Intelligent traffic violation identification optimization method and system

Abstract

The application discloses an intelligent traffic violation identification optimization method and system, and belongs to the technical field of intelligent traffic management.The application generates a cross-frame stable vehicle feature symbol through a key feature point at the front end of a vehicle, effectively overcomes the defects that a license plate recognition is easily affected by shielding and pollution, and improves the robustness of vehicle identity binding in a complex environment.A micro-violation behavior chain is constructed, continuous quantitative analysis of subjective intentions such as driver exploration and hesitation is realized, single result judgment of violation determination is deepened into fine description of a behavior process, intelligent decision support is provided for implementation of graded and classified accurate treatment and optimization of resource allocation, and the scientificity and efficiency of traffic management are improved as a whole.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic management technology, and in particular to an intelligent traffic violation identification and optimization method and system. Background Technology

[0002] With the rapid development of intelligent transportation systems, the automated analysis of traffic monitoring videos using computer vision technology has become a core means of confirming traffic violations and improving road safety and management efficiency. In the specific field of red-light violation detection, related technologies have evolved from the early stage of relying on single physical sensors such as geomagnetic sensors and radar to the current stage dominated by video analysis.

[0003] In recent years, in order to further improve the accuracy of judgment, some companies have begun to delve into the refined analysis of vehicle motion status. However, these technological evolutions have mostly focused on optimizing the back-end judgment logic or seeking to improve the general accuracy of detection and tracking. They have not yet formed an end-to-end systematic solution that goes from accurate vehicle identification to continuous quantitative analysis of driving intentions, and then to attribution of violation scenarios and correlation with historical behavior.

[0004] While existing technologies have automated basic functions, several inherent limitations remain when dealing with complex real-world scenarios. For example, in terms of robustness for vehicle identification and tracking, mainstream systems heavily rely on license plate recognition. However, license plates are susceptible to interference from factors such as obstruction, damage, forgery, and nighttime glare. Furthermore, blind spots exist before vehicles enter a clear recognition distance, leading to trajectory identity jumps or false associations in congested, closely following, and lane-changing scenarios. Secondly, existing methods often rely on binary instantaneous judgments of location and traffic light status, or simply introduce simple speed thresholds, lacking a micro-behavioral characterization of the driver's continuous decision-making process before a violation occurs. This makes it difficult to effectively distinguish between passive line crossings caused by clear subjective or objective factors. Summary of the Invention

[0005] In view of the problems existing in the above-mentioned background technology, the present invention is proposed.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent traffic violation recognition optimization method, comprising: acquiring video streams from traffic monitoring equipment; extracting a continuous frame trajectory sequence of a target vehicle within a set time period; and generating a vehicle feature symbol of the target vehicle based on the immutable features of the vehicle's appearance; analyzing vehicle motion parameters according to the continuous frame trajectory sequence to construct a micro-violation behavior chain of the target vehicle, wherein the micro-violation behavior chain consists of an intent vector sequence used to characterize continuous exploratory behaviors before the violation; when the violation behavior of the target vehicle is identified, tracing back the micro-violation behavior chain and the trajectory sequences of surrounding vehicles, calculating relative behavior parameters between vehicles, evaluating whether the violation behavior is affected by the behavior of the preceding vehicle, and generating a responsibility correlation score; matching the vehicle feature symbol with a historical violation database, and outputting differentiated processing warning information in conjunction with the responsibility correlation score.

[0007] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the extraction of the continuous frame trajectory sequence includes: a first preset time period before the traffic light at the current intersection turns red and a second preset time period after the traffic light turns red are set as a set time period; continuous video frames in a specified area in front of the stop line within the set time period are extracted from the video stream; vehicle detection and tracking are performed on the continuous video frames to obtain the center point coordinates of the target vehicle in the image coordinate system, and the center point coordinates are connected in chronological order to form a continuous frame trajectory sequence.

[0008] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the vehicle feature symbol includes: locating key feature points of the target vehicle in the first frame of the continuous video frames; the key feature points include at least the center points of the two headlights and the four corner points of the air intake grille; calculating the relative positional relationship vector composed of the key feature points, and encoding the relative positional relationship vector as the vehicle feature symbol.

[0009] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the construction of the micro-violation behavior chain includes: calculating the velocity component and acceleration component of the target vehicle in the direction perpendicular to the stop line based on the changes of the continuous frame trajectory sequence in the image coordinate system, as basic motion parameters; calculating the forward step length and hesitation index in fixed time windows based on the basic motion parameters; the forward step length is the positive cumulative displacement of the velocity component in a unit time; the hesitation index is the frequency of the sign change of the acceleration component in the same time window; comparing the forward step length corresponding to each time window with a preset first displacement threshold and comparing the hesitation index with a preset frequency threshold within the set time period; if the forward step length of a time window exceeds the first displacement threshold and the hesitation index exceeds the frequency threshold at the same time, the vehicle movement is determined as a micro-violation segment; combining the forward step length and hesitation index extracted from all the micro-violation segments arranged in chronological order into an intent vector sequence, and defining it as a micro-violation behavior chain.

[0010] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the velocity component and acceleration component are calculated by a difference-calibration algorithm based on image sequences.

[0011] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the calculation of relative behavior parameters between vehicles includes: determining the time and location of the violation of the target vehicle; extracting the nearest preceding vehicle located directly in front of the target vehicle within a third preset time period before the violation time as the associated preceding vehicle; and obtaining the continuous frame trajectory sequence of the associated preceding vehicle within the same time period as the preceding vehicle trajectory sequence; extracting the corresponding part of the target vehicle within the third preset time period from the micro-violation behavior chain; and simultaneously calculating the visual distance occlusion ratio of the associated preceding vehicle on the driver's line of sight of the target vehicle based on the preceding vehicle trajectory sequence and the continuous frame trajectory sequence of the target vehicle.

[0012] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the line-of-sight occlusion ratio includes: in the road plane coordinate system, based on the trajectory of the preceding vehicle, the standard contour data of the vehicle model and the trajectory of the target vehicle, simulating the line-of-sight cone of the traffic lights or road surface observed from the perspective of the driver of the target vehicle; calculating the length of the line-of-sight cone occluded by the associated three-dimensional contour projection of the preceding vehicle, and calculating the ratio of the occlusion length to the total length of the line-of-sight cone as the line-of-sight occlusion ratio.

[0013] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the following steps are included: evaluating whether the violation is affected by the behavior of the preceding vehicle and generating a responsibility correlation score: calculating the theoretical braking distance based on the speed of the target vehicle at the time of the violation and a preset standard deceleration; determining whether the actual vehicle distance between the target vehicle and the associated preceding vehicle at the time of the violation is less than the sum of the theoretical braking distance and a preset safety margin, and generating a first judgment result; and executing the following logical judgment conditions based on a combination of the following three conditions: Condition A: whether the visual distance occlusion ratio exceeds a preset occlusion threshold; Condition B: whether there is at least one micro-violation segment in the micro-violation behavior chain of the target vehicle within the third preset time period; Condition C: whether the first judgment result is yes; and the responsibility correlation score includes three levels: high correlation, medium correlation, and low correlation.

[0014] As a preferred embodiment of the intelligent traffic violation recognition optimization method of the present invention, the logical combination judgment includes: if the following conditions are met simultaneously, it is rated as high correlation: condition A is true and condition C is true; if the following conditions are met simultaneously, it is rated as low correlation: condition B is true and condition A is false and condition C is false; the rest are rated as intermediate correlation.

[0015] Secondly, the present invention provides an intelligent traffic violation recognition and optimization system, comprising: a vehicle feature extraction module, used to acquire video streams from traffic monitoring equipment, extract continuous frame trajectory sequences of a target vehicle within a set time period, and generate vehicle feature symbols of the target vehicle based on the immutable features of the vehicle's appearance; a micro-violation chain construction module, used to analyze vehicle motion parameters based on the continuous frame trajectory sequences, construct a micro-violation behavior chain of the target vehicle, the micro-violation behavior chain being composed of an intent vector sequence, used to characterize continuous exploratory behaviors before the violation; a liability association assessment module, used to, when the violation behavior of the target vehicle is identified, backtrack the micro-violation behavior chain and the trajectory sequences of surrounding vehicles, calculate the relative behavior parameters between vehicles, assess whether the violation behavior is affected by the behavior of the preceding vehicle, and generate a liability association score; and a differential warning output module, used to match the vehicle feature symbols with a historical violation database, and output differentiated warning information in conjunction with the liability association score.

[0016] The beneficial effects of this invention are as follows: This invention generates stable vehicle feature symbols across frames by using key feature points at the front of the vehicle, effectively overcoming the shortcomings of license plate recognition being susceptible to occlusion and dirt, and improving the robustness of vehicle identity binding in complex environments. By constructing a micro-violation behavior chain, it achieves continuous quantitative analysis of the driver's subjective intentions such as probing and hesitation, deepening violation judgment from a single result judgment to a detailed characterization of the behavioral process. This provides intelligent decision support for implementing hierarchical and classified precise processing and optimizing resource allocation, thus improving the scientific nature and efficiency of traffic management overall. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for optimizing intelligent traffic violation recognition methods.

[0019] Figure 2 Flowchart for generating vehicle feature symbols.

[0020] Figure 3 This is a structural diagram of an optimized intelligent traffic violation recognition system. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Figure 1 This is a flowchart of an intelligent traffic violation recognition optimization method according to an embodiment of the present invention. Figure 1 As shown, the intelligent traffic violation recognition optimization method includes: S1: Obtain video stream from traffic monitoring equipment, extract the continuous frame trajectory sequence of the target vehicle within a set time period, and generate the vehicle feature symbol of the target vehicle based on the immutable features of the vehicle's appearance.

[0025] In this embodiment of the invention, step S1 aims to perform stable and traceable temporal modeling and identity representation of the target vehicle within the critical time window before and after the violation occurs, so as to provide basic data support for subsequent violation triggering condition judgment, trajectory consistency analysis and data integrity verification.

[0026] Unlike existing technologies that rely on single or limited frames for vehicle identification, this invention, by defining time windows, spatial regions, and feature construction methods, enables the generated vehicle feature symbols to possess cross-frame consistency, resistance to occlusion, and resistance to changes in lighting conditions. This avoids misassociation or incorrect judgment of vehicles due to instantaneous image anomalies. Figure 2 As shown, the details are as follows: S1.1: The first preset time before the traffic light at the current intersection turns red and the second preset time after it turns red are the set time period. Continuous video frames in the specified area in front of the stop line within the set time period are extracted from the video stream.

[0027] Specifically, the system accesses the traffic management database to obtain the average queue length and traffic flow of vehicles at a certain intersection within a set period (e.g., 30 days) before and after the red light turns on in each cycle of the traffic light (N seconds before and after the red light turns on).

[0028] The first preset duration is determined based on the P percentile of the queue length distribution (e.g., P=85) to ensure that the process of the vast majority of vehicles starting to approach the stop line is covered; the second preset duration is determined based on the probability that vehicles still cross the stop line in the early stage of the red light (e.g., the first 3 seconds) to ensure that possible red light running behavior can be fully captured. The values ​​of some of the data mentioned above (in this invention) can be selected based on expert experience.

[0029] The designated area before the stop line is a static, polygonal region of interest in the image coordinate system. It is defined as follows: on the as-built drawings of the road project or a high-precision electronic map, using the stop line as a baseline, extend a preset distance (e.g., 50 meters) in the opposite direction of vehicle travel, and simultaneously expand to both sides according to the lane width, forming a rectangular area on the road surface. Through pre-completed camera calibration, the camera's intrinsic parameter matrix, distortion coefficients, and extrinsic parameter matrix are obtained. ,in For rotation matrix, As the translation vector, using the perspective projection transformation formula, the four corner points of the rectangular area of ​​the ground are transformed from the world coordinate system. Projecting onto the image coordinate system yields the quadrilateral region in the image. Only retain the data located in each frame. The pixel sub-images within the image are used to form a continuous video frame sequence.

[0030] The perspective projection transformation formula is as follows: ; in, The scale factor is non-zero and is a scaling factor in homogeneous coordinates. Its specific value is determined during the solution process. It has no independent physical meaning and is used to ensure that the third component of the vector on the left side of the equation is 1. These are the column coordinates (in pixels) of the target point in the image, usually with the top left corner as the origin. The target point is located in the row coordinates (in pixels) of the image. This is the intrinsic parameter matrix of the camera. Let be the extrinsic parameter matrix of the camera. The rotation matrix is ​​an orthogonal identity matrix that describes the rotation transformation from the world coordinate system to the camera coordinate system. The three column vectors represent the directions of the X, Y, and Z axes of the world coordinate system in the camera coordinate system. The translation vector represents the coordinates of the origin of the world coordinate system in the camera coordinate system.

[0031] It should be noted that by precisely synchronizing traffic light events and defining regulatory-guided time windows, the collected data is ensured to be highly correlated with potential violations in the time dimension, avoiding the enormous computational resource consumption caused by processing massive amounts of irrelevant video data. Furthermore, by using camera calibration parameters to accurately map geographic spatial areas to image pixel areas, spatial focus of analysis is achieved, effectively eliminating interference from other lanes, sidewalks, and distant backgrounds, significantly improving the accuracy and processing speed of subsequent vehicle detection and tracking.

[0032] S1.2: Perform vehicle detection and tracking on the continuous video frames, obtain the center point coordinates of the target vehicle in the image coordinate system, and connect the center point coordinates in time sequence to form a continuous frame trajectory sequence.

[0033] In this embodiment of the invention, a deep learning-based target detection model (such as an optimized YOLO or SSD architecture, etc., which is not the only limitation in this embodiment) is used to process each frame of image in real time, identify all vehicle targets, and output their bounding boxes. To associate the same vehicle in different frames, a multi-target tracking algorithm is used, such as a tracking framework based on Kalman filter prediction and Hungarian algorithm data association.

[0034] The system uses a multi-task detection model (such as the YOLOv8+ classification branch) to simultaneously output bounding boxes, vehicle categories (sedans / SUVs / trucks / buses, etc.) and confidence scores; vehicle model information is stored in the tracking trajectory to establish a vehicle model-3D contour mapping database.

[0035] Specifically, for the i-th vehicle detected in the k-th frame, the midpoint of the bottom edge of its bounding box is taken as the representative point, and its coordinates in the image pixel coordinate system are calculated. , These represent the x-axis and y-axis values ​​of the i-th vehicle detected in the k-th frame, respectively.

[0036] The tracker achieves optimal matching by comparing the combined costs of feature similarity (such as the cosine distance between appearance feature embedding vectors) and spatial distance (such as the IoU intersection-union ratio) between the current frame detection box and the predicted box of the existing tracking trajectory.

[0037] For successfully matched vehicles, their trajectories are updated; for newly appearing vehicles, new trajectories are initialized. Ultimately, a set of center point coordinates ordered by timestamps is generated for each continuously tracked vehicle target, forming a sequence of continuous frame trajectories on the image plane.

[0038] It should be noted that by combining detection and tracking, not only is the instantaneous position of the vehicle obtained, but more importantly, its continuous motion path is constructed. This provides a dynamic basis for determining whether the vehicle continues to cross the stop line after the red light is on, overcoming the deficiency of single-frame judgment in distinguishing between inertial gliding and intentional red light running.

[0039] S1.3: Locate key feature points of the target vehicle in the first frame of the continuous video frames. The key feature points include at least the center points of the two headlights and the four corner points of the grille.

[0040] It should be noted that this step aims to extract visual feature components with strong structure and relative invariance from the front of the vehicle. The frame in which the target vehicle first appears completely and clearly within the ROI in the tracking sequence is selected as the first frame image. For the detection bounding box region of the vehicle in this frame, a dedicated component localization model is used to detect key feature points.

[0041] This component localization model is a convolutional neural network trained on a large amount of vehicle front-end annotation data, outputting a series of heatmaps or coordinate regression values ​​of predefined feature points. Specifically, the component localization model uses a keypoint detection network architecture as its backbone, such as Stacked Hourglass Networks, High-Resolution Net (HRNet), or networks based on encoder-decoder structures (such as variants of U-Net), etc. This embodiment is not limited to any single one; the specific choice and setting can be made according to actual needs. The final output layer of the model outputs heatmaps or coordinate regression values ​​for each predefined keypoint (center of the left front headlight, center of the right front headlight, and four corner points of the air intake grille, a total of 6 points). The model is trained using a large-scale vehicle front-end image dataset. The data needs to cover different vehicle models (sedans, SUVs, trucks, etc.), different lighting conditions (daytime, nighttime, backlight), different weather conditions, and multiple viewing angles. In each training image, the pixel coordinates of the 6 keypoints are accurately labeled by the annotator according to the clearly defined criteria. For example, the center point of the left front headlight is defined as the center of the headlight's geometric contour; the upper left corner of the grille is defined as the intersection of the grille's outer contour and its upper left edge. During training, data augmentation techniques such as random cropping, rotation, scaling, brightness and contrast adjustment, and noise addition are used to improve the model's robustness to different scenarios.

[0042] If a heatmap is used as the output, a pixel-wise weighted mean squared error (MSE) or Focal Loss is used as the loss function to address the imbalance between positive and negative samples (keypoint regions have very few pixels). If coordinate regression is used as the output, a smoothed L1 loss is used to directly supervise the deviation between the predicted and true coordinates. The model is trained end-to-end using a backpropagation algorithm and an optimizer (such as Adam) until the localization accuracy on the validation set converges.

[0043] The key feature points that must be specified include: the center point of the left front headlight. Center point of the right front headlight And the upper left of the air intake grille Top right Bottom left and the bottom right Four corner points. The component positioning model achieves high-precision positioning by leveraging the relative stability of these points within the vehicle structure (e.g., headlights are typically symmetrically distributed on both sides of the license plate, and the grille is located in the center above the front bumper), maintaining robustness even under partial obstruction or uneven lighting conditions.

[0044] It should be noted that this invention focuses on the feature points of the rigid structure at the front of the vehicle. These points are less affected by the vehicle's motion posture (translation) and are not easily changed by time or operating conditions (e.g., they do not deform), thus providing a reliable geometric basis for generating stable vehicle feature symbols.

[0045] Optionally, if the front is not visible, features of the vehicle's sides / rear are extracted, such as taillight layout, window shape, and body outline. If exterior features cannot be obtained, license plate recognition can be used as an auxiliary method. This invention uses only key front feature points as the ideal and primary source of features for generating vehicle identifiers because of their structural rigidity, direct view from the surveillance camera, stable geometric relationships between feature points, and high discriminative power. The side and rear features, along with license plate information, serve as supplementary feature sources when front features are unavailable, collectively ensuring the robustness of the identity representation process.

[0046] S1.4: Calculate the relative positional relationship vector composed of key feature points, and encode the relative positional relationship vector as the vehicle feature symbol.

[0047] First, select a reference point, for example, the center point of the left front headlight as the origin. Then, calculate the offset vectors of the other five key feature points relative to the origin. These offset vectors are represented in the image coordinate system as follows: ,in Index other feature points, among which, , These are the pixel offsets of the j-th feature point relative to the reference point in the horizontal X and vertical Y directions of the image, respectively.

[0048] To eliminate scale differences caused by varying distances between the vehicle and the camera, all offset vectors were normalized.

[0049] Choose a distance with stable physical meaning as a reference scale, such as the pixel distance between the center points of two headlights. ,in, , The center points of the right and left front headlights are respectively, and each offset vector is... Divide by pixel distance This yields a scale-normalized relative position vector. .

[0050] Finally, all these normalized vectors are arranged in a preset order (e.g., They are concatenated into a one-dimensional total feature vector.

[0051] To further compress and facilitate comparison, the floating-point vector can be converted into a fixed-length numeric string or hash code through quantization (e.g., uniform quantization to the range of integers 0-255) and concatenation, which serves as the vehicle feature identifier for the target vehicle.

[0052] It should be noted that the generated vehicle feature is scale-invariant, relying only on the inherent geometric layout of the vehicle's front components. It is insensitive to variable factors such as the vehicle's color, texture, and rear decals, exhibiting high uniqueness and stability. This feature can serve as a vehicle fingerprint, continuously verifying the consistency of the tracked vehicle's identity in complex scenarios. Particularly in scenarios involving multiple vehicles traveling in parallel or re-identification after a brief loss of tracking, it effectively prevents erroneous trajectory associations, ensuring that traffic violations are accurately attributed to the same vehicle entity.

[0053] S2: Analyze the vehicle motion parameters based on the continuous frame trajectory sequence, and construct the micro-violation behavior chain of the target vehicle. The micro-violation behavior chain consists of an intent vector sequence, which is used to characterize the continuous exploratory behaviors before the violation.

[0054] It should be noted that this invention goes beyond the simple binary judgment logic of location-red light state in traditional red-light violation detection. By deeply deconstructing the continuous motion state of the vehicle before the violation occurs, it constructs a micro-violation behavior chain that can finely characterize the driver's decision-making process. Traditional methods usually only focus on whether the vehicle has crossed the stop line, and cannot distinguish between unintentional crossing (such as inertial coasting) and intentional violation, nor can they provide warnings before the vehicle has completely crossed the line. This invention extracts feature sequences representing complex driving behaviors by parameterizing and fragmenting the vehicle's motion over a set time period, thereby achieving earlier and more accurate violation intent recognition at the behavioral level, providing dynamic behavioral data for subsequent comprehensive judgment. Specifically: S2.1: Based on the changes in the trajectory sequence of consecutive frames in the image coordinate system, the velocity component and acceleration component of the target vehicle in the direction perpendicular to the stop line are calculated using a difference-calibration algorithm based on the image sequence, and used as the basic motion parameters.

[0055] First, using pre-defined camera calibration parameters (including intrinsic matrix, distortion coefficients, and extrinsic matrix), the center point coordinates of the target vehicle in each frame are determined. Back-projected onto the road plane coordinate system.

[0056] Specifically, by solving the collinearity equation and combining it with the known road level assumption, the corresponding coordinates of the center point in the world coordinate system established with the stop line as the reference are obtained. The positive direction of the Y-axis is defined as perpendicular to the stop line and pointing towards the intersection. The above image coordinate to world coordinate transformation can be performed based on direct linear transformation (DLT) and plane homography matrix, accurately mapping the image observation to the actual road plane, ensuring that the subsequent calculation of motion parameters (velocity, acceleration) has real physical meaning and dimensions.

[0057] Subsequently, regarding the world coordinate sequence arranged in chronological order... Calculate the instantaneous velocity components of the vehicle in the Y-axis direction (perpendicular to the stop line direction). The central difference method is used for calculation: for the k-th time point (corresponding to the timestamp) The instantaneous velocity component is calculated as follows: ; in, , , These represent the target vehicle at time [time]. , , At that time, the coordinate value in the world coordinate system of the road plane, perpendicular to the stop line direction (Y-axis), is obtained from the image coordinates through calibration transformation.

[0058] The above calculations utilize information from previous and subsequent frames, which can effectively reduce velocity calculation noise caused by position errors compared to forward or backward differential calculations.

[0059] Furthermore, in obtaining the velocity component sequence Then, the same central difference method was used to calculate the acceleration component sequence perpendicular to the stop line direction. : ; Since motion parameters obtained directly from image sequences by difference calculation may contain high-frequency noise (originating from minor vehicle vibrations, slight shifts in detection boxes, etc.), temporal smoothing filters (such as Savitzky-Golay filters or low-pass Kalman filters) can be introduced before and after the difference calculation to smooth the position or velocity sequences.

[0060] The filter window length needs to be configured according to the video frame rate and vehicle motion characteristics in order to suppress noise while maintaining the realistic motion trend.

[0061] As can be seen, through rigorous coordinate transformation, this invention correlates image observation with real physical motion, ensuring the objectivity and comparability of motion parameter measurements. By employing the central difference method combined with filtering, the smoothness and accuracy of velocity and acceleration estimations are significantly improved without excessive response lag, providing a high-quality data foundation for subsequent microscopic behavior analysis.

[0062] S2.2: Based on the aforementioned basic motion parameters, calculate the forward stride length and hesitation index in fixed time windows; the forward stride length is the positive cumulative displacement of the velocity component within a unit time; the hesitation index is the frequency at which the sign of the acceleration component changes within the same time window.

[0063] Specifically, a fixed-duration time window (e.g., 0.5 seconds) is set. Starting from the beginning of the set time period, this fixed-duration time window is slid in fixed increments (e.g., 1 frame). For each fixed-duration time window, the velocity components perpendicular to the stop line at all time points within the window are extracted. Calculate the step length within this window. Defined as all positive velocity components within the window (i.e. , representing the approximate integral of the motion (moving towards the intersection) over time. The specific calculation formula is: ; in, This represents the total number of sampling points within the window. For the first A timestamp at a specific point in time. The instantaneous velocity component in the Y direction.

[0064] This indicator reflects the net distance a vehicle can effectively move in the direction of the violation (crossing the stop line) within the unit's observation window.

[0065] Analyze acceleration components within the same sliding time window The sign change. Define the hesitation index. Within this time window The number of times the sign changes.

[0066] Technically, it can be detected The number of times the sequence crosses zero is used to determine this. To prevent false zero crossings caused by noise, a small dead-zone threshold is set. (For example, small positive values ​​close to zero), only when The absolute value is greater than And adjacent time intervals A sign change is counted only when the product is less than zero. Ultimately, the hesitation index... This is the count of the number of valid symbol changes within the window. This indicator reflects the driver's control state as they frequently switch between acceleration and deceleration (or vice versa), and is a direct manifestation of the uncertainty of intent.

[0067] It should be noted that the forward step length effectively filters out irrelevant displacements such as reversing and lateral movement, focusing on the substantive advancement in the direction relevant to the violation. The use of a sliding window calculation provides a clear temporal localization for the behavioral characteristics. The hesitation index, through zero-crossing statistics, cleverly captures the oscillation frequency of the acceleration sign, transforming hesitant behavior, which is difficult to quantify directly, into a calculable indicator. The combination of these two methods provides a quantitative tool for describing tentative driving behavior.

[0068] S2.3: Within the set time period, the forward step length corresponding to each time window is compared with a preset first displacement threshold, and the hesitation index is compared with a preset frequency threshold; if the forward step length of a time window exceeds the first displacement threshold and the hesitation index exceeds the frequency threshold at the same time, the vehicle movement is determined as a minor violation segment.

[0069] The first displacement threshold is set based on the minimum displacement observation of typical tentative forward behavior, for example, it can be set to 0.5 meters to 1 meter, which ensures that the segment includes substantial displacement that cannot be ignored. The frequency threshold is set based on statistical analysis of normal smooth driving and hesitant driving, for example, it can be set to at least two changes in the acceleration sign within a 0.5-second window. To enhance adaptability to traffic flow characteristics at different intersections and at different times, an adaptive threshold mechanism can be introduced. For example, the first displacement threshold can be dynamically adjusted based on the average coasting distance of vehicles at the intersection during the yellow light period in history; the frequency threshold can be normalized based on the average hesitation index of all vehicles in the current time period.

[0070] The system sequentially traverses all sliding time windows within a set time period. For the first... The system uses a window to retrieve the calculated forward step length and hesitation index. The execution logic and decision are as follows: if both the forward step length and hesitation index are greater than the corresponding thresholds, then the behavior of the target vehicle within the time window constitutes a minor traffic violation segment. The system records the start and end timestamps of this segment, as well as its core feature values. , which serves as the basis for subsequent analysis.

[0071] S2.4: Combine the forward step length and hesitation index extracted from all the aforementioned micro-violation segments arranged in chronological order into an intent vector sequence, and define it as a micro-violation behavior chain.

[0072] For each minor violation segment identified in S2.3, an intent vector is formed by using the core feature quantity as two components of a two-dimensional vector. This vector simultaneously implies both the intensity (how far forward) and the characteristics (how hesitant) of the behavior.

[0073] Subsequently, according to the order in which these minor traffic violations occurred on the original timeline, all the corresponding intent vectors were arranged to form an ordered sequence. : ,in The total number of segments identified. This corresponds to the Gth segment.

[0074] The constructed micro-violation behavior chain has a clear temporal attribute, with each vector in the chain carrying the start and end time information of the corresponding segment. This behavior chain, as a compact and quantitative description of the evolution of the target vehicle's driving intention within a set time period, is output to subsequent judgments. It not only records whether probing behavior occurred, but more importantly, it records how these behaviors change over time, such as whether the hesitation index gradually decreases while the forward step length gradually increases, providing direct evidence for judging the accumulation and escalation of violation intentions.

[0075] S2.5: Violation determination based on continuous frame trajectory sequence and traffic light status: Acquire traffic light status timing data to determine the moment the red light turns on; compare the world coordinate trajectory of the target vehicle with the stop line position; violation determination conditions: after determining the moment the red light turns on, the vehicle's front wheels cross the stop line, continue moving towards the intersection after crossing the line, and the red light status remains valid while crossing the stop line; if the conditions are met, it is determined as a violation, and the moment of violation is recorded. Location of violation Identify the type of violation (running a red light) and trigger S3 execution.

[0076] It should be noted that this invention elevates the behavioral representation from a single judgment label to a rich sequence of temporal vectors, greatly enhancing the system's ability to describe complex and progressive violations. This behavioral chain, as a standardized intermediate representation, allows subsequent comprehensive violation judgment logic to be based on richer and more interpretable behavioral features, rather than solely relying on the final crossing position, thereby improving the accuracy of judgments, the feasibility of early warnings, and their reliability.

[0077] S3: When the violation of the target vehicle is detected, the chain of minor violations and the trajectory sequence of surrounding vehicles are traced back, the relative behavioral parameters between vehicles are calculated, the violation is evaluated as to whether it is affected by the behavior of the vehicle in front, and a responsibility correlation score is generated.

[0078] It should be noted that in complex traffic flows, the traffic violations of following vehicles may be influenced by external factors such as abnormal driving behavior of preceding vehicles (e.g., emergency braking) or obstruction of vision by large vehicles. This invention, based on the confirmation of a traffic violation by the target vehicle, uses intelligent backtracking and correlation analysis to quantitatively assess whether the violation was influenced by the preceding vehicle and generates a responsibility correlation score. This score can serve as a supplementary reference, enabling violation determination to evolve from a single factual determination to a comprehensive contextual assessment. Specifically: S3.1: Determine the time and location of the violation of the target vehicle, extract the nearest preceding vehicle located directly in front of the target vehicle within a third preset time period before the time of the violation as the associated preceding vehicle, and obtain the continuous frame trajectory sequence of the associated preceding vehicle within the same time period as the preceding vehicle trajectory sequence.

[0079] First, obtain the precise time when the target vehicle's violation (such as the wheels crossing the stop line) occurred from the upstream violation determination. and the vehicle's position in the road plane coordinate system at that moment. .by Using this as a baseline, trace back one third preset duration. (For example, 5 seconds), define the time window for association analysis as... This window covers the critical decision-making and response period before a violation occurs.

[0080] In the time window Within each frame (or keyframe), the following operations are performed in the road plane coordinate system: 1) Determine the target vehicle's driving direction vector; 2) Define a positive sector area along this direction vector as the search area directly ahead (e.g., 15 degrees to the left and right of the centerline); 3) Within this search area, search for all other vehicles besides the target vehicle and calculate their distances to the target vehicle's position; 4) Select the vehicle with the shortest distance that is continuously present (tracked) within this window and establish it as the associated preceding vehicle. Subsequently, retrieve the associated preceding vehicle from the system's tracking database within the exact same time window. The continuous frame trajectory sequence within the data ensures complete temporal synchronization with the target vehicle's data.

[0081] Additionally, if the third preset duration is longer than the first preset duration, the third preset duration will be automatically adjusted to 80% of the first preset duration to match the data window.

[0082] It should be noted that by locating the time of the violation and tracing back over a fixed period, the relevance of the analyzed time period is ensured. By using dynamic spatial direction and distance judgments, rather than fixed lane binding, the system can more accurately identify the vehicle actually affecting following vehicles, adapting to common situations such as lane changes and slight deviations from the lane centerline. Obtaining synchronized trajectories lays the foundation for subsequent refined interactive analysis.

[0083] S3.2: Extract the corresponding part of the target vehicle within the third preset time period from the chain of minor traffic violations. At the same time, calculate the visual distance occlusion ratio of the associated preceding vehicle to the driver's line of sight of the target vehicle based on the trajectory sequence of the preceding vehicle and the continuous frame trajectory sequence of the target vehicle.

[0084] Specifically, based on the generated complete chain of minor traffic violations, all timestamps falling within the associated time window are extracted. The intent vector within the vector forms a sub-behavior chain. This sub-behavior chain reflects the probing and hesitation patterns exhibited by the driver of the target vehicle as they approach the point of violation.

[0085] The proportion of visual distance occlusion includes: In the road plane coordinate system, based on the trajectory of the preceding vehicle, the standard contour data of the vehicle model, and the trajectory of the target vehicle, the line-of-sight cone of the target vehicle driver's view (driver's line-of-sight height and direction) is simulated to observe the traffic lights or road surface ahead; the length of the line-of-sight cone obscured by the associated three-dimensional contour projection of the preceding vehicle is calculated, and the ratio of the obscured length to the total length of the line-of-sight cone is calculated as the line-of-sight obscuration ratio.

[0086] In this embodiment of the invention, in the road plane coordinate system, based on the trajectory points of the target vehicle at each moment (e.g., one keyframe per second), and according to typical data of the vehicle model, a driver's line-of-sight origin is preset above the vehicle. (e.g., 1.2 meters above ground). Define the line of sight direction as the vehicle's forward direction (horizontal), and set a vertical field of vision angle (e.g., ±10 degrees), thus forming a three-dimensional field of vision originating from the driver's line of sight. A frustum-shaped cone of vision extending forward, with its length extending to the key observation target ahead (such as a traffic light or the road surface at a certain distance ahead).

[0087] Based on the vehicle model recognition result of the preceding vehicle (which can be provided by the S1 detection and classification model), a simplified 3D bounding box model (including length, width, and height) is obtained from the standard vehicle model 3D contour database. At the same time, this 3D model is placed at the position corresponding to the preceding vehicle's trajectory point and the heading angle. The projection of this 3D model into the line-of-sight cone space is calculated. Specifically, the length of the line segment intersecting the centerline of the line-of-sight cone and the preceding vehicle's 3D model is calculated. The length of the obstruction reflects the depth of visual obstruction to which the driver cannot directly see traffic lights or further road surfaces. Therefore, the visibility obstruction ratio at that moment is... ,in, This represents the total length of the line-of-sight cone from the origin to the target. Finally, the proportions calculated at multiple sampling moments within the time window are averaged over time to obtain the overall line-of-sight occlusion proportion within that window.

[0088] It should be noted that the extraction of sub-behavior chains retains the target vehicle's own behavioral responsibility data. The calculation of the line-of-sight occlusion ratio, through the introduction of three-dimensional geometric simulation, quantifies the fuzzy concept of occlusion into an objective and continuous proportional value. This not only considers the presence of the vehicle in front but also accurately assesses its actual shielding effect on the critical line of sight of the vehicle behind at different distances and positions.

[0089] S3.3: Calculate the theoretical braking distance based on the speed of the target vehicle at the time of the violation and the preset standard deceleration.

[0090] The speed is directly taken as the calculated value at the time of the violation by the target vehicle. Velocity component perpendicular to the stop line direction This speed is the basis for determining braking demand.

[0091] The theoretical braking distance is calculated using classic kinematic formulas. : ; in, This is the preset standard deceleration value. Standard deceleration The setting should refer to road traffic safety regulations, vehicle engineering standards, and the typical adhesion coefficient of dry paved roads, and is usually taken as a reasonable constant value (e.g., 3.5 m / s). 2 up to 4.5 m / s 2 (between). This calculation assumes an ideal situation where the driver's reaction time is zero and the braking system reaches its maximum deceleration instantaneously, thus yielding the theoretical minimum safe distance.

[0092] S3.4: Based on the actual distance between the target vehicle and the associated preceding vehicle at the time of the violation, determine whether the actual distance is less than the sum of the theoretical braking distance and the preset safety margin, and generate a first judgment result.

[0093] Specifically, at the time of the violation Based on the location of the target vehicle and the position of the vehicle in front The Euclidean distance between the two vehicles is calculated in the road plane coordinate system and used as the actual vehicle spacing. The calculation typically uses the distance between the center points of the two vehicles or the nearest boundary point.

[0094] Set a preset safety margin , can be set to The value is dynamically adjusted according to speed, with 0.5 in seconds. This margin is used to compensate for the idealized assumptions in theoretical calculations. The safe distance threshold is then calculated. The actual vehicle distance is compared with this safety threshold. If the actual vehicle distance is less than the safety distance threshold, the vehicle will be considered safe. If the following distance is too close, the first judgment result is "yes" (i.e., following too closely); otherwise, the first judgment result is "no". This judgment quantifies whether the target vehicle has lost the space for safe braking due to insufficient following distance.

[0095] S3.5: Based on the combined state of the following three conditions, perform logical judgment to output a responsibility correlation score: Condition A: Does the line-of-sight occlusion ratio in S3.2 exceed the preset occlusion threshold (e.g., 0.6)? A high occlusion ratio means that the target vehicle driver's ability to obtain critical road condition information is severely weakened by the vehicle in front.

[0096] Condition B: Determine whether the target vehicle sub-behavior chain extracted in S3.2 is not empty, that is, whether there is at least one micro-violation segment in the micro-violation behavior chain of the target vehicle within the third preset time period. This indicates that the driver has shown clear proactive behavioral characteristics before the violation.

[0097] Condition C: Is the result of the first judgment yes? This indicates that the target vehicle itself is in an unsafe following state, which may limit its autonomous decision-making space.

[0098] A vehicle is considered to have a high degree of correlation if both of the following conditions are met: Condition A and Condition C are met. This logic states that if the vehicle in front causes severe obstruction of vision and forces the target vehicle to follow too closely, then the vehicle in front has a very significant role in inducing or contributing to the traffic violation of the vehicle behind, and thus has a high degree of correlation.

[0099] If all of the following conditions are met, the correlation is assessed as low: Condition B is true and Condition A and Condition C are false; otherwise, the correlation is assessed as intermediate. This logic assumes that if the target vehicle itself has shown a clear probing intent to violate the rules (Condition B), and at the same time there is no serious obstruction of vision (Condition A is not true), and a sufficient safe following distance is maintained (Condition C is not true), then the violation is mainly caused by the target vehicle's own proactive decision, with minimal influence from the vehicle in front, resulting in low correlation.

[0100] The responsibility relevance score includes three levels: high relevance, medium relevance, and low relevance.

[0101] Furthermore, all other situations that do not meet the above logic for determining high and low correlation. For example, severe obstruction of vision but sufficient following distance, or following too closely but also exhibiting probing behavior and without obstruction, etc., are complex mixed situations. Intermediate correlation indicates that the vehicle in front has some influence, but it is not a decisive factor, or multiple factors are intertwined, and responsibility needs further specific analysis.

[0102] It should be noted that the above scoring logic, through the combination of multiple conditions, achieves differentiated processing of complex situations. Combining physical data (A, C) with behavioral data (B) yields a more interpretable and reasonable conclusion regarding the degree of responsibility. The ternary rating output (high, medium, low) provides a clear and evidence-based reference for subsequent manual review or processing, avoiding arbitrary judgments.

[0103] S4: Match the vehicle identifier with the historical violation database, and output differentiated warning information based on the liability correlation score.

[0104] S4.1: Using the vehicle identifier as the search key, query the historical violation database. The historical records include the types and frequencies of historical violations. If there are records with a similarity exceeding a preset similarity threshold, it is determined that the target vehicle has a habitual violation pattern.

[0105] In practice, the vehicle identifiers associated with each record in the historical traffic violation database are pre-indexed using Locality Sensitive Hash (LSH) or similar techniques to support fast approximate nearest neighbor search for high-dimensional vectors. Once the vehicle identifiers of the target vehicle are obtained, they are input into the retrieval system as the query vector.

[0106] The system returns several historical records with the highest similarity to the vehicle's signature, along with their similarity scores. Similarity is calculated using the cosine similarity or the normalized reciprocal of the Euclidean distance between feature vectors. A preset similarity threshold is set (e.g., 0.95). If a historical record's signature satisfies a similarity greater than or equal to the similarity threshold, a match is considered successful, and the system extracts a list of the vehicle's historical traffic violations and their corresponding frequencies.

[0107] Based on a successful match, further analysis of historical violation types is conducted. If the frequency of violations of the same type as the current violation (e.g., both being running red lights) exceeds a preset frequency threshold (e.g., 2 times) within a preset statistical period (e.g., the past year), it is ultimately determined that the target vehicle exhibits a habitual violation pattern for this type of violation. This pattern characterizes a repetitive and habitual tendency to commit specific violations by the vehicle or its driver.

[0108] As can be seen, this invention utilizes vehicle feature symbols to achieve high-precision, cross-time association of vehicle identity, overcoming the limitations of traditional license plate recognition which is susceptible to occlusion, damage, or forgery. By combining similarity thresholds with historical frequency analysis, it can reliably identify frequently violating vehicles that truly require focused attention, providing crucial historical behavioral evidence for differentiated processing.

[0109] S4.2: Make a joint judgment based on the matching result and the responsibility correlation score: if it is determined that there is a habitual violation pattern and the responsibility correlation score is low correlation, then generate a first type of warning information; if it is not determined that there is a habitual violation pattern and the responsibility correlation score is high correlation or intermediate correlation, then generate a second type of warning information.

[0110] The input to the warning generator includes two key dimensions: first, the result of the inertial violation pattern judgment from the historical database matching module (Boolean value: yes / no); second, the responsibility correlation score from S3.5 (discrete value: high, medium, low).

[0111] The first type of warning information includes at least the current violation type of the target vehicle, the frequency of historical violations matched, and the suggested violation level; the second type of warning information includes at least the current violation type of the target vehicle, the responsibility correlation score level, and the suggested review priority.

[0112] Specifically, the first type of warning is generated under the following conditions: if the habitual violation pattern is determined to be present and the responsibility correlation score is low, then the first type of warning information is triggered. This logic indicates that the vehicle has a history of similar violations, and this violation was mainly caused by its own voluntary intent, with little external influence, which falls under the warning situation that the system needs to focus on.

[0113] The second type of warning is generated under the following conditions: If the habitual violation pattern is determined to be negative and the responsibility correlation score is high or intermediate, then the second type of warning information is triggered. This logic indicates that the vehicle has no significant historical violations of the same type, and the current violation is significantly affected by the vehicle in front or involves a complex situation, which requires handling in conjunction with the context.

[0114] Furthermore, the first type of warning information should include at least the following fields: current violation type; historical violation frequency (specific number of times); and suggested violation level (e.g., minor, severe). This information directly points to the vehicle's historical violation records and suggests taking higher-level warning measures.

[0115] The second type of warning information should include at least the following fields: current violation type; responsibility correlation rating (high / medium); and recommended review priority (e.g., high, medium). This information emphasizes the specific circumstances of the current violation and suggests prioritizing it in the manual review process, and combining it with specific data from S3 (such as visibility obstruction ratio and following distance data) for violation analysis.

[0116] It should be noted that by jointly judging two orthogonal dimensions—historical behavior patterns and current situational responsibility—a more comprehensive assessment of traffic violations is achieved. Differentiated early warning information upgrades simple violation alerts into intelligent work orders with processing suggestions. This effectively guides limited resources to prioritize key violations, while also marking and diverting violations that may be affected by objective factors, thus improving the overall accuracy of traffic management.

[0117] Furthermore, such as Figure 3As shown, this embodiment also provides an intelligent traffic violation recognition and optimization system, including: a vehicle feature extraction module, used to acquire video streams from traffic monitoring equipment, extract continuous frame trajectory sequences of a target vehicle within a set time period, and generate vehicle feature symbols of the target vehicle based on the immutable features of the vehicle's appearance; a micro-violation chain construction module, used to analyze vehicle motion parameters based on the continuous frame trajectory sequences, construct a micro-violation behavior chain of the target vehicle, the micro-violation behavior chain being composed of an intent vector sequence, used to characterize continuous exploratory behaviors before the violation; a responsibility association assessment module, used to, when the violation behavior of the target vehicle is identified, backtrack the micro-violation behavior chain and the trajectory sequences of surrounding vehicles, calculate the relative behavior parameters between vehicles, assess whether the violation behavior is affected by the behavior of the preceding vehicle, and generate a responsibility association score; and a differential warning output module, used to match the vehicle feature symbols with a historical violation database, and output differentiated warning information in conjunction with the responsibility association score.

[0118] This embodiment also provides a computer device applicable to the intelligent traffic violation recognition optimization method, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the intelligent traffic violation recognition optimization method proposed in the above embodiment.

[0119] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0120] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the intelligent traffic violation recognition optimization method proposed in the above embodiments.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent traffic violation identification optimization method, characterized in that: include: The video stream is acquired from the traffic monitoring equipment, the continuous frame trajectory sequence of the target vehicle within a set time period is extracted, and the vehicle feature symbol of the target vehicle is generated based on the immutable features of the vehicle's appearance. Based on the analysis of the vehicle motion parameters according to the continuous frame trajectory sequence, a micro-violation behavior chain of the target vehicle is constructed. The micro-violation behavior chain consists of an intent vector sequence, which is used to characterize the continuous exploratory behaviors before the violation. When a traffic violation by the target vehicle is detected, the chain of minor violations and the trajectory sequence of surrounding vehicles are traced back, the relative behavioral parameters between the vehicles are calculated, the violation is assessed as to whether it is affected by the behavior of the vehicle in front, and a responsibility correlation score is generated. The vehicle identifier is matched with the historical violation database, and the differentiated warning information is output based on the responsibility correlation score. The extraction of the continuous frame trajectory sequence includes: a set time period consisting of a first preset time before the traffic light at the current intersection turns red and a second preset time period after the light turns red; continuous video frames in a specified area in front of the stop line within the set time period are extracted from the video stream; vehicle detection and tracking are performed on the continuous video frames to obtain the center point coordinates of the target vehicle in the image coordinate system, and the center point coordinates are connected in chronological order to form a continuous frame trajectory sequence. The vehicle feature identifier includes: locating key feature points of the target vehicle in the first frame of the continuous video frames; the key feature points include at least the center points of the two headlights and the four corner points of the air intake grille; calculating the relative positional relationship vector formed by the key feature points, and encoding the relative positional relationship vector as the vehicle feature identifier; The construction of the micro-violation behavior chain includes: calculating the velocity and acceleration components of the target vehicle in the direction perpendicular to the stop line based on the changes of the continuous frame trajectory sequence in the image coordinate system, as basic motion parameters; calculating the forward stride length and hesitation index in fixed time windows based on the basic motion parameters; the forward stride length is the positive cumulative displacement of the velocity component in a unit time; the hesitation index is the frequency of the sign change of the acceleration component within the same time window; comparing the forward stride length corresponding to each time window with a preset first displacement threshold and comparing the hesitation index with a preset frequency threshold within the set time period; if the forward stride length of a time window exceeds the first displacement threshold and the hesitation index simultaneously exceeds the frequency threshold, the vehicle movement is determined as a micro-violation segment; combining the forward stride length and hesitation index extracted from all the micro-violation segments arranged in chronological order into an intent vector sequence, and defining it as a micro-violation behavior chain.

2. The intelligent traffic violation recognition optimization method as described in claim 1, characterized in that: The velocity and acceleration components are calculated using a differential calibration algorithm based on image sequences.

3. The intelligent traffic violation recognition optimization method as described in claim 2, characterized in that: The parameters for calculating the relative behavior between vehicles include: Determine the time and location of the traffic violation of the target vehicle, extract the nearest preceding vehicle located directly in front of the target vehicle within a third preset time period before the time of the violation as the associated preceding vehicle, and obtain the continuous frame trajectory sequence of the associated preceding vehicle within the same time period as the preceding vehicle trajectory sequence. Extract the corresponding part of the target vehicle within the third preset time period from the chain of minor traffic violations. At the same time, calculate the visual distance occlusion ratio of the associated preceding vehicle to the driver's line of sight of the target vehicle based on the trajectory sequence of the preceding vehicle and the continuous frame trajectory sequence of the target vehicle.

4. The intelligent traffic violation recognition optimization method as described in claim 3, characterized in that: The line-of-sight occlusion ratio includes: In the road plane coordinate system, based on the trajectory of the preceding vehicle, the standard contour data of the vehicle model, and the trajectory of the target vehicle, the line-of-sight cone of the target vehicle is simulated from the perspective of the driver of the target vehicle observing the traffic lights or the road surface ahead; the occlusion length of the line-of-sight cone by the associated three-dimensional contour projection of the preceding vehicle is calculated, and the ratio of the occlusion length to the total length of the line-of-sight cone is calculated as the line-of-sight occlusion ratio.

5. The intelligent traffic violation recognition optimization method as described in claim 4, characterized in that: The assessment of whether the traffic violation was influenced by the behavior of the vehicle preceding it and the generation of a liability correlation score include: Based on the speed of the target vehicle at the time of the violation and the preset standard deceleration, the theoretical braking distance is calculated; Based on the actual distance between the target vehicle and the associated preceding vehicle at the time of the violation, determine whether the actual distance is less than the sum of the theoretical braking distance and the preset safety margin, and generate a first judgment result; Based on the combined state of the following three conditions, the logical judgment conditions are as follows: Condition A: Whether the line-of-sight occlusion ratio exceeds a preset occlusion threshold; Condition B: Within the third preset time period, is there at least one minor violation segment in the chain of minor traffic violations of the target vehicle? Condition C: Whether the result of the first judgment is yes; The responsibility correlation score includes three levels: high correlation, medium correlation, and low correlation.

6. The intelligent traffic violation recognition optimization method as described in claim 5, characterized in that: Logical combination judgments include: If the following conditions are met simultaneously, it is rated as a high degree of correlation: condition A is true and condition C is true. If the following conditions are met simultaneously, it is rated as low correlation: condition B is true and condition A is false and condition C is false. The remaining cases are assessed as intermediate correlation.

7. An intelligent traffic violation recognition and optimization system, based on the intelligent traffic violation recognition and optimization method according to any one of claims 1 to 6, characterized in that: Also includes: The vehicle feature extraction module is used to acquire video streams from traffic monitoring equipment, extract the continuous frame trajectory sequence of the target vehicle within a set time period, and generate the vehicle feature symbol of the target vehicle based on the immutable features of the vehicle's appearance. The micro-violation chain construction module is used to analyze vehicle motion parameters based on the continuous frame trajectory sequence and construct the micro-violation behavior chain of the target vehicle. The micro-violation behavior chain consists of an intent vector sequence, which is used to characterize continuous exploratory behaviors before the violation. The responsibility correlation assessment module is used to backtrack the chain of minor violations and the trajectory sequence of surrounding vehicles when the violation of the target vehicle is identified, calculate the relative behavior parameters between vehicles, assess whether the violation is affected by the behavior of the preceding vehicle, and generate a responsibility correlation score. The differential warning output module is used to match the vehicle feature with the historical violation database and output differential warning information in combination with the responsibility correlation score.

Images