Intelligent traffic illegal parking monitoring management method based on image recognition

Abstract

The invention discloses an intelligent traffic illegal parking monitoring management method based on image recognition, and relates to the technical field of intelligent traffic, and the method comprises the steps: obtaining a multi-view synchronous video stream of a target area; detecting and continuously tracking the vehicle in the multi-view synchronous video stream, generating a spatial-temporal trajectory of the vehicle, and determining a stationary state of the vehicle according to the spatial-temporal trajectory; carrying out three-dimensional geometric reconstruction on the static vehicle to obtain a spatial position of the vehicle in a world coordinate system, carrying out geometric relationship judgment on the basis of the spatial position and a preset no-parking area, and generating a geometric relationship judgment result; and detecting a legal intervention visual signal corresponding to the vehicle, and carrying out causal analysis on the illegal parking behavior in combination with a geometrical relationship judgment result. According to the method, through multi-view collaborative perception and three-dimensional space reconstruction, multi-modal semantic analysis and a causal association mechanism of time sequence visual signals are fused, so that the method has the capability of identifying legal intervention behaviors in a complex city scene, and the interpretability of illegal parking judgment is also enhanced.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to an intelligent traffic illegal parking monitoring and management method based on image recognition. Background Technology

[0002] With the increasing informatization and intelligence of urban traffic management, video analytics-based illegal parking monitoring has been gradually applied to road management and law enforcement evidence collection scenarios. Current technologies commonly employ a single-channel camera to acquire video footage, utilize target detection algorithms to identify vehicles, and then compare the detection bounding box with a pre-defined no-parking zone on the image plane to trigger subsequent alarms or capture processes. Some solutions only introduce simple rule or signal recognition mechanisms, but the overall processing flow still primarily relies on two-dimensional image information, with limited coordination between functional modules.

[0003] Existing methods for monitoring illegal parking generally suffer from the following limitations: First, they lack spatial positioning accuracy and are susceptible to interference from viewing angle and occlusion. Most existing methods rely on single-view images for two-dimensional coordinate determination, failing to consider perspective distortion and depth blur in camera imaging, which severely restricts system reliability. Second, they lack the ability to causally infer the legality of parking behavior. Most systems rely solely on the appearance of "whether the vehicle is stationary in a no-parking zone" for judgment, failing to identify and verify legitimate intervention visual signals. Even when some solutions attempt to introduce simple signal detection, it is often handled in isolation, without multi-dimensional collaborative verification with the vehicle's stationary time period, signal duration, confidence level, and spatial location, resulting in the inability to distinguish between "malicious illegal parking" and "legitimate temporary parking." Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent traffic illegal parking monitoring and management method based on image recognition, which solves the problems of insufficient spatial positioning accuracy and lack of causal reasoning ability for the legality of parking behavior.

[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 illegal parking monitoring and management method based on image recognition, comprising, Obtain multi-view synchronized video streams of the target area; Vehicles in multi-view synchronous video streams are detected and continuously tracked to generate their spatiotemporal trajectories, and the stationary state of the vehicles is determined based on these trajectories. Perform three-dimensional geometric reconstruction on stationary vehicles to obtain their spatial position in the world coordinate system. Determine the geometric relationship between the spatial position and the preset no-parking zone, and generate the result of the geometric relationship determination. The system detects the legal intervention visual signals corresponding to the vehicle, and combines the results of geometric relationship determination to conduct causal analysis of illegal parking behavior, outputs the illegal parking event determination result, and uploads the illegal parking event determination result information to the traffic management platform.

[0007] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the step of obtaining a multi-view synchronous video stream of the target area includes: Cameras are deployed in multiple directions in the target area, imaging parameters are uniformly configured and fixed, and video streams from various sources are connected to the processing platform and synchronized based on a unified time reference to form a time-consistent multi-view synchronous video stream.

[0008] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the step of detecting and continuously tracking vehicles in multi-view synchronous video streams to generate the spatiotemporal trajectory of the vehicles includes: Read synchronized frame groups in chronological order from a multi-view synchronized video stream; In a multi-view scenario, vehicle detection results are obtained through object detection, and the continuous vehicle trajectories within the view are updated based on trajectory prediction and data association, with the center of the bottom edge of the bounding box as the representative point of the trajectory. By integrating the spatial location and appearance features of a vehicle at the same moment from different perspectives, cross-perspective trajectory association is performed, a unified vehicle identifier is assigned, a multi-perspective observation set is formed, and multi-perspective observations at consecutive moments are aggregated based on the identifier to construct the complete spatiotemporal trajectory of the vehicle.

[0009] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the step of determining the stationary state of the vehicle based on the spatiotemporal trajectory includes: Based on the vehicle's spatiotemporal trajectory information, corresponding multi-view observation data are collected at synchronous moments to calculate the vehicle's representative position on the image plane. By analyzing the displacement changes of representative positions at consecutive moments frame by frame and evaluating the stability of the displacement changes within a preset time window, a stationary determination mark for the vehicle is generated. Based on the continuity of the static determination markers in the time dimension, the moments that continuously meet the static condition are merged into static time periods, and the static time periods are associated with the corresponding vehicle identifiers to form a set of static vehicles.

[0010] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the step of performing three-dimensional geometric reconstruction of stationary vehicles to obtain the spatial position of the vehicles in the world coordinate system includes: Read the corresponding stationary vehicle and stationary time period from the set of stationary vehicles based on the vehicle identifier, and select at least one synchronization moment from the time period as the 3D reconstruction moment. Based on multi-view synchronous video, the two-dimensional observation points of the vehicle are obtained from the perspectives of each camera device. Combined with the calibration parameters of the camera devices, the two-dimensional observation points are mapped to a unified world coordinate system with the ground of the target area as the reference, and the three-dimensional position that satisfies the multi-view consistency constraint is solved. If the static time period contains multiple synchronization moments, the reconstruction is repeated at M moments to obtain a three-dimensional position sequence, calculate the fused three-dimensional spatial position, and determine the final spatial position of the vehicle when the position stability meets the preset conditions.

[0011] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the step of determining the geometric relationship between spatial location and preset no-parking zones, and generating the result of the geometric relationship determination, includes: Obtain the boundary vertex information of the preset no-parking zones from the traffic management database and construct a polygonal geometric model; The stable three-dimensional spatial position of the vehicle is projected onto the ground coordinate system to obtain the corresponding two-dimensional projection points. Geometric relationships are then determined based on the spatial relationship between the projection points and the polygonal model of the no-stopping area. When the projection point is inside the polygon, the vehicle is determined to be in the no-parking area. When the projection point is outside the polygon and the distance from the boundary is greater than a preset threshold, the vehicle is determined to be outside the no-parking area. Otherwise, the vehicle is determined to be in the boundary critical state of the no-parking area. The geometric relationship determination results are correlated with the vehicle identification and the corresponding stationary time period.

[0012] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the legal intervention visual signal corresponding to the detected vehicle includes: Based on the vehicle's spatiotemporal trajectory, within the time window corresponding to the vehicle's stop, candidate regions of visual signals associated with the vehicle are dynamically extracted from multi-view videos, and multimodal recognition is performed: The periodic flashing characteristics of the vehicle's rear hazard lights are detected by brightness timing analysis to determine the validity of the hazard light signal. Use a pre-trained convolutional neural network model to identify police car and ambulance vehicle markings; Combine OCR technology with template matching methods to identify text or icon information such as "taxi" and "accepting orders"; Standard fault warning devices are detected using color and morphological characteristics; Recognition results that meet the preset minimum duration condition are determined as valid and legitimate intervention events, and consecutive events of the same type are merged to generate a set of legitimate intervention visual signals corresponding to the vehicle.

[0013] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the causal analysis of illegal parking behavior based on the results of geometric relationship determination includes: Acquire the vehicle's stationary time period, geometric relationship determination results, and set of legal intervention visual signals; When a vehicle is determined to be in a no-parking zone, if there is a valid and legal intervention event that overlaps with the stationary period, it is determined to be a legal temporary stop; otherwise, it is determined to be illegal parking. When a vehicle is determined to be in a critical state, if there are valid and legitimate intervention events with overlapping time, it is determined to be a legitimate temporary stop; otherwise, it is determined to require manual review.

[0014] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the output of the illegal parking event determination result includes: Based on the causal analysis results of illegal parking, the system associates vehicle identification, stationary time period, geometric relationship determination results, whether there are legal intervention events overlapping with the stationary time period, and parking behavior determination conclusions to construct structured illegal parking event determination result data. Based on the determination conclusions, it generates illegal parking event records, legal parking records, or records pending review.

[0015] As a preferred embodiment of the intelligent traffic illegal parking monitoring and management method based on image recognition described in this invention, the step of uploading the illegal parking event determination result information to the traffic management platform includes: Based on the results of illegal parking incident determination, the data is packaged into a unified data packet in a standard format and uploaded to the traffic management platform through a communication interface. After receiving the data, the platform assigns a unique event identifier, writes it into the storage unit, and marks the status according to the determination conclusion, thereby realizing classified storage and management.

[0016] The beneficial effects of this invention are as follows: This invention constructs a vehicle spatiotemporal modeling and illegal parking determination process based on multi-view synchronous video streams. By fusing observation information from multiple camera devices under a unified time reference, it reduces the positioning instability caused by perspective distortion, occlusion, and scale changes under single-view imaging conditions. Compared to existing illegal parking monitoring methods that mainly rely on two-dimensional image coordinates and simple region overlap judgments, this invention introduces three-dimensional geometric reconstruction based on vehicle stationary time periods, allowing the vehicle's spatial position to be described in a unified world coordinate system, thereby improving the accuracy and stability of geometric relationship determination in no-parking areas. Simultaneously, this invention uses the vehicle's spatiotemporal trajectory as its core, limiting the detection of legal intervention visual signals to a clearly defined parking time window, and filters signal validity through multimodal recognition and duration constraints, reducing the interference of isolated frames or instantaneous features on the determination results. Based on this, it performs joint causal analysis of vehicle spatial position relationships and legal intervention events, enabling illegal parking determination to no longer rely solely on the superficial condition of stationary status, but to distinguish between illegal parking, legal temporary parking, and situations requiring manual review, improving the consistency and interpretability of the determination results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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 This is a flowchart of an intelligent traffic illegal parking monitoring and management method based on image recognition in Example 1.

[0019] Figure 2 This is a flowchart of the legal intervention visual signal multimodal recognition and fusion process in Example 1.

[0020] Figure 3 This is a flowchart of the causal analysis and event output for illegal parking behavior in Example 1. 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] Example 1, referring to Figures 1-3 This is the first embodiment of the present invention, which provides an intelligent traffic illegal parking monitoring and management method based on image recognition, including the following steps: S1. Obtain multi-view synchronized video streams of the target area; Specifically, obtaining multi-view synchronized video streams of the target area includes: Multiple cameras are set up at different locations in the target area to capture real-time images from each camera. The imaging parameters of each camera device are uniformly configured and fixed. The video streams output by each camera device are connected to the video processing platform. Based on a unified time reference, the multiple video streams are time-aligned to form a multi-view synchronous video stream arranged in chronological order.

[0025] This invention deploys camera equipment in multiple directions within the target area and configures the imaging parameters uniformly. By combining a unified time reference with time alignment of multiple video streams, a synchronous multi-view video stream with consistent timing is formed. This makes the vehicle observation results from different perspectives comparable in the time dimension, providing a reliable data foundation for subsequent cross-view trajectory association, spatial location reconstruction, and behavior analysis. This reduces the impact of temporal deviations between multiple video sources on the analysis results and improves the stability and consistency of the overall monitoring and processing flow.

[0026] S2. Detect and continuously track vehicles in multi-view synchronous video streams, generate spatiotemporal trajectories of vehicles, and determine the stationary state of vehicles based on spatiotemporal trajectories. Specifically, the detection and continuous tracking of vehicles in multi-view synchronized video streams, generating the spatiotemporal trajectories of the vehicles, includes: Read synchronous video frame groups in chronological order from multi-view synchronous video streams. :

[0027] in, The k-th synchronization moment under a unified time base. Refers to the camera's field of view. Refers to the first Each camera device is constantly The video frames, where N refers to the number of camera devices; For synchronization time , for the Video frames corresponding to each camera device use The object detection algorithm performs vehicle object detection and obtains the corresponding vehicle detection result set. :

[0028] in, Refers to the first Perspective The total number of vehicles detected at any given time; The YOLOv5 object detection algorithm is a single-stage real-time object detection algorithm based on a deep convolutional neural network. It can predict the category and bounding box of multiple objects in an image simultaneously through a single forward inference. It features high accuracy, high speed and easy deployment, and is widely used in intelligent transportation scenarios such as vehicle detection. Each test result Defined as:

[0029] in, Refers to the first Perspective 1 The bounding box coordinates of each vehicle. Confidence level Refers to vehicle category, Refers to the appearance feature vector; From the perspective of each camera device, maintain the corresponding set of vehicle trajectories, and based on the detection results at the current moment. Compared to the previous moment The trajectory status is updated to update the vehicle trajectory within the viewpoint, forming a continuous trajectory within the viewpoint; For each existing trajectory, a state prediction is performed, and the representative point of each trajectory on the image plane is defined as the center point of the bottom edge of the bounding box. :

[0030] in, Indicates the vehicle is in Spatial location characteristics from the perspective of a camera device; After completing the trajectory update within the field of view, the spatial position characteristics of the vehicle under different viewpoints are used as a basis. Appearance features Cross-view correlation is performed on trajectories from different camera devices, and a unified vehicle identifier is assigned to the multi-view trajectories corresponding to the same physical vehicle. ; Under the constraint of unified vehicle identification, the vehicles converge at the same time. The multi-view observation information below forms a multi-view observation set of the vehicle. :

[0031] in, Indicates that the trajectory belongs to ; Based on unified vehicle identification By converging cross-perspective observations of the vehicle at continuous moments, the spatiotemporal trajectory of the vehicle is generated. :

[0032] in, Refers to time index. This refers to the number of time points included in the vehicle's trajectory.

[0033] This invention introduces multi-view synchronous video streams to collaboratively process vehicle observation information from different camera devices under a unified time reference, achieving continuous cross-view construction of vehicle spatiotemporal trajectories. By maintaining vehicle trajectories separately in each viewpoint and fusing spatial location and appearance features of the vehicle at synchronous moments to correlate trajectories across camera devices, a unified vehicle identifier is assigned to the same physical vehicle. This effectively avoids trajectory interruptions caused by occlusion, viewpoint changes, or short-term target loss under single-view conditions. Furthermore, by aggregating vehicle information using multi-view observation sets as basic units, a spatiotemporal trajectory structure that evolves over time is formed. This not only improves the consistency and stability of vehicle tracking results in both time and space dimensions but also provides a reliable data foundation for subsequent vehicle stationary state determination, 3D geometric reconstruction, and illegal parking behavior analysis. Compared to existing technologies that rely solely on two-dimensional image coordinates or single-frame detection results, this solution can more comprehensively depict the continuous motion state of vehicles in real-world road scenarios, thereby enhancing the adaptability and reliability of illegal parking monitoring systems in complex traffic environments.

[0034] Furthermore, determining the stationary state of a vehicle based on its spatiotemporal trajectory includes: For vehicles spacetime trajectory According to time index Process sequentially, utilizing the multi-view observation set at each synchronization moment. Calculate the representative positions of the vehicle; For each moment ,from Select representative points from all visible viewpoints. And calculate the representative position of the vehicle on the image plane. :

[0035] in, Used only for determining static states, not as input for 3D reconstruction. Refers to the moment Vehicles can be observed A collection of camera device indexes; For vehicles Position sequence Perform frame-by-frame processing, from Begin by calculating the displacement change between adjacent time points. :

[0036] in, Refers to the vehicle at the previous moment The three-dimensional spatial position, Refers to spatial distance measurement; In continuous Within a time window comprised of several time points, the change in vehicle displacement... The fluctuations were analyzed to obtain the motion stability characteristics. :

[0037] in, In a time series, starting from the first... The amount of displacement change starting at a certain moment. Refers to the length of the time window. Variance calculation; At every moment A static determination mark is generated based on the displacement change and motion stability characteristics. :

[0038] in, The displacement change threshold, ranging from 0.05 meters to 0.2 meters, is determined using existing technologies such as statistical analysis of the measured error distribution of vehicle positioning systems, analysis of historical stationary vehicle trajectory data, or empirical calibration experiments in traffic scenarios. The motion stability threshold, ranging from 0.001 to 0.01 square meters, is determined using existing technologies such as statistical analysis of measured error distribution of vehicle positioning systems, analysis of historical stationary vehicle trajectory data, or empirical calibration experiments in traffic scenarios. Refers to the logical AND operation; Based on the static determination marker sequence , will satisfy The continuous moments are merged into a continuous static time period. The vehicle identifier is then associated with a continuous period of stationary time to form a set of stationary vehicles. .

[0039] By introducing a joint judgment mechanism of displacement change and motion stability based on multi-view spatiotemporal trajectory, the system can distinguish between short-term vehicle vibration, detection noise and real motion behavior, and achieve robust recognition of vehicle stationary state. The continuous judgment method based on time window can effectively avoid single-frame misjudgment, improve the stability and continuity of static time period extraction, and provide reliable static vehicle screening results for subsequent 3D reconstruction and illegal parking judgment.

[0040] S3. Perform three-dimensional geometric reconstruction on the stationary vehicle to obtain the vehicle's spatial position in the world coordinate system. Based on the spatial position and the preset no-parking area, determine the geometric relationship and generate the result of the geometric relationship determination. Specifically, performing three-dimensional geometric reconstruction on a stationary vehicle to obtain its spatial position in the world coordinate system includes: From the output set of stationary vehicles In the middle, according to vehicle identification Read the vehicles that are stationary one by one and the corresponding stationary time periods. and from the static time period Select at least one synchronization time As a moment in 3D reconstruction; At the same time Extracting vehicles from multi-view synchronized video streams A multi-view observation set is formed from the two-dimensional observation points under each visible camera device. :

[0041] in, The vehicle is in the Representative points on the image plane from each perspective; Establish a unified world coordinate system with the ground of the target area as the reference plane. and acquire each camera device The calibration parameters, including the intrinsic parameter matrix. and extrinsic parameter matrix ,in, Refers to the first The intrinsic parameter matrix of the camera equipment includes focal length and principal point parameters. For rotation matrix, It is a translation vector; Two-dimensional observation points Convert to homogeneous pixel coordinates And solve for the three-dimensional position of the vehicle in the world coordinate system. Satisfying the multi-view projection consistency constraint:

[0042] in, Refers to the vehicle At any moment 3D world coordinates Rotate to the In the coordinate system of each camera, Scale equivalence in homogeneous coordinates Set Each camera device in the system is numbered ; If the static time period contains multiple synchronous moments, At that moment Repeat the reconstruction to obtain a three-dimensional position sequence. And calculate the fused three-dimensional spatial position :

[0043] in, Vehicles In the The single 3D position estimate obtained by reconstructing from a number of synchronization moments, where M refers to the number of valid synchronization moments selected during the vehicle's stationary period. The normalization factor is used to calculate the arithmetic mean. Calculate position stability metrics :

[0044] in, Euclidean distance; when At that time, confirm For the final stable spatial position of the vehicle, among which, The preset position stability threshold is typically between 0.1 meters and 0.3 meters, and can be determined using existing technologies such as positioning accuracy calibration of multi-view 3D reconstruction systems, statistical analysis of position fluctuations of stationary vehicles in real-world scenarios, or engineering experience in the field of traffic monitoring.

[0045] This invention performs 3D geometric reconstruction on vehicles already determined to be stationary. By selecting one or more synchronous moments within a stationary time period, and utilizing 2D observation points from multi-view synchronous video combined with the camera's intrinsic and extrinsic parameter calibration information, the 3D spatial position of the vehicle is solved in a unified world coordinate system. This effectively reduces the impact of vehicle motion, viewpoint changes, and detection jitter on the spatial positioning results. Furthermore, by repeatedly performing reconstruction at multiple synchronous moments within the stationary time period and fusing and measuring the stability of the resulting 3D position sequences, the final spatial position is confirmed only when the position fluctuation meets a preset stability condition, thus avoiding interference from occasional observation errors. Compared to existing schemes that directly use single reconstruction or 2D position approximation, this invention significantly improves spatial positioning accuracy and result reliability, providing a more robust spatial foundation for subsequent geometric determination of no-parking zones and analysis of illegal parking behavior.

[0046] Furthermore, the geometric relationship is determined based on the spatial location and the preset no-parking zone, and the results of the geometric relationship determination are generated, including: Geometric information of preset no-parking zones is obtained from the traffic management database. The no-parking zones are represented by a set of ordered two-dimensional boundary vertices and have been converted to the same ground coordinate system used for vehicle three-dimensional reconstruction. Constructing a polygonal geometric model of the no-stopping area based on the boundary vertex sequence. :

[0047] in, This refers to the total number of boundary vertices. Refers to the polygon constructor function. Refers to the two-dimensional boundary vertex; The vehicle's stable three-dimensional spatial position Projecting onto the ground coordinate system yields the projected position:

[0048] in, The projection point of the vehicle on the ground coordinate system. The horizontal and vertical coordinates of the vehicle in the ground coordinate system; Based on projection point Polygonal model of no-parking zone Perform geometric relationship determination: when When located inside a polygon, let the geometric relationship determination result... ; when Located outside the polygon and with a minimum distance from the boundary greater than a preset threshold. season ; when Located outside the polygon and with a minimum distance from the boundary less than a preset threshold. ,make ; Determine the geometric relationship results With vehicle identification and static time period Related; Among them, threshold The range of values ​​is usually 100. The accuracy of road markings, vehicle positioning errors, and actual law enforcement tolerance can be determined using existing technologies such as statistical analysis of historical illegal parking cases, positioning system error calibration, or traffic management department regulations. This refers to a vehicle's projected location being outside the no-parking zone but not close to the boundary. This refers to the vehicle's projected location being within a no-parking zone. This refers to a vehicle's projected location being outside the no-parking zone and close to its boundary.

[0049] This invention projects the stable three-dimensional spatial position of a vehicle onto a ground coordinate system and accurately determines its spatial relationship with the polygonal geometric model of a pre-defined no-parking zone. This effectively avoids misjudgments caused by inconsistent scales in single-view images. By introducing a boundary distance threshold to distinguish the adjacent states of the no-parking zone, it achieves refined discrimination of clearly illegal parking, clearly non-illegal parking, and boundary-critical situations. This improves the spatial accuracy of no-parking determination and the tolerance for law enforcement errors, and provides a reliable geometric basis for subsequent causal analysis.

[0050] S4. Detect the legal intervention visual signals corresponding to the vehicle, and perform causal analysis of the illegal parking behavior based on the results of geometric relationship determination, output the illegal parking event determination result, and upload the illegal parking event determination result information to the traffic management platform; Specifically, the legally permissible visual intervention signals for detecting vehicles include: Based on the vehicle's spatiotemporal trajectory, the vehicle In the trajectory time window Inside, detect legitimate intervention visual signals related to vehicle parking behavior; Within the time window, based on real-time updates of vehicle detection bounding boxes, the YOLO object detection algorithm is used to dynamically extract objects related to vehicles from multi-view synchronous video streams. The associated visual signal candidate areas (ROIs) include the emergency hazard lights area at the rear of the vehicle, the police and special vehicle markings area on the vehicle body, the temporary passenger pick-up and drop-off indicators on the roof and side windows, and the fault warning triangle area that may appear on the ground behind the vehicle. Each ROI is updated in real time with the vehicle detection frame to keep track. The YOLO object detection algorithm is a deep learning object detection algorithm that combines single-pass inference, full-image detection, and speed and accuracy. It uses a single-pass convolutional neural network to forward propagate and simultaneously output the position, category, and confidence of all objects in the image. It has high real-time performance and detection accuracy and is suitable for the recognition of signs, lights, and warning objects in traffic scenarios. For each moment Perform multimodal recognition on each ROI: The system uses a time-series brightness variation analysis algorithm to detect the periodic flashing characteristics of the vehicle's rear hazard lights and determine whether the signal is valid, including: Based on the vehicle's spatiotemporal trajectory, the region of interest (ROI) of the vehicle's rear hazard lights is located in consecutive video frames. For every moment The ROI is determined, the average luminance intensity is calculated, and a luminance time series is generated. ; The luminance sequence is subjected to moving average filtering to suppress transient noise, resulting in a smoothed luminance sequence. ; Set brightness threshold Binarize the smooth sequence: when When the light is on, mark it as "on" (1); otherwise mark it as "off" (0). Merge consecutive "on" states and extract the time period during which the light is on. ; Among them, the brightness threshold The value range is usually 50% to 80% of the maximum possible brightness of the ROI area, and can be determined by existing technologies such as adaptive thresholding, offline calibration experiments or statistical analysis based on historical video data. Calculate the center time interval between adjacent lit intervals If there are at least 3 consecutive lit intervals, and each interval If all the flashes fall within the period range (0.5 seconds to 2.0 seconds) corresponding to the standard double flash frequency, then the product is determined to have periodic flashing characteristics. If both periodic flickering and total duration are satisfied... If so, the signal is determined to be a valid hazard light signal; Using a pre-trained convolutional neural network classification model, the system identifies the special markings for police cars, ambulances, and fire trucks on vehicles. The convolutional neural network classification model includes: an input layer, a fusion layer, and an output layer; The input layer is a 224×224×3 RGB image. The fusion layer consists of: three convolutional layers (each with a 3×3×3 kernel, numbering 64, 128, and 256 respectively, using ReLU activation, Same padding, and stride of 1), two max pooling layers (window size 2×2×2, stride of 2), and two fully connected layers (with 512 and 256 neurons respectively, both using ReLU activation and 0.5 Dropout regularization). The output layer consists of three neurons, corresponding to the categories "police car", "ambulance", and "fire truck", and outputs the predicted probability of each category through the Softmax activation function. Model training steps: Using an offline-built special vehicle identification image dataset, the training samples are standardized and scaled to a uniform size of 224×224×3 before being input into the convolutional neural network model. The multi-layer features are integrated through a fusion layer, and the output layer generates the predicted probabilities for each category. The error between the prediction and the true label is calculated using the cross-entropy loss function. The optimizer iteratively updates the weights of each layer of the network based on the loss gradient. The training process iterates multiple times on the training set until the classification accuracy of the model on the validation set converges. The text detection algorithm based on OCR technology and the icon recognition algorithm based on template matching are used to identify the "Taxi" and "Taxi Accepting Orders" passenger pick-up and drop-off signs displayed on the vehicle body; Based on morphological operations and color feature matching algorithms, the system detects whether a standard warning triangle sign appears on the ground and identifies its shape and color features. All recognition results are recorded in a structured data format, which includes signal type, recognition confidence level, timestamp, corresponding camera device identifier, and viewpoint information. Will meet the requirements of the vehicle Signals within the ROI and lasting for a duration not less than the preset minimum effective duration (e.g., 2 seconds) are considered valid and legitimate intervention events. Continuous valid and legitimate intervention event segments of the same type of signal are time-merged to form a legitimate intervention event time segment. and generate vehicles The corresponding set of legal intervention visual signals :

[0051] in, This is a legitimate type of intervention signal. A predefined set of legal intervention types.

[0052] This invention focuses on the spatiotemporal trajectory of vehicles, performing unified modeling and collaborative verification of legal intervention visual signals within a defined stopping time window. Through a multi-view candidate region extraction mechanism that dynamically updates with the vehicle detection frame, it effectively ensures that various signals maintain spatial consistency with the target vehicle, avoiding background interference or false associations. Furthermore, for different types of legal intervention signals, it introduces various recognition methods, including brightness temporal analysis, convolutional neural network classification, OCR and template matching, and morphological and color feature detection. Constraints such as duration, periodicity, and temporal overlap are used to filter and merge the recognition results, thereby avoiding the impact of single-frame false detections or instantaneous noise on the judgment results. Compared to existing schemes that only perform simple signal detection or static sign recognition, this invention correlates legal intervention signals with vehicle stopping behavior at both the temporal and spatial levels, providing a more discriminative evidentiary basis for subsequent causal analysis of illegal parking.

[0053] Furthermore, the causal analysis of illegal parking behavior based on the results of geometric relationship determination includes: Obtain the vehicle's stationary time period Geometric relationship determination results and the set of legally intervened visual signals : when At that time, determine the period during which the vehicle is stationary. Does the entity contain at least one legitimate intervention event? And it meets the condition of overlapping time periods:

[0054] If it exists, it is determined to be a legal temporary stop; if it does not exist, it is determined to be an illegal stop. when At that time, determine the period during which the vehicle is stationary. Are there any legitimate intervention events within the time frame that meet the condition of overlapping time periods? If it exists, it is determined to be a legal temporary stop; if it does not exist, the vehicle stop is determined to require manual review.

[0055] This invention achieves refined causal determination of illegal parking behavior by performing spatiotemporal correlation analysis of vehicle stationary time periods, geometric relationship determination results, and legal intervention visual signals. When a vehicle is located in a no-parking zone or sensitive area, the system automatically checks whether there is a legal intervention event during the stationary period and distinguishes between legal temporary parking, illegal parking, and situations requiring manual review based on time overlap conditions. This method effectively avoids misjudgments caused by relying solely on spatial location, significantly improves the accuracy of illegal parking identification and the rationality of law enforcement, while reducing the burden of manual review and enhancing the credibility and explainability of the intelligent traffic management system.

[0056] Furthermore, the system outputs the results of illegal parking incident determinations and uploads this information to the traffic management platform, including: Based on the causal analysis results of illegal parking, obtain vehicle identification numbers. static time period Geometric relationship determination results And the conclusion of the judgment on vehicle parking behavior; The determination results include illegal parking, legal temporary parking, and cases requiring manual review. And determine whether there are any legitimate intervention events that overlap with the static time period. :

[0057] in, This is a boolean value indicating whether overlapping events exist. Vehicle identification static time period Geometric relationship determination results Are there any legitimate intervention events that overlap with the static time period? The results of illegal parking incident determinations are linked to the conclusions of parking behavior determinations to construct structured data. The event status is marked based on the judgment conclusion, and the corresponding event record is generated: When a parking violation is determined to be "illegal parking", a parking violation event record is generated. When determined to be a "legal temporary stop", a legal stop event record is generated; When a record is determined to be "requires manual review", a record of events pending review is generated. Based on the results of the illegal parking incident determination, the data information is encapsulated according to a standard format to form a data packet with a unified structure; The standard format refers to a data format that uniformly organizes and encodes various information fields in the judgment result of illegal parking incidents according to a pre-agreed data structure specification; The packaged illegal parking incident judgment results are uploaded to the traffic management platform via the communication interface; The traffic management platform receives the judgment results of illegal parking events, writes the corresponding events into the storage unit, assigns a unique event identifier to each event record, and marks the event status according to the parking behavior judgment conclusion, thereby storing and managing illegal parking events.

[0058] This invention, based on causal analysis of illegal parking, constructs structured event data including vehicle identification, stationary period, geometric relationship, legal intervention event markers, and judgment conclusions. This data is then packaged in a unified format and uploaded to the traffic management platform in real time. The platform automatically receives, stores, and assigns a unique event identifier to each record, classifying and managing it according to the judgment conclusions. This mechanism standardizes, automates, and makes traceable the illegal parking handling process, ensuring not only the integrity and reliability of enforcement data but also supporting efficient review, statistical analysis, and system linkage, significantly improving the intelligence level and business collaboration efficiency of urban traffic management.

[0059] 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. A method for intelligent traffic violation monitoring and management based on image recognition, characterized in that: include, Obtain multi-view synchronized video streams of the target area; Vehicles in multi-view synchronous video streams are detected and continuously tracked to generate their spatiotemporal trajectories, and the stationary state of the vehicles is determined based on these trajectories. Perform three-dimensional geometric reconstruction on stationary vehicles to obtain their spatial position in the world coordinate system. Determine the geometric relationship between the spatial position and the preset no-parking zone, and generate the result of the geometric relationship determination. The system detects the legal intervention visual signals corresponding to the vehicle, and combines the results of geometric relationship determination to conduct causal analysis of illegal parking behavior, outputs the illegal parking event determination result, and uploads the illegal parking event determination result information to the traffic management platform.

2. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 1, characterized in that: The method of obtaining the multi-view synchronized video stream of the target area includes: Cameras are deployed in multiple directions in the target area, imaging parameters are uniformly configured and fixed, and video streams from various sources are connected to the processing platform and synchronized based on a unified time reference to form a time-consistent multi-view synchronous video stream.

3. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 2, characterized in that: The process of detecting and continuously tracking vehicles in a multi-view synchronized video stream to generate the vehicle's spatiotemporal trajectory includes: Read synchronized frame groups in chronological order from a multi-view synchronized video stream; In a multi-view scenario, vehicle detection results are obtained through object detection, and the continuous vehicle trajectories within the view are updated based on trajectory prediction and data association, with the center of the bottom edge of the bounding box as the representative point of the trajectory. By integrating the spatial location and appearance features of a vehicle at the same moment from different perspectives, cross-perspective trajectory association is performed, a unified vehicle identifier is assigned, a multi-perspective observation set is formed, and multi-perspective observations at consecutive moments are aggregated based on the identifier to construct the complete spatiotemporal trajectory of the vehicle.

4. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 3, characterized in that: Determining the stationary state of a vehicle based on its spatiotemporal trajectory includes: Based on the vehicle's spatiotemporal trajectory information, corresponding multi-view observation data are collected at synchronous moments to calculate the vehicle's representative position on the image plane. By analyzing the displacement changes of representative positions at consecutive moments frame by frame and evaluating the stability of the displacement changes within a preset time window, a stationary determination mark for the vehicle is generated. Based on the continuity of the static determination markers in the time dimension, the moments that continuously meet the static condition are merged into static time periods, and the static time periods are associated with the corresponding vehicle identifiers to form a set of static vehicles.

5. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 4, characterized in that: The process of performing three-dimensional geometric reconstruction on a stationary vehicle to obtain its spatial position in the world coordinate system includes: Read the corresponding stationary vehicle and stationary time period from the set of stationary vehicles based on the vehicle identifier, and select at least one synchronization moment from the time period as the 3D reconstruction moment. Based on multi-view synchronous video, the two-dimensional observation points of the vehicle are obtained from the perspectives of each camera device. Combined with the calibration parameters of the camera devices, the two-dimensional observation points are mapped to a unified world coordinate system with the ground of the target area as the reference, and the three-dimensional position that satisfies the multi-view consistency constraint is solved. If the static time period contains multiple synchronization moments, the reconstruction is repeated at M moments to obtain a three-dimensional position sequence, calculate the fused three-dimensional spatial position, and determine the final spatial position of the vehicle when the position stability meets the preset conditions.

6. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 5, characterized in that: The geometric relationship determination based on spatial location and preset no-parking zones, and the generation of the geometric relationship determination results, include: Obtain the boundary vertex information of the preset no-parking zones from the traffic management database and construct a polygonal geometric model; The stable three-dimensional spatial position of the vehicle is projected onto the ground coordinate system to obtain the corresponding two-dimensional projection points. Geometric relationships are then determined based on the spatial relationship between the projection points and the polygonal model of the no-stopping area. When the projection point is inside the polygon, the vehicle is determined to be in the no-parking area. When the projection point is outside the polygon and the distance from the boundary is greater than a preset threshold, the vehicle is determined to be outside the no-parking area. Otherwise, the vehicle is determined to be in the boundary critical state of the no-parking area. The geometric relationship determination results are correlated with the vehicle identification and the corresponding stationary time period.

7. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 6, characterized in that: The legitimate intervention visual signals corresponding to the detection vehicle include: Based on the vehicle's spatiotemporal trajectory, within the time window corresponding to the vehicle's stop, candidate regions of visual signals associated with the vehicle are dynamically extracted from multi-view videos, and multimodal recognition is performed: The periodic flashing characteristics of the vehicle's rear hazard lights are detected by brightness timing analysis to determine the validity of the hazard light signal. Use a pre-trained convolutional neural network model to identify police car and ambulance vehicle markings; Combine OCR technology with template matching methods to identify the text or icon information of "taxi" and "accepting orders"; Standard fault warning devices are detected using color and morphological characteristics; Recognition results that meet the preset minimum duration condition are determined as valid and legitimate intervention events, and consecutive events of the same type are merged to generate a set of legitimate intervention visual signals corresponding to the vehicle.

8. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 7, characterized in that: The causal analysis of illegal parking behavior based on the results of geometric relationship determination includes: Acquire the vehicle's stationary time period, geometric relationship determination results, and set of legal intervention visual signals; When a vehicle is determined to be in a no-parking zone, if there is a valid and legal intervention event that overlaps with the stationary period, it is determined to be a legal temporary stop; otherwise, it is determined to be illegal parking. When a vehicle is determined to be in a critical state, if there are valid and legitimate intervention events with overlapping time, it is determined to be a legitimate temporary stop; otherwise, it is determined to require manual review.

9. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 8, characterized in that: The output of the illegal parking event determination result includes: Based on the causal analysis results of illegal parking, the system associates vehicle identification, stationary time period, geometric relationship determination results, whether there are legal intervention events overlapping with the stationary time period, and parking behavior determination conclusions to construct structured illegal parking event determination result data. Based on the determination conclusions, it generates illegal parking event records, legal parking records, or records pending review.

10. The intelligent traffic illegal parking monitoring and management method based on image recognition as described in claim 9, characterized in that: The process of uploading the results of illegal parking incident determinations to the traffic management platform includes: Based on the results of illegal parking incident determination, the data is packaged into a unified data packet in a standard format and uploaded to the traffic management platform through a communication interface. After receiving the data, the platform assigns a unique event identifier, writes it into the storage unit, and marks the status according to the determination conclusion, thereby realizing classified storage and management.

Images