; Figure 4 A flowchart illustrating the method for determining illegally operating vehicles provided in this application embodiment; Figure 5 Another flowchart of the method for determining illegally operating vehicles provided in this application embodiment. Detailed Implementation

[0019] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0021] For example, the terms "first" and "second" used in this application are only used to distinguish similar objects and differentiate the first object from another object, rather than to describe a specific order or sequence, nor should they be interpreted as indicating or implying relative importance.

[0022] This application proposes a method for detecting illegally operating vehicles, combining... Figure 1 As shown, the method includes: Step S101: Obtain the bidirectional driving time sequence data of the target vehicle on the target road segment.

[0023] In this embodiment, the target road segment refers to a monitoring section located along the highway, defined around a single checkpoint monitoring point. This segment emphasizes the bidirectional coverage capability of the checkpoint cross-section, by selecting specific checkpoint devices with high-definition capture capabilities as monitoring points within the highway network. This checkpoint has bidirectional lane coverage capability, or includes paired bidirectional capture units, such as two sets of back-to-back capture units installed on a single pole or gantry, facing the oncoming and outgoing lanes respectively, or two independent capture devices symmetrically deployed on both sides of the road centerline, together forming a complete bidirectional checkpoint. The logical layout of this checkpoint is pre-configured, clearly defining and marking the lane attributes for outgoing directions (e.g., leaving the city) and incoming directions (e.g., entering the city), and establishing a mapping relationship between directions and device channel numbers.

[0024] In one embodiment, by acquiring vehicle images captured by monitoring points under the target road segment, the acquired vehicle images captured in the outgoing and incoming directions are integrated to form a two-way vehicle image set; the vehicle identifiers in the vehicle images in the two-way vehicle image set are identified, and the two-way driving time sequence data is constructed based on the vehicle identifiers and the shooting time.

[0025] In this embodiment, the vehicle identifier can be the license plate number. First, standardize the vehicle identifier for each frame of passing vehicle image in the set. Specifically, remove the leading and trailing blank characters, forcibly convert full-width characters (such as Chinese brackets and numbers) to half-width characters, uniformly convert letters to uppercase (for example, convert "a" to "A" and "Jing b" to "JING B"), eliminate encoding differences, represent them in a unified character form, and delete common delimiters in the license plate, including but not limited to the midpoint "·", dash "-", space, etc. For example, standardize "Yue A·12345" to "YUE A12345", set filtering rules, and eliminate obviously invalid license plate strings such as unlicensed (records with an empty recognition result or a confidence level lower than the preset threshold), garbled codes (records containing unrecognizable special characters or non-standard combinations of Chinese characters, letters, and numbers), and license plate strings with a severely inconsistent length (such as less than 5 digits or more than 8 digits) to obtain the vehicle index plate_norm.

[0026] At the same time, uniformly convert the capture time to the preset standard time zone. Preferably, use Beijing time (UTC+8). Regardless of whether the original time of the front-end device is local time or other time zones, convert it to the standard time, and generate multi-dimensional time features based on the standardized timestamp. The multi-dimensional time features can include one or more of the passing date, hour number, day of the week, etc. Among them, pass_date: natural date (format: YYYY-MM-DD), used to count the passing frequency by day, pass_hour: hour number (0-23), used to analyze peak hours, pass_dow: day of the week (1-7, corresponding to Monday to Sunday), used to identify the operating rules on weekdays and weekends.

[0027] Next, process adjacent records of the same vehicle identifier in the same direction. If the time difference between two records is less than the second preset threshold dedup_sec, only retain the earlier record. Preferably, the second preset threshold dedup_sec can be 2 seconds, 5 seconds, and this application does not make specific limitations on this.

[0028] Finally, after the above-mentioned standardization process of the vehicle identifier, time zone unification, and deduplication, construct two-way driving time series data with the vehicle index plate_norm. Aggregate all two-way passing vehicle records belonging to the same vehicle index plate_norm within the statistical time window w (for example, w = 7 days or w = 30 days) into an ordered set, which is defined as the vehicle sequence , where represents the target vehicle The total number of passes within the statistical time window w, represents the th passing record. The vehicle sequence After sorting the capture times in ascending order, we construct bidirectional time series, daily count series, hourly count series, adjacent passage time interval series, and bidirectional round-trip pairing series. We set bidirectional round-trip pairing rules based on time window constraints. For example, from... Filter all records where the direction of travel is "out". Sort by time in ascending order, and in The records that meet the criteria of entering from all directions within a preset time window are selected. ,for The departure time of any departure record in the data is recorded as follows: ,for The time when any entry record in the record is reached is recorded as :

[0029] in, Indicates the first The second departure and the first The time interval between each entry. A round trip must meet the following requirements. ,in, This represents the threshold for the shortest return trip time. This represents the threshold for the longest return trip.

[0030] Preferably, a greedy nearest neighbor matching algorithm is used to pair the two sets, meaning that each outgoing record is matched with at most one incoming record, and each incoming record is matched at most once. A successfully paired record pair is counted as one round trip, and the number of round trip pairs is accumulated. This is used to calculate the round-trip pairing rate. :

[0031] in, This indicates the total number of times the vehicle passed through within the statistical time window w. It is a very small smoothing constant used to prevent when The occasional division-by-zero error has improved the accuracy of identifying illegal carpooling vehicles.

[0032] Step S102: Extract vehicle behavior features that characterize the vehicle's operational tendency from the driving time series data, and perform preliminary screening of suspected operations based on the vehicle behavior features to obtain preliminary screening results.

[0033] In this embodiment, the vehicle behavior characteristics include one or more of the following: number of trips, trip frequency, trip time period, two-way round-trip pairing rate, and number of two-way round-trip pairings. Trip frequency refers to one or more of the following: number of active days, average daily pass rate, and weekday / weekend intensity ratio.

[0034] Travel time period includes hourly coverage Normalized hourly entropy Nighttime percentage Peak concentration For each 24-hour period of a day (with values ​​from 0 to 23), count the number of vehicle passages within that hour, and record it as... (in (Representing the hour number), calculated as the proportion of hours with passage records within the statistical time window out of the total number of hours, i.e., hour coverage. , where the molecule represents satisfying hours The number of times (i.e. how many different hours the vehicle traveled) is calculated, with the denominator being 24 hours in a day. If the value is close to 1 (e.g., above 0.8), it indicates that the vehicle is active all day long. If the value is low (e.g., concentrated only in 2-3 hours), it is more in line with the characteristics of a commuter vehicle.

[0035] To quantify the randomness of travel time, information entropy calculation is introduced. To eliminate the influence of different statistical period lengths, normalized hourly entropy is calculated. ,in, The probability of passage per hour ,in, This represents the total number of times the vehicle passes through within the statistical time window w. Preset nighttime time period set. The study counted the abnormal activity of target vehicles during nighttime hours, with nighttime accounting for a certain percentage. The numerator represents the travel time falling during the nighttime period. The number of records, This represents the total number of trips made by the vehicle within the statistical time window w. If this ratio is significantly higher than the average for ordinary commuter vehicles, it indicates that the vehicle exhibits characteristics of nighttime operation. Peak concentration is calculated by identifying the hour with the highest trip frequency within the statistical time window and calculating the concentration for that hour. ,in, This represents the maximum number of passages in all hours. This indicates the total number of times the vehicle passed through within the statistical time window w. It is a very small smoothing constant. When the peak concentration is extremely high (e.g., greater than 0.5), it indicates that more than half of the traffic is concentrated in a particular hour.

[0036] The aforementioned vehicle behavior characteristics are used for preliminary screening of suspected operational situations. The preliminary screening rules mainly include sample validity and rule sets. Sample validity includes one or more of the following conditions: 1) there are valid vehicle passage records in at least two directions; 2) there is at least one computable round-trip candidate pair; 3) the total number of passages within the statistical time window w. 4) The number of active days is not less than the number of days threshold. For example, the number of times threshold can be 20, and the number of days threshold can be 3. The rule group has multiple sets of rules, such as: 1) High-frequency bidirectional round-trip rule M1, used to determine whether the vehicle exhibits stable and high-frequency round-trip operation behavior within the statistical time window, that is, "driving out - driving in" appears in pairs and the frequency is higher than that of ordinary private cars; 2) Atypical commuter distribution rule M2, used to exclude commuter vehicles for fixed time periods and capture the typical concentrated travel time characteristics of operating vehicles (concentrated morning and evening peak hours, uneven hourly distribution, and limited coverage hours); 3) Interval and rhythm rule M3, used to determine the frequency of adjacent travel time intervals. Based on the weekday distribution dimension, determine whether the vehicles meet the regular characteristics of operating vehicles.

[0037] In one embodiment, the initial screening score of the target vehicle is calculated based on the vehicle behavior characteristics according to a preset initial screening calculation model, and the initial screening result is obtained based on the initial screening score. When the initial screening score exceeds a preset first score threshold, it is determined that the initial screening result indicates that the target vehicle meets the preliminary operational behavior.

[0038] In this embodiment, one or more of the following features within the statistical time window w—total number of passes, round-trip pairing rate, hourly coverage, normalized hourly entropy, nighttime proportion, peak concentration, and adjacent pass time interval—are numerically quantified. The initial screening calculation model can be a weighted sum of the quantified values ​​for the corresponding dimensions, and the resulting value is used as the initial screening score. For example, the six features within the statistical time window w—total number of passes, round-trip pairing rate, normalized hourly entropy, nighttime proportion, peak concentration, and adjacent pass time interval—are quantified and weighted summed to obtain the initial screening score for the target vehicle. :

[0039] in, Represents the normalization function. Indicates configurable weights, satisfying .

[0040] Initial screening score When the score is greater than or equal to a preset first score threshold, or when the target vehicle's behavioral characteristics satisfy at least two sets of rules in rule groups M1 to M3, the initial screening result indicates that the target vehicle meets the preliminary operational behavior, and the target vehicle that meets the preliminary operational behavior is added to the first suspected vehicle set V. s_1 .

[0041] Step S103: When the initial screening result indicates that the target vehicle meets the initial operational behavior, vehicle type identification is performed based on the bidirectional driving time sequence data to obtain the vehicle category of the target vehicle.

[0042] In this embodiment, vehicle types can include various types such as Car, SUV, MPV, Van, Bus, and Truck. After the target vehicle passes the initial screening, one or more frames from its bidirectional driving time-series data are selected to contain a frontal view of the vehicle and a front-view perspective view. This ensures that the vehicle type recognition model can capture the vehicle bounding box location result or the license plate location result. Since the imaging quality of the same vehicle varies at different times and under different lighting conditions, a multi-image selection strategy is adopted to improve the recognition robustness.

[0043] Obtain the first set of suspected vehicles V s_1 and the corresponding set of captured images for each vehicle. For each target vehicle in the set Simultaneously, it acquires multiple captured images within the statistical time window w, along with the corresponding vehicle bounding box localization results or license plate localization results. The quality of each single-frame image is then evaluated and filtered, specifically targeting vehicles. Each candidate image Calculate its overall quality score The score is a weighted fusion of one or more sub-indicators, such as vehicle frame area, sharpness score, brightness / exposure score, occlusion ratio, and vehicle front / body integrity.

[0044] Vehicle frame area: Prioritize images where the vehicle occupies a suitable proportion of the frame, discarding samples that are too far away (few pixels) or too close (severe truncation). Sharpness score: Calculate image gradients based on the Laplacian or Sobel operator to evaluate focus sharpness. Brightness / exposure score: Analyze image histograms, discarding overexposed (overly bright license plate) or underexposed (failed nighttime lighting). Occlusion ratio: Based on semantic segmentation results, evaluate the proportion of the vehicle obscured by guardrails, other vehicles, or shadows. Front / body integrity: Detect the visibility of key components (headlights, A-pillar, side windows) at the front or side of the vehicle. Overall quality score. The top k images are selected as the high-quality candidate set. Where k is a configurable parameter, preferably 1, 3 or 5.

[0045] Secondly, inference is performed on the vehicle model from a single frame image, and... Each image in The inputs are fed into a pre-trained vehicle classification model (e.g., based on a deep convolutional neural network), which outputs probability distributions for three key vehicle types:

[0046] in, This represents the probability that the vehicle type is a sedan. This represents the probability that the vehicle category is truck. This represents the probability of vehicle category as a miniature / closed van. Simultaneously, it records the Top-1 category (the vehicle category with the highest probability in the output probability distribution), Top-1 confidence (the maximum value in the probability distribution, i.e., the model's prediction confidence for the Top-1 category), Top-2 confidence (the second largest probability value in the output probability distribution), and the marginal confidence of the difference between the two for each image. , used to represent the target object The edge score or interval score, among which, Indicates the target The confidence level of the category with the highest predicted probability. Indicates the target The confidence score of the class with the second highest predicted probability is used to measure the certainty of the model's prediction.

[0047] Next, the vehicle-level probabilities are fused. To eliminate random errors in single-frame images, a quality-weighted averaging method is used to fuse the results from multiple frames into a vehicle-level probability distribution. The calculation formula is as follows:

[0048] The category with the highest probability is selected as the Top-1 category for vehicle-level classification, denoted as... And calculate the vehicle-level confidence difference. .

[0049] Then, based on the above vehicle-level probability fusion results, classification is performed, and vehicles that meet the following conditions are used to generate the second suspected vehicle set V. s_2 One method is to directly determine if the vehicle-level Top-1 category is Van (i.e., ...). Or the probability of a van satisfying ,in A preset threshold is set for the probability of a van being illegally operated; preferably, it can be between 0.6 and 0.8. If the above conditions are met, the vehicle is directly identified as a key vehicle for illegal operation. Secondly, it can be downgraded and retained, or manually reviewed. ,but If the model falls within a preset gray range (e.g., [0.3, 0.6]), or if the model lacks sufficient determinism, i.e. Vehicles falling below the preset minimum threshold for certainty (and belonging to the above categories) are marked as "uncertain," and their model weight is reduced in subsequent comprehensive scoring, or they are sent to a manual review platform for secondary confirmation. Thirdly, they are prioritized for removal; if... and If the vehicle is identified as a freight vehicle, it is generally not an illegally operated vehicle, or the vehicle image quality does not meet the minimum recognition requirements (such as being extremely blurry or severely obscured), or historical data shows that the vehicle category has been consistently classified as a non-priority vehicle type for a long time (such as being consistently identified as a Car). Vehicles that meet any of the above conditions will be preferentially removed.

[0050] Finally, calculate the category score that matches the vehicle category. .

[0051]

[0052] Among them, the weight coefficients satisfy 1> > ≥0. For example, configurable. =0.6 (indicating high risk of uncertainty) =0.2 (indicating low-risk private cars).

[0053] The final output is the second set of suspected vehicles, V. s_2 and category scores This provides a behavioral and visual basis for subsequent detection.

[0054] This embodiment overcomes the problem of inaccurate single-frame image recognition caused by weather, lighting, and vehicle speed in checkpoint images by introducing a multi-image quality assessment and quality-weighted probability fusion mechanism, thereby reducing detection errors and improving usability in complex road conditions.

[0055] Step S104: When the target vehicle belongs to a preset vehicle type, perform a front passenger feature analysis based on the bidirectional driving time sequence data to obtain the front passenger characteristics.

[0056] In this embodiment, based on the vehicle behavior features and vehicle category features identified in the above embodiments, the characteristics of the front passenger are further analyzed. During the operation of illegal carpooling or charter vehicles, the occupants in the front passenger position are often not fixed and may have one-way empty or full-load characteristics. By longitudinally comparing the front passenger features of the same vehicle in two-way traffic in the time dimension, it is possible to effectively distinguish between fixed commuting and mobile passenger transport. By utilizing the vehicle frame and license plate positioning results from the previous steps, there is no need to re-inspect the entire vehicle, and high-precision region cropping can be performed directly, which improves the accuracy of front passenger identification and reduces computational costs.

[0057] Step S105: Based on the vehicle behavior characteristics, vehicle category, and passenger characteristics, determine whether the target vehicle is a suspected illegal operating vehicle.

[0058] In this embodiment, by fusing vehicle behavior characteristics, vehicle category, and front passenger characteristics, the problem of large errors caused by relying on a single feature in the prior art is solved. By introducing the changing characteristics of the front passenger, the boundary between commuter and illegal operation vehicles is effectively distinguished, thereby improving the accuracy and efficiency of illegal operation detection on highways.

[0059] This application focuses on a single two-way checkpoint. It constructs two-way driving time-series data of vehicles from captured images, extracts vehicle behavior characteristics representing operational tendencies from this time-series data, and filters out a first set of suspected vehicles that conform to operational time patterns. Based on this, vehicle categories are identified, and key vehicles meeting these criteria are used to generate a second set of suspected vehicles and vehicle category scores. Finally, combined with the characteristics of the front passenger, the application determines whether the vehicle is suspected of being illegally operating. This improves detection accuracy and efficiency, reduces detection errors, and enhances the intelligence and precision of illegal vehicle operation detection.

[0060] Optionally, such as Figure 2 As shown, in step S104, the characteristics of the front passenger include identity features. The step of analyzing the front passenger characteristics based on the bidirectional driving time-series data to obtain the front passenger characteristics includes: Step S201: Perform face recognition on the passenger side area in each frame of the vehicle passing image in the bidirectional driving time sequence data to obtain the face features at each vehicle passing time.

[0061] In this embodiment, based on the vehicle frame obtained in step S103, the cabin area is cropped according to a preset geometric mapping relationship. The cabin area preferably covers the windshield and the visible area of ​​the front passengers, ensuring that it includes the driver's seat, the passenger seat, and the visible space from the top edge of the windshield to the hood. Because the checkpoint camera is usually installed on the centerline of the lane or on the side of the road, the passenger seat may be mirrored in the image (i.e., the passenger seat is on the left in some lanes and on the right in others). It is necessary to dynamically determine the position of the passenger seat. If it is determined that the passenger seat is on the left side of the image (such as the oncoming lane captured by the roadside camera), the left half of the cabin area is extracted. If it is determined that the passenger seat is on the right side of the image (such as the gantry camera capturing the center of the image or the same-direction lane captured by the roadside camera), the right half of the cabin area is extracted. To avoid the passenger seat being missed because it is exactly near the centerline, an overlapping tolerance area is set at the centerline (e.g., offset by 10% width to the left and right of the centerline). That is, the passenger seat area not only includes the strict half but also the intersection area near the centerline, ensuring that the human body can still be completely captured when the posture is shifted.

[0062] Scan the passenger side region in the vehicle passing images from the two-way driving time series data, and output a set of all possible face candidates. :

[0063] Among them, each candidate face Includes bounding box coordinates and detection confidence. and the area of ​​the face frame If multiple faces are detected, calculate the number of each candidate. Overall rating :

[0064] To further eliminate interference, filtering conditions can be set in combination. First, location prioritization: prioritize faces whose center point is closer to the geometric center of the passenger area, and farther from the window frame or A-pillar shadow. Second, minimum area threshold: set a minimum face pixel area to filter out small targets that are too far away to extract effective features. Third, sharpness requirements: calculate the Laplacian variance of the face region to ensure the image remains in focus and meets the minimum quality requirements for feature extraction.

[0065] Select those that satisfy the above constraints and The highest candidate is selected as the best face. If no face is detected or all faces do not meet the constraints, the passenger seat status at that moment is marked as "unoccupied" or "obscured".

[0066] Next, facial feature vector extraction and normalization are performed. The selected optimal face... The input is fed into a pre-trained face recognition feature extraction network, which outputs a high-dimensional floating-point feature vector. (where d is preferably 128 or 512), this vector encodes the key biometric features of a face, so that the vectors of the same person at different times and under different lighting conditions are close in distance in high-dimensional space.

[0067] To eliminate the modulus differences caused by light intensity and shooting distance, L2 normalization is performed on the feature vectors. L2 normalization involves dividing a non-zero vector by its L2 norm (Euclidean length) to transform it into a unit vector of length 1, while maintaining its direction. This yields the unit feature vector. The normalized vectors are easier to calculate the cosine similarity directly, and have higher numerical stability.

[0068] To accelerate the large-scale face comparison process, the normalized floating-point vector can be further processed. The data is converted to a binary hash code. For example, each dimension is mapped to +1 or -1 using a sign function. This hash code is primarily used to quickly recall candidate matching objects from massive amounts of historical data, followed by secondary refinement using high-precision floating-point vectors. This solves the problems of driver / passenger mirror image confusion and missed detection of the center line in checkpoint images, significantly improving the signal-to-noise ratio of passenger detection. It ensures that even in low-resolution or long-distance capture scenarios, the most suitable face samples for feature extraction can be selected, improving the reliability and efficiency of face screening.

[0069] Step S202: Analyze the facial similarity between different passing vehicle images based on the facial features, and obtain identity features to reflect the changes in the identity of the front passenger based on the facial similarity.

[0070] In this embodiment, for each record containing a valid co-driver's face, its high-dimensional feature vector after L2 normalization is extracted. Aggregate all successfully extracted feature vectors to construct a set of feature vectors for the passenger side of the vehicle. :

[0071] in, This represents the total number of times the front passenger was effectively detected within the statistics window for this vehicle.

[0072] To quantify the similarity of front passenger passengers at different time points, a set was calculated. Any two eigenvectors and ( Cosine similarity between :

[0073] Since the vectors have been L2 normalized (magnitude of 1), the above dot product operation is directly equivalent to calculating the cosine of the angle between the two vectors. The closer the cosine value is to 1, the more similar the facial features detected at the two time points are, meaning the greater the probability that they belong to the same person. A threshold for determining if someone is the same person is set. Iterate through all calculated similarity values ,like Then determine the first Second and third The same front passenger was observed in the previous observations. Then determine the first Second and third The co-pilot observed this time has changed, meaning it is not the same occupant.

[0074] To quantify occupant mobility within a statistical time window, the rate of change of front passenger occupants was defined and calculated. :

[0075] Where the numerator represents the number of pairs that were determined to be "different persons" in all pairwise comparisons, and the denominator represents the number of pairs that were determined to be "different persons". This represents the number of possible pairwise combinations among n samples. The value range is [0,1]. The closer it is to 1, the more frequently the front passenger changes, and the more obvious the carpooling characteristics of the vehicle. The closer it is to 0, the more stable the front passenger is, which is consistent with the characteristics of family car use or fixed commuting. The identity characteristics may include the change rate of the front passenger.

[0076] Optionally, identity features may also include the estimated number of different front-seat passengers and the effective observation quality score. As a supplement to or alternative to the aforementioned threshold determination method, an unsupervised clustering algorithm can be used to analyze the front-seat feature vector set. Analysis was performed using density-based clustering algorithms (such as DBSCAN or K-Means). Clustering is performed on high-dimensional vectors in the algorithm, and the number of cluster centers or clusters output by the algorithm is denoted as . , This is the estimated number of different front passenger passengers appearing within the statistical window w, with a threshold set for the number of identities. ,like (For example 3) If the passenger in the front seat of the vehicle is deemed to have poor stability and exhibits obvious behavior of carrying passengers in multiple batches, then there is a very high suspicion of illegal operation. If the passenger is identified as a fixed individual, the vehicle can be removed from the suspect list. Based on the number of successful extractions of valid passenger feature vectors, image clarity, and the time span covered by the samples, the sufficiency and reliability of the samples are comprehensively evaluated to obtain an effective observation quality score. If the number of valid samples is too small, the score will be lowered, indicating that subsequent judgments should be made with caution.

[0077] By calculating the above parameter values, illegal operations are quantified, effectively solving the problem of confusion between fixed commuter vehicles and illegal operating vehicles, and reducing detection errors.

[0078] Optionally, such as Figure 3 As shown, in step S104, the characteristics of the front passenger include quantitative characteristics. The step of analyzing the front passenger characteristics based on the bidirectional driving time series data to obtain the front passenger characteristics includes: Step S301: Perform face recognition on the passenger area in each frame of the vehicle passing image in the bidirectional driving time sequence data to obtain the number of passengers in the passenger seat each time the vehicle passes.

[0079] In this embodiment, the passing images of the target vehicle in the bidirectional driving time series data within the statistical time window are traversed, and the following operations are performed for each passing image frame: First, based on the acquired vehicle bounding box and license plate key points, the image block of the passenger seat area is accurately cropped according to the preset geometric mapping relationship; then, the trained dense head detection model is called to perform visual analysis on the area. This model focuses on recognizing the head contour, shoulder features and hair texture on the passenger seat, and can effectively perceive even slight occlusion. The number of independent heads detected in the area is counted to determine the number of people in the passenger seat at the time of the passing, and recorded as zero people, one person or more people.

[0080] Step S302: Determine whether the target vehicle has an overload of passengers in the passenger seat based on the number of passengers in the passenger seat, and count the number of times the passenger seat is overloaded.

[0081] In this embodiment, after obtaining the number of front passenger seats for each passing vehicle, compliance is determined based on the vehicle type. According to road traffic safety regulations, the permitted number of front passenger seats for various vehicle types is pre-set. Typically, sedans, SUVs, and MPVs have a permitted number of one front passenger. The actual number of front passenger seats is compared with the permitted number. If the actual number exceeds the permitted standard, it is determined that there is overcrowding in the front passenger seat during that passage. Then, all vehicle passage records within the entire statistical time window are scanned to record the cumulative number of times front passenger overcrowding occurred. .

[0082] Step S303: Compare whether the number of front passenger seats is the same in adjacent round-trip vehicle images, and obtain quantitative features to reflect the change in the number of front passenger seats based on the comparison results and the number of overcrowding.

[0083] In this embodiment, based on the pairing relationship in the two-way driving time sequence, the number of passengers in adjacent or corresponding round-trip journeys is compared. Specifically, paired entry and exit records are extracted, and the difference in the number of passengers in the front passenger seat is compared. If the front passenger seat is empty or has only a few passengers when leaving the city, but the number of passengers in the front passenger seat increases significantly when returning, or if the number of passengers in the front passenger seat is unstable in round-trip journeys on different dates, it is determined that the number of passengers in the front passenger seat has changed. The frequency of such changes in number of passengers is further statistically analyzed throughout the entire statistical time window. Finally, a quantitative feature reflecting the change in the number of passengers in the front passenger seat is generated. This quantitative feature may include the number of times the vehicle was overloaded, the frequency / number of changes in the number of passengers in the front passenger seat, and the difference in the change value.

[0084] For example, from time series data Filter all records where the direction of travel is "out". Sort by time in ascending order, and in The records that meet the criteria of entering from all directions within a preset time window are selected. Calculate the difference in the number of co-pilots between the two. ,like If the number of passengers in the front passenger seat is inconsistent when leaving the city and entering the city (e.g., 0 passengers when leaving the city and 2 passengers when entering the city), it is considered a change in the number of passengers. The number of times the number of passengers in the front passenger seat changes within the time window w is counted. .

[0085] By introducing passenger count statistics, the limitations of relying solely on facial recognition are overcome, and the detection accuracy of illegal operations is improved.

[0086] Optionally, the front passenger features also include interaction features between the front passenger and the driver. The step of analyzing the front passenger features based on the bidirectional driving time-series data to obtain the front passenger features includes: recognizing one or more interaction actions of the driver and front passenger in each frame of the bidirectional driving time-series image, including mouth shape, body movements, interaction distance, and head angle, and obtaining the interaction features based on the interaction action recognition.

[0087] In this embodiment, interaction features refer to the set of features extracted from the synchronous visual behavior analysis of the driver and passenger in the same vehicle passing image (or the same trip), which describes the communication, distance, posture synchronization, and object transfer relationship between the two. The pre-trained cockpit behavior analysis model is used to perform parallel analysis of the occupants in the driver's and passenger's areas, focusing on identifying the following interaction features: detecting the mouth states (e.g., speaking, stillness) and head posture angles of the driver and passenger. If both mouths are simultaneously open and their head angles are both biased towards the vehicle's centerline (i.e., they are looking at each other or talking), then "verbal communication" is identified. Key hand points and limb bones are captured. If the driver's right hand leaves the steering wheel and extends towards the passenger area, or the passenger's hand extends towards the center console or glove box, then physical interaction is identified. The pixel distance between the center point of the driver's shoulder (or head) and the corresponding point of the passenger is calculated and converted into physical distance using calibration parameters. If the interaction distance is significantly less than a certain threshold, it is identified as close contact, indicating that the relationship between the two may not be that of acquaintances, but rather a temporary driver-passenger relationship. Analyzing the synchronicity of head turning in the driver and passenger seats, if multiple consecutive checkpoint captures show that the driver and passenger heads simultaneously turn towards each other or simultaneously look to a certain side and rear, this synchronicity indicates that there is interactive characteristic between the two parties.

[0088] By detecting the interaction characteristics between the front passenger and the driver, the blind spots of existing technologies are filled, the effective detection rate under complex road conditions is improved, and the detection error is reduced.

[0089] Optionally, such as Figure 4 As shown, determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle's behavioral characteristics, vehicle category, and the characteristics of the front passenger includes: Step S401: Determine the initial screening score that matches the vehicle behavior characteristics.

[0090] Step S402: Determine the category score that matches the vehicle category.

[0091] Step S403: Determine the co-driver score that matches the characteristics of the co-driver.

[0092] In this embodiment, one or more parameters of the front passenger characteristics in the above embodiments are numerically quantified, and the quantified values ​​are weighted and calculated to obtain the front passenger score. :

[0093] in, Indicates the rate of change in the number of front passenger passengers. This indicates the estimated number of different front passenger occupants appearing in the statistics window. Indicates the quality score of valid observations. This indicates the cumulative number of times the co-pilot's seat has been overcrowded. Indicates the number of changes in the number of front passenger occupants. This represents the difference in the number of passengers in the front seat. This represents the quantitative score reflecting the interaction characteristics between the driver and the passenger. This represents the weighting coefficients. Among them, the closer the interaction between the driver and passenger, the lower the passenger's score, indicating a negative correlation. The other six parameters are positively correlated with the passenger's score. Optionally, the parameters involved in calculating the passenger's score can include only a few of the seven parameters in the above formula, such as only including... , and .

[0094] The execution order of steps S401 to S403 is not limited; for example, they can be executed in parallel or in a specific order. Figure 4 The execution will proceed in the order specified in the instructions.

[0095] Step S404: Calculate the illegal operation risk score of the target vehicle based on the initial screening score, category score, and co-driver score; determine whether the target vehicle is a suspected illegal operation vehicle based on the illegal operation risk score.

[0096] In this embodiment, based on the initial screening score Category Score And the co-pilot's score Calculate the illegal operation risk score of the target vehicle. :

[0097] in, For weighting coefficients. When Greater than or equal to the preset risk threshold If so, the target vehicle is determined to be a suspected illegally operating vehicle.

[0098] Optionally, before determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle behavior characteristics, vehicle category, and front passenger characteristics, the method further includes: analyzing the front passenger's riding situation in vehicles other than the target vehicle to obtain the associated occupant characteristics of the target vehicle.

[0099] In this embodiment, after identifying the front passenger of the target vehicle, the passenger's biometric vector (such as L2-normalized facial features) is used as the query key to perform a cross-vehicle search in the entire index database, comparing whether the front passenger of the current target vehicle is the same person as the passenger captured in other vehicles. The characteristics of associated occupants mainly reflect the breadth of activity of the front passenger in a specific time period (such as the last 7 days). First, it counts the number of different vehicles the occupant has ridden in, i.e., the number of associated vehicles. Second, it analyzes the time density and frequency of the occupant changing vehicles. For example, if it is found that a front passenger in a target vehicle not only appears in the current vehicle, but also appears in two or more other vehicles within a week, and the time interval between the changes is extremely short, then the occupant is marked as a high-frequency mobile person. This phenomenon of one person in multiple vehicles is a typical characteristic of illegal chartering. This characteristic of associated occupants is fed back into the detection logic of the target vehicle. If it is confirmed that the front passenger of the target vehicle has obvious cross-vehicle mobility attributes, even if the frequency of the target vehicle's round trips or the change rate of the front passenger does not reach the highest threshold, the comprehensive risk score of the vehicle will be increased based on this strong correlation, which helps to reduce detection errors.

[0100] The step of determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle behavior characteristics, vehicle category, and front passenger characteristics includes: determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle behavior characteristics, vehicle category, front passenger characteristics, and associated passenger characteristics.

[0101] In this embodiment, the vehicle behavior characteristics, vehicle category, front passenger characteristics, and associated passenger characteristics are weighted and fused to reduce detection errors, avoid the ineffective use of resources for non-target vehicles, and improve detection efficiency.

[0102] Optionally, the passenger's riding situation in vehicles other than the target vehicle is analyzed to obtain the associated passenger characteristics of the target vehicle, including: obtaining the passenger identifiers in the passenger feature database that pass through the target road segment within a statistical time window; obtaining the vehicle identifiers associated with the passenger identifiers; and determining the associated passenger characteristics based on the number of associated vehicle identifiers and / or the vehicle analysis data corresponding to the associated vehicle identifiers.

[0103] In this embodiment, the passenger feature database is used to store passenger information for all vehicles passing through the target road segment within a statistical time window. Each record in the database contains at least the following fields: capture timestamp, vehicle identifier (such as license plate number), and corresponding passenger identifier (i.e., L2-normalized face feature vector or its binary hash code). When it is necessary to detect a target vehicle, the set of all passenger identifiers for that vehicle is first extracted from the target vehicle's historical capture data. , Subsequently, using these identifiers as query keys, perform a reverse search in the co - passenger feature database of the said vehicle. Specifically, filter out all the capture records in which any one of the identifiers has appeared within the statistical time window. Exemplarily, for the co - passenger , it is retrieved that he / she took a vehicle with license plate "Yue Axxxxx" last Monday, took a vehicle with license plate "Yue Bxxxxx" last Wednesday, and took "Yue Axxxxx" again last Friday. Then, "Yue Axxxxx" and "Yue Bxxxxx" will be recognized as two different vehicle identifiers. Summarize all the associated vehicles. If is associated with 2 different vehicles, is associated with 1 vehicle, and and the associated vehicles do not overlap, then the co - passenger carried by the target vehicle is associated with a total of 3 different vehicles. Record this value as the number of associated vehicle identifiers .

[0104] In this embodiment, based on the number of associated vehicle identifiers , quantify and determine the associated passenger characteristics of the target vehicle . If , it indicates that the co - passenger only appears in the target vehicle and belongs to a fixed passenger. If , it indicates that the co - passenger has cross - vehicle activities. The larger the value, the stronger the mobility of the passenger and the higher the risk of being an illegal carpool passenger. The associated passenger characteristics reflect the degree of correlation between the target vehicle and illegal operation on the social surface. If the co - passenger of the target vehicle frequently changes to other vehicles, it indicates that the target vehicle is more likely to be an illegal operation vehicle.

[0105] The vehicle analysis data corresponding to the associated vehicle identifiers can be one or more of the vehicle behavior characteristics, vehicle category, co - passenger characteristics of the vehicle, initial screening score of the vehicle, category score, co - passenger score, illegal operation risk score, etc. of the corresponding vehicle. When the associated vehicle identifiers include multiple ones, the corresponding vehicle analysis data may include multiple copies. The vehicle analysis data can be determined by the method in the above embodiment.

[0106] Optionally, as Figure 5 described, determining whether the target vehicle belongs to a suspected illegal operation vehicle based on the vehicle behavior characteristics, vehicle category, co - passenger characteristics, and the associated passenger characteristics includes: Step S501, determine the associated passenger score that matches the associated passenger characteristics.

[0107] In this embodiment, if , it indicates that the co - passenger has cross - vehicle activities, and calculate the associated passenger score Among them, the score of associated passengers. The numerical value and It is positively correlated with the initial screening score, category score, co-pilot score, and illegal operation risk score, respectively. Through a pre-set calculation model, the corresponding... For example, it could be based solely on the number of associated vehicle identifiers. Quantify and determine the associated occupant characteristics of the target vehicle. It can also be combined with the number of vehicle identification tags. The characteristics of associated occupants are calculated by combining the illegal operation risk score of each associated vehicle. .

[0108] Step S502: Determine the initial screening score that matches the vehicle behavior characteristics.

[0109] Step S503: Determine the category score that matches the vehicle category.

[0110] Step S504: Determine the co-driver score that matches the characteristics of the co-driver.

[0111] The execution order of steps S501 to S503 is not limited; for example, they can be executed in parallel or sequentially. Figure 5 The process is performed in the order described above. The calculation methods for the initial screening score, category score, and co-pilot score can be found in the above-described embodiment, and will not be repeated here.

[0112] Step S505: Calculate the illegal operation risk score of the target vehicle based on the initial screening score, category score, associated passenger score, and co-driver score.

[0113] In this embodiment, if the initial screening score of the target vehicle is... Passenger's score The score far exceeds the preset threshold, and the category score is... If a vehicle is highlighted as a priority model (e.g., a van), it is immediately identified as an illegally operating vehicle. If the initial screening score for the target vehicle is... Passenger's score Far below the preset threshold, and the category score If the vehicle is displayed as a non-priority model (such as a Truck), it is directly determined that the target vehicle is not an illegally operating vehicle. If the target vehicle's initial screening score is... Passenger's score If the vehicle's risk level is close to a preset threshold (e.g., slightly above the threshold or in a gray area) and the vehicle type is of medium to low risk (e.g., a regular sedan or SUV), then it is necessary to introduce associated occupant scores to further determine whether the target vehicle is an illegal operating vehicle. The illegal operating risk score is calculated using the following formula. .

[0114]

[0115] in, These are the weighting coefficients.

[0116] Step S506: Determine whether the target vehicle is a suspected illegal operation vehicle based on the illegal operation risk score.

[0117] In this embodiment, when Greater than or equal to the preset risk threshold If this occurs, the target vehicle is determined to be a suspected illegally operating vehicle. Risk threshold. It can be any preset, suitable value, for example, its value can be related to the aforementioned risk threshold. same.

[0118] This embodiment optimizes the allocation of computing resources, improves computing efficiency and accuracy, and enhances the intelligence level of illegal vehicle detection.

[0119] In one embodiment, the illegal operation risk score of the target vehicle can be calculated first based solely on the initial screening score, category score, and co-driver score. If the calculated illegal operation risk score is less than the risk threshold... But greater than the basic threshold If so, the system returns to analyze the passenger's riding situation in vehicles other than the target vehicle, obtains the associated passenger characteristics of the target vehicle, and determines whether the target vehicle belongs to a suspected illegal operating vehicle based on the vehicle behavior characteristics, vehicle category, passenger characteristics, and associated passenger characteristics.

[0120] Among them, when the illegal operation risk score calculated based on the above three factors is at the basic threshold and risk threshold If the situation is as described above, it means that the possibility of the target vehicle not being a suspected illegal operating vehicle cannot be ruled out. At this point, the associated passenger score can be further calculated. The corresponding illegal operation risk score can be recalculated by combining the initial screening score, category score, associated passenger score, and the co-driver score. The illegal operation risk score calculated based on these four factors can be used to further determine whether the target vehicle is a suspected illegal operating vehicle.

[0121] This application focuses on a single two-way checkpoint. It constructs two-way driving time-series data of vehicles from captured images, extracts vehicle behavior characteristics representing operational tendencies from this time-series data, and filters out a first set of suspected vehicles that conform to operational time patterns. Based on this, vehicle categories are identified, and key vehicles meeting these criteria are used to generate a second set of suspected vehicles and vehicle category scores. Finally, combined with the characteristics of the front passenger, the application determines whether the vehicle is suspected of being illegally operating. This improves detection accuracy and efficiency, reduces detection errors, and enhances the intelligence and precision of illegal vehicle operation detection.

[0122] In one embodiment, a computer-readable storage medium is provided having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the steps in the above method embodiments.

[0123] In one embodiment, an electronic device is also provided, including one or more processors; and a memory storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the steps in the above-described method embodiments. The electronic device may be a backend server of a traffic management system.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting illegally operating vehicles, characterized in that, include: Acquire the two-way traffic timing data of the target vehicle on the target road segment; Vehicle behavior features that characterize the vehicle's operational tendency are extracted from the driving time series data. Suspected operations are initially screened based on the vehicle behavior features to obtain the initial screening results. When the initial screening results indicate that the target vehicle meets the initial operational behavior, vehicle type identification is performed based on the two-way driving time sequence data to obtain the vehicle category of the target vehicle; When the target vehicle belongs to a preset vehicle type, the characteristics of the front passenger are obtained by analyzing the characteristics of the front passenger based on the bidirectional driving time sequence data. Based on the vehicle's behavioral characteristics, vehicle category, and the characteristics of the front passenger, it is determined whether the target vehicle is a suspected illegally operating vehicle.

2. The method for detecting illegally operating vehicles according to claim 1, characterized in that, The vehicle behavior characteristics include one or more of the following: number of trips, trip frequency, trip time period, two-way round-trip pairing rate, and number of two-way round-trip pairings. The preliminary screening of suspected operations based on the vehicle's behavioral characteristics, to obtain the preliminary screening results, includes: Based on the vehicle behavior characteristics, the initial screening score of the target vehicle is calculated according to the preset initial screening calculation model. The initial screening result is obtained based on the initial screening score. When the initial screening score exceeds the preset first score threshold, it is determined that the initial screening result indicates that the target vehicle meets the preliminary operational behavior.

3. The method for detecting illegally operating vehicles according to claim 1, characterized in that, The characteristics of the front passenger include identity features. The front passenger feature analysis based on the bidirectional driving time series data yields the following characteristics: Face recognition is performed on the passenger side area of ​​each frame of the vehicle passing image in the bidirectional driving time sequence data to obtain the face features at each vehicle passing time; Based on the facial feature analysis, the facial similarity between different vehicle images is obtained, and based on the facial similarity, identity features that reflect the changes in the identity of the front passenger are obtained.

4. The method for detecting illegally operating vehicles according to claim 3, characterized in that, The characteristics of the co-pilot also include quantitative characteristics. The method of performing front passenger characteristic analysis based on the bidirectional driving time sequence data to obtain front passenger characteristics includes: Face recognition is performed on the passenger side area of ​​each frame of the vehicle passing image in the bidirectional driving time sequence data to obtain the number of passengers in the passenger side each time the vehicle passes. Determine whether the target vehicle has an overload of passengers in the front passenger seat based on the number of passengers in the front passenger seat, and count the number of times the vehicle has an overload of passengers; By comparing whether the number of front-seat passengers is the same in adjacent round-trip vehicle images, quantitative features reflecting changes in the number of front-seat passengers are obtained based on the comparison results and the number of times the number of passengers is overloaded.

5. The method for detecting illegally operating vehicles according to claim 3, characterized in that, The front passenger characteristics also include the interaction characteristics between the front passenger and the driver. The front passenger characteristics are obtained by analyzing the front passenger characteristics based on the bidirectional driving time-series data, including: For each frame of the vehicle passing image in the bidirectional driving time sequence data, one or more interactive actions of the driver and passenger are identified, including mouth shape, body movements, interaction distance, and head angle. The interactive features are obtained based on the identified interactive actions.

6. The method for detecting illegally operating vehicles according to claim 1, characterized in that, The determination of whether the target vehicle is a suspected illegal operating vehicle based on the vehicle's behavioral characteristics, vehicle category, and passenger characteristics includes: Determine the initial screening score that matches the vehicle's behavioral characteristics; Determine the category score that matches the vehicle category; Determine the co-pilot score that matches the characteristics of the co-pilot; The illegal operation risk score of the target vehicle is calculated based on the initial screening score, the category score, and the co-driver's score. The illegal operation risk score is then used to determine whether the target vehicle is a suspected illegal operation vehicle.

7. The method for detecting illegally operating vehicles according to claim 1, characterized in that, Before determining whether the target vehicle is a suspected illegal operating vehicle based on the vehicle behavior characteristics, vehicle category, and front passenger characteristics, the method further includes: Analyze the passenger's riding situation in vehicles other than the target vehicle to obtain the associated occupant characteristics of the target vehicle; The determination of whether the target vehicle is a suspected illegal operating vehicle based on the vehicle's behavioral characteristics, vehicle category, and passenger characteristics includes: Based on the vehicle's behavioral characteristics, vehicle category, front passenger characteristics, and associated passenger characteristics, it is determined whether the target vehicle is a suspected illegally operating vehicle.

8. The method for detecting illegally operating vehicles according to claim 1, characterized in that, Before acquiring the bidirectional driving time sequence data of the target vehicle on the target road segment, the method further includes: Acquire vehicle images captured by monitoring points under the target road segment to form a two-way vehicle image set; Identify the vehicle identifiers in the passing images in the bidirectional vehicle passing image set, and construct the bidirectional driving time sequence data based on the vehicle identifiers and the shooting time.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores executable instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, include: One or more processors; A memory for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to perform the method as described in any one of claims 1 to 8.

Images