Vehicle violation identification method and system based on three-dimensional pose perception

Abstract

The application relates to the technical field of artificial intelligence, and provides a vehicle violation identification method and system based on three-dimensional posture perception. The method comprises the following steps: obtaining a vehicle violation detection point through violation site positioning; pre-disposing equipment at the detection site to obtain a high-position image collection array; setting a trigger distance to determine a trigger range; when a target vehicle enters the range, activating the image collection array to obtain a real-time image set; realizing three-dimensional posture perception based on the image set to obtain real-time three-dimensional posture parameters; taking the trigger range as a constraint to obtain local road geometric parameters; and combining the posture parameters and the road geometric parameters to identify the violation to obtain a vehicle violation identification result. The application realizes three-dimensional posture perception on the real-time images collected by the high-position image collection array, solves the technical problem that a traditional monocular camera cannot construct a three-dimensional scene, and ensures high-precision identification of the vehicle position.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, specifically to the technical field of deep learning and computer vision, and particularly to a vehicle violation identification method and system based on three-dimensional pose perception. BACKGROUND

[0002] With the continuous growth of urban traffic and the complexity of road use, the monitoring and identification of vehicle violations have become increasingly important. Traditional vehicle violation identification methods have low identification accuracy and poor real-time performance, making it difficult to meet the growing demand for traffic management.

[0003] In this context, a vehicle violation identification method and system based on three-dimensional pose perception has emerged, aiming to solve the technical problem of low-precision vehicle location identification caused by the inability of traditional monocular cameras to construct three-dimensional scenes by performing three-dimensional pose perception on real-time images collected by a high-position image collection array. SUMMARY

[0004] The present application provides a vehicle violation identification method and system based on three-dimensional pose perception. The method aims to solve the technical problem of low-precision vehicle location identification caused by the inability of traditional monocular cameras to construct three-dimensional scenes by performing three-dimensional pose perception on real-time images collected by a high-position image collection array.

[0005] In view of the above problems, the present application provides a vehicle violation identification method and system based on three-dimensional pose perception.

[0006] The first aspect of the present application provides a vehicle violation identification method based on three-dimensional pose perception, which comprises: locating a target road at a violation site to obtain a vehicle violation detection point; pre-disposing a violation detection device at the vehicle violation detection point to obtain a high-position image collection array; presetting a trigger distance and determining a preset trigger range based on the preset trigger distance and the vehicle violation detection point; activating the high-position image collection array to collect images of a target vehicle when the target vehicle enters the preset trigger range to obtain a real-time image set; performing three-dimensional pose perception on the target vehicle based on the real-time image set to obtain real-time three-dimensional pose parameters; taking the preset trigger range as an interaction constraint to obtain local road geometric parameters; and performing violation identification based on the real-time three-dimensional pose parameters and the local road geometric parameters to obtain a vehicle violation identification result.

[0007] In another aspect of the present disclosure, a vehicle violation identification system based on three-dimensional pose perception is provided, which comprises: a site positioning module configured to position a target road for a violation site, and obtain a vehicle violation detection point; a device pre-disposition module configured to pre-dispose a violation detection device at the vehicle violation detection site, and obtain a high-position image acquisition array; a trigger range determination module configured to pre-set a trigger distance, and determine a pre-set trigger range based on the pre-set trigger distance and the vehicle violation detection point; an image acquisition module configured to activate the high-position image acquisition array to acquire images of a target vehicle when the target vehicle enters the pre-set trigger range, and obtain a real-time image set; a three-dimensional pose perception module configured to perceive a three-dimensional pose of the target vehicle based on the real-time image set, and obtain a real-time three-dimensional pose parameter; a road geometric parameter acquisition module configured to take the pre-set trigger range as an interaction constraint, and obtain a local road geometric parameter; and a violation identification module configured to identify a violation based on the real-time three-dimensional pose parameter and the local road geometric parameter, and obtain a vehicle violation identification result.

[0008] One or more technical solutions provided in the present disclosure have at least the following technical effects or advantages:

[0009] The above vehicle violation identification method based on three-dimensional pose perception performs detailed analysis on a target road, determines potential violation sites, i.e., specific positions where a vehicle may violate rules. Then, efficient violation detection devices are pre-disposed at these sites to form a high-position image acquisition array for capturing accurate images of violation behaviors. To ensure the timeliness and accuracy of the system, a proper trigger distance is set, and a pre-set trigger range is accurately determined based on the distance and the violation sites. Once a vehicle enters the range, the system immediately starts the high-position image acquisition array to acquire real-time high-definition images of the target vehicle. These real-time images are then processed by three-dimensional pose perception technology to obtain the current three-dimensional pose parameter of the vehicle, reflecting the accurate position and pose of the vehicle. Meanwhile, the geometric parameters of the road itself are also considered as important basis for violation judgment. By combining the above real-time three-dimensional pose parameter and road geometric parameter, it can be accurately determined whether the vehicle has violated rules, and a clear violation identification result is given. This method not only successfully constructs a three-dimensional scene to ensure high-precision identification of the position of the vehicle, but also improves the accuracy and timeliness of violation identification, which helps to improve the safety and order of the entire road traffic system.

[0010] The above description is only a summary of the technical solutions of the present application. In order to make the technical means of the present application more clear and understandable, and to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following will describe the specific embodiments of the present application in detail. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0012] Figure 1 A flowchart of a vehicle violation identification method based on three-dimensional pose perception in an embodiment;

[0013] Figure 2 A system architecture diagram of a vehicle violation identification system based on three-dimensional pose perception in an embodiment.

[0014] Explanation of reference signs: site positioning module 1, device pre-disposition module 2, trigger range determination module 3, image acquisition module 4, three-dimensional pose perception module 5, road geometric parameter acquisition module 6, violation identification module 7. DETAILED DESCRIPTION

[0015] The embodiments of the present application provide a vehicle violation identification method and system based on three-dimensional pose perception, which solves the technical problem that the vehicle position cannot be identified with high precision due to the fact that the traditional monocular camera cannot construct a three-dimensional scene.

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] It should be noted that the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server comprising a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to the process, method, product or device.

[0018] Embodiment one

[0019] As Figure 1As shown, the present application provides a vehicle violation identification method based on three-dimensional pose perception, which comprises:

[0020] Violating site positioning is performed on the target road to obtain vehicle violation detection points;

[0021] The three-dimensional pose of the vehicle refers to the direction and posture of the vehicle in three-dimensional space, including the position of the vehicle and the orientation of the vehicle. Among them, the position of the vehicle refers to the coordinates of the vehicle in three-dimensional space; the orientation of the vehicle refers to the rotation angle of the vehicle in three-dimensional space. The vehicle state perception technology in three-dimensional scene aims to accurately judge the three-dimensional pose of the vehicle in complex environment.

[0022] In the embodiment of the present application, in order to find out the area on the road where the violation behavior is easy to occur, the system terminal first analyzes and positions the target road in detail, virtually restores the target road according to the road design information, and identifies the violation point based on the restored virtual model. In this way, the system terminal can determine the key vehicle violation detection points, that is, the places where the vehicle is most likely to violate. The selection of these detection points is crucial for the accuracy and effectiveness of the subsequent violation detection work. Therefore, this step is a key link in building an efficient vehicle violation identification system.

[0023] Further, the present application provides a method for positioning the violation site on the target road to obtain the vehicle violation detection point, which further comprises:

[0024] Interactively obtaining the road design information of the target road;

[0025] The road design information is used as a modeling reference to restore the modeling of the target road by using digital twin technology to obtain a road restoration model;

[0026] Preferably, in order to more accurately understand the actual situation of the target road, the system terminal first obtains the design information of the road, which includes the layout of the road, the lane division, the traffic signs and various elements, and classifies these elements. Then, a terrain model is constructed according to the classified data by using digital twin technology, and on the basis of the terrain model, a model almost identical to the target road, i.e. a road restoration model, is constructed. This model can restore the terrain structure and traffic flow of the target road, helping the system terminal to more intuitively understand the road situation, and providing strong support for subsequent monitoring point identification, vehicle violation identification and other work.

[0027] A preset geometric segmentation constraint is obtained, and the road restoration model is cut based on the geometric segmentation constraint to obtain a set of restoration sub-blocks;

[0028] Preferably, the geometric partitioning constraints refer to a set of rules or conditions that the system terminal follows when cutting the road reconstruction model. The reconstruction sub-block set refers to a collection of sub-models or sub-regions obtained by applying geometric partitioning constraints to the road reconstruction model. These sub-block sets are a division of the original road model, with each sub-block representing a part of the road, such as a specific lane, intersection, road segment, or region with specific geometric characteristics. The system terminal identifies key geometric features from the road reconstruction model, including lane lines, traffic signs, intersections, road edges, and others. These features will serve as reference points for cutting. Then, according to the pre-set geometric partitioning constraints, the specific location and method of cutting are determined. For example, cutting along the lane lines, cutting at intersections, cutting according to specific distances or angles, etc. Then, according to the identified geometric features and applied geometric partitioning constraints, the road reconstruction model is cut. The cutting process involves model segmentation, reconstruction, and other operations to ensure that each sub-block retains the detailed information and features of the original model. Finally, the reconstructed sub-block set obtained after cutting is verified to ensure that each sub-block meets the expected partitioning effect and does not lose or incorrectly include key road information. If the cutting result is found to be unsatisfactory or problematic, the system terminal adjusts the geometric partitioning constraints and re-cuts to obtain a more accurate reconstructed sub-block set. Each sub-block in the obtained reconstructed sub-block set is a part of the road model, containing specific road features and traffic information. This allows the system terminal to analyze and process each sub-block independently, simplifying and speeding up the analysis process. At the same time, because each sub-block is derived from the original road design information, they can well restore the true road conditions.

[0029] Synchronize the reconstructed sub-block set to the pre-constructed violation site identification network, identify the violation site based on the violation site identification network, and obtain a set of detection point identifiers;

[0030] Locate the detection point identifiers set in the road reconstruction model to obtain the vehicle violation detection point.

[0031] Preferably, the system terminal pre-constructs a violation site identification network, which includes multiple identification channels for identifying the positions where violations are most likely to occur. Then, the previously cut road sub-block set is synchronized into the violation site identification network, which allocates an identification channel for each sub-block to identify the violation site, obtaining a detection point identification set. Through the detection point identification set, the system terminal can understand which positions are high-risk areas for vehicle violations. Then, the system terminal determines the correspondence between the detection point identification set and the road restoration model, unifies the coordinate systems, and matches the identification marks. Then, each identification point in the detection point identification set is matched with the corresponding position in the road restoration model to determine its accurate position in the road restoration model. After the matching is completed, the system terminal verifies the positioning restoration result to ensure that each detection point is accurately restored to the corresponding position in the road restoration model. If the positioning restoration result is inaccurate or has problems, adjustments and optimizations are made, such as re-coordinating or adjusting the accuracy of the model, to improve the accuracy of the positioning restoration. If the positioning restoration result is accurate, each point in the detection point identification set is restored to the specific position in the road restoration model, and a mark is drawn on the model for highlighting. In this way, the system terminal can intuitively understand which positions are violation detection points on the road restoration model, making it easier to conduct subsequent traffic management and monitoring work.

[0032] Further, the application provides that the road design information is used as a modeling reference to model and restore the target road using digital twin technology to obtain a road restoration model. The method further comprises:

[0033] The first modeling data attribute and the second modeling data attribute are preset, the road design information is called based on the first modeling data attribute to obtain a terrain-related data set, and the road design information is called based on the second modeling data attribute to obtain a road-related data set.

[0034] Optionally, the road design information is taken as a modeling reference, and digital twin technology is used for modeling and restoration of the target road to obtain a road restoration model. The first modeling data attribute refers to a set of preset data filtering conditions specially used to extract terrain-related data from the road design information. The second modeling data attribute refers to a set of preset data filtering conditions specially used to extract road-related data from the road design information. The system terminal is pre-configured with two modeling data attributes, namely the first modeling data attribute and the second modeling data attribute. These attributes are all filtering conditions that help the system terminal filter the data needed to build a model from a large amount of road design information. Then, the road design information is called according to the first modeling data attribute, and information related to the terrain is filtered out and arranged into a data set, namely a terrain-related data set. This data set contains all road design information related to the terrain, such as height, slope, geological structure, etc. Then, the road design information is called according to the second modeling data attribute. This time, the system terminal filters information related to the road itself, and similarly, arranges this information into a data set, namely a road-related data set. This data set contains all information related to the design of the road itself, such as length, width, road surface material, number of lanes, etc. The system terminal filters the data related to the terrain and the road itself from the road design information and arranges them into two data sets, respectively, to facilitate subsequent analysis and modeling work.

[0035] The terrain-related data set is taken as a terrain modeling reference, and digital twin technology is used for terrain modeling and restoration to obtain a terrain model.

[0036] The road-related data set is taken as a road modeling reference, and digital twin technology is used for modeling and restoration based on the terrain model to obtain an initial restoration model.

[0037] Optionally, the system terminal takes the previously obtained terrain-related data set as a detailed terrain blueprint for reference during terrain modeling. This data set contains information such as the ups and downs of the terrain and the characteristics of the landscape. Then, features related to terrain modeling are extracted from the data set, such as the height of the terrain, the slope, and the watershed analysis. The purpose of feature selection is to reduce the complexity of the model and improve the efficiency and accuracy of modeling. Then, according to the extracted features and actual requirements, digital twin modeling methods are used for terrain modeling and restoration, thereby obtaining a terrain model that accurately reflects the characteristics of the real terrain. Then, the system terminal takes the terrain model as a basis and takes the road-related data set as a reference for road modeling. Using the same method as before, digital twin technology is used again to model and restore the road-related data set based on the terrain model. In this way, the system terminal obtains an initial road restoration model that accurately reflects the position and shape of the road on the real terrain.

[0038] The upstream road traffic data set and the downstream road traffic data set of the target road are obtained through interaction, and traffic fitting is performed on the initial restoration model using the upstream road traffic data set and the downstream road traffic data set, so as to obtain the road restoration model.

[0039] Optionally, the system terminal obtains the upstream road traffic data set and the downstream road traffic data set of the target road through interaction with the road traffic database. These data sets contain traffic information upstream and downstream of the road, which is crucial for the system terminal to understand the road traffic situation. Then, these upstream and downstream traffic data sets are decomposed into different components, such as seasonal decomposition, trend decomposition, etc., using time series decomposition techniques. This helps the system terminal understand the changing trend, seasonal variation and periodic variation of traffic flow. Then the decomposed data is input into the initial restoration model for traffic fitting. This process is to continuously adjust the parameters of the initial restoration model so that it can better describe the actual traffic situation. Through continuous adjustment and optimization, the system terminal finally obtains a road restoration model that more accurately reflects the real road traffic situation.

[0040] Further, the application provides synchronizing the restoration atom block set to a pre-constructed violation site identification network, identifying a violation site based on the violation site identification network to obtain a detection point identification set, and the method further comprises:

[0041] The violation site identification network comprises K identification channels, and the K identification channels are respectively built based on K preset operators;

[0042] A plurality of sample restoration atom blocks are obtained through interaction, wherein the plurality of sample restoration atom blocks have sample detection point identifications and sample traffic identifications based on artificial semantics;

[0043] The K identification channels are trained using the plurality of sample restoration atom blocks until the detection point identification accuracy of the K identification channels is higher than a preset value;

[0044] Optionally, before building the illegal site identification network, the system terminal analyzes a large number of machine learning algorithms in advance according to factors such as the adaptability of the problem, historical performance, computational efficiency, and whether it can handle large-scale data, and selects K from them, which will serve as the basis for building K identification channels. Once the preset operator selection is complete, the system terminal begins to build the identification channels. Each identification channel will be built based on a specific preset operator. This means that each channel will use the corresponding machine learning algorithm to process and analyze the input data. During the building process, parameter adjustment and optimization are also needed to ensure that the performance of each channel is at its best. When all K identification channels are built, the system terminal integrates them into a unified network structure, i.e., the illegal site identification network. Then, the system terminal obtains multiple sample restoration atomic blocks through interaction. These sample restoration atomic blocks contain two key pieces of information: one is the sample detection point identification based on artificial semantics, which indicates certain specific locations or features; the other is the sample traffic volume identification, i.e., the traffic volume data represented by these sub-blocks. After obtaining these sample restoration atomic blocks, the system terminal uses them to train the K identification channels. The purpose of training is to enable these identification channels to accurately identify and analyze the detection points in the samples. The training process will continue until the detection point identification accuracy of the K identification channels is higher than the preset value, and the training process will stop. In summary, this process is to make the illegal site identification network more accurate and reliable through continuous learning and adjustment, so as to accurately identify and analyze the illegal sites in the traffic volume data.

[0045] Call a first restoration atomic block from the set of restoration atomic blocks, and synchronize the first restoration atomic block to the K identification channels of the illegal site identification network for illegal site identification, to obtain K groups of candidate detection point identifications;

[0046] Pre-set an identification repetition frequency threshold, and perform screening processing on the K groups of candidate detection point identifications based on the identification repetition frequency threshold, to obtain a first detection point identification;

[0047] By analogy, synchronize the set of restoration atomic blocks to the illegal site identification network for illegal site identification, to obtain the set of detection point identifications.

[0048] Optionally, the system terminal takes out the first restoration atom block, i.e., the first restoration atom block, from the restoration atom block set in order. This sub-block contains part of the information of the road. Then, the first restoration atom block is synchronized to the K identification channels in the violation site identification network. Each identification channel is built based on a different preset operator, and they will independently analyze and process this sub-block. After processing by these identification channels, the system terminal obtains K sets of candidate detection point identifiers. Each set of identifiers represents the violation site that the identification channel considers to be possible. Then, in order to filter out the most likely violation site from these candidate detection point identifiers, the system terminal presets an identifier repetition frequency threshold. The threshold sets a standard, and only a candidate detection point identifier that meets or exceeds this standard will be considered a true violation site. Based on this threshold, the system terminal performs filtering processing on the K sets of candidate detection point identifiers. After filtering, the first detection point identifier, i.e., the most likely violation site, is obtained. Further, using the same method as described above, the system terminal will continue to call the next sub-block from the restoration atom block set and synchronize it to the identification network for violation site identification. Each time a new set of candidate detection point identifiers is obtained, and after filtering processing, a new detection point identifier is obtained. Finally, after all the restoration atom blocks have been processed, the system terminal obtains a complete detection point identifier set. This set contains all the sites that are judged to be violation sites.

[0049] In the pre-deployment of the violation detection equipment at the vehicle violation detection site, a high-position image acquisition array is obtained.

[0050] In one embodiment, after the violation site identification is completed and the detection point identifier set is obtained, the pre-deployment of the violation detection equipment at these identified possible violation sites is performed, including the installation or arrangement of high-position image acquisition equipment at the positions indicated by the detection point identifier set. The purpose of this process is to ensure that the violation behavior can be accurately and effectively captured when the detection equipment is actually deployed, thereby improving the accuracy of violation detection. Through the deployment of these high-position image acquisition equipment, the system terminal can obtain a high-position image acquisition array. This array is composed of multiple image acquisition equipment that work together to cover the entire traffic scene. Each image acquisition equipment will capture the traffic situation within its field of view and transmit this information to the processing center in real time for analysis and processing. In summary, the pre-deployment of the vehicle violation detection site is to ensure that the corresponding image acquisition equipment is arranged at the location where the violation behavior may exist, thereby forming a wide-ranging image acquisition array. In this way, real-time and accurate monitoring of traffic conditions can be achieved, and violation behavior can be discovered and handled in a timely manner.

[0051] A preset trigger distance is preset, and a preset trigger range is determined based on the preset trigger distance and the vehicle violation detection point;

[0052] In one embodiment, during the vehicle violation detection process, the preset trigger distance is a key parameter. This distance is set according to the actual traffic situation and the characteristics of the violation behavior, with the purpose of determining when to start the violation detection equipment and record the violation evidence. Specifically, the preset trigger distance refers to the distance between the vehicle and the violation detection point during the vehicle's travel. When the distance is less than or equal to the preset value, the relevant detection equipment will be triggered and start working. This distance is usually set according to traffic rules, road conditions, and the severity of the violation behavior. For example, at a complex intersection, 10 meters before the stop line at a red light can be set as the preset trigger distance. Once the preset trigger distance is set, the system terminal determines a preset trigger range based on this distance and the vehicle violation detection point. This range is an area centered on the violation detection point with the preset trigger distance as the radius. When the vehicle enters this area, the detection equipment will be activated and start recording the vehicle's behavior for subsequent analysis of whether there is a violation behavior. Through such setting, it can be ensured that only when the vehicle approaches the violation detection point, the detection equipment will be activated, thereby avoiding unnecessary waste of resources and misjudgment. At the same time, this also provides reliable evidence support for subsequent violation behavior processing.

[0053] When the target vehicle enters the preset trigger range, the high-position image acquisition array is activated to acquire images of the target vehicle, obtaining a real-time image set;

[0054] In one embodiment, when the target vehicle approaches the preset trigger range, it represents that the vehicle has entered an alert area. Once the vehicle enters this range, the system terminal will immediately respond and activate the high-position image acquisition array. When they are activated, they will immediately start image acquisition of the target vehicle. These image acquisition devices will capture the real-time dynamics of the target vehicle, including its driving trajectory, speed, etc. Through this series of image acquisition process, the system terminal will obtain a real-time image set. This image set is continuous, dynamic, and contains all important information of the target vehicle from entering the preset trigger range to leaving the range. These images not only have high definition, but also can provide rich details, providing strong evidence for subsequent violation behavior analysis and processing.

[0055] Based on the real-time image set, the target vehicle is subjected to three-dimensional pose perception, obtaining real-time three-dimensional pose parameters;

[0056] In one embodiment, the system terminal performs three-dimensional pose perception on the target vehicle based on the obtained real-time image set. This process mainly uses feature point extraction and matching on the target vehicle to analyze the three-dimensional pose of the vehicle, including parameters such as the position, direction, and inclination of the target vehicle. First, the system terminal processes the real-time image set to locate the position of the target vehicle and extract the feature points of the target vehicle. Then, by comparing images at different time points and angles, the system terminal can establish the relationship between the features and calculate the vehicle's motion trajectory and pose change. Then, the system terminal constructs a three-dimensional model of the vehicle based on the feature points and motion trajectory in these two-dimensional images. This model can reflect the position and pose of the vehicle in the actual space. Finally, through analysis and processing of the three-dimensional model, the system terminal can obtain the real-time three-dimensional pose parameters of the target vehicle. These parameters not only help it understand the actual state of the vehicle in the traffic scene, but also provide important basis for subsequent violation behavior judgment and handling. In summary, three-dimensional pose perception on the target vehicle based on the real-time image set is a complex but necessary process. Through this process, the three-dimensional pose parameters of the vehicle can be obtained, so that the behavior and state of the vehicle in the traffic scene can be more accurately understood.

[0057] Further, the application provides three-dimensional pose perception on the target vehicle based on the real-time image set to obtain real-time three-dimensional pose parameters. The method further comprises:

[0058] The high-position image acquisition array includes M image sensors, and the real-time image set includes M groups of real-time image sequences;

[0059] Timestamp synchronization and synchronized image extraction are performed on the real-time image set to obtain a plurality of synchronized image sets, wherein each synchronized image set includes M real-time images with the same timestamp;

[0060] Preferably, the high-position image acquisition array is composed of multiple image sensors that work simultaneously to capture real-time images of the target vehicle from different angles and positions. Therefore, the real-time image set contains multiple sets of real-time image sequences generated by these sensors, each reflecting the state of the vehicle from different perspectives. To more accurately analyze the three-dimensional pose and violations of the vehicle, the system terminal timestamps synchronize these real-time image sets. Timestamp synchronization is a process that ensures that images captured by all image sensors at the same time can be accurately aligned and processed. This eliminates the problem of image mismatch caused by time differences and improves the accuracy of analysis. After timestamp synchronization is achieved, the system terminal further extracts real-time images corresponding to each timestamp to form multiple synchronized image sets. Each synchronized image set contains M real-time images captured by different image sensors at the same time. These images have the same timestamp, so they show different angle views of the vehicle at the same time. In summary, by timestamp synchronizing the real-time image set and extracting the synchronized image set, the system terminal can ensure that the image data used in analyzing the three-dimensional pose and violations of the vehicle is accurately aligned and synchronized, thereby improving the accuracy and reliability of the analysis.

[0061] Based on the timing sequence, a first synchronized image set is extracted from the multiple synchronized image sets, and a tracking algorithm is used to track and locate the target vehicle in the first synchronized image set to obtain a first limited image set;

[0062] A preset feature extraction rule is extracted, and a feature point automatic extractor is constructed based on the feature extraction rule;

[0063] Preferably, after the system terminal completes the timestamp synchronization and the extraction of the synchronized image sets, the next step is to select the first synchronized image set from these synchronized image sets based on the chronological order. Here, the group of images with the earliest timestamp is selected as the starting point for tracking and positioning, i.e., the first synchronized image set. Then, the system terminal uses a tracking algorithm to track and locate the target vehicle on the first synchronized image set. The tracking algorithm is based on feature points, contours, or color information, and its goal is to accurately identify and locate the same vehicle in consecutive image frames. Through the tracking algorithm, the system terminal can obtain the precise position and motion trajectory of the target vehicle in the image. During the tracking process, the system terminal generates a first cropped image set. This image set contains only the target vehicle after being processed by the tracking algorithm. These images have been cropped and scaled to only retain the pixel coordinates or the position of the bounding box where the target vehicle is located, in order to highlight the target vehicle and reduce the interference of background information, thereby facilitating subsequent analysis and processing. In order to more effectively perform feature extraction, the system terminal also predefines some feature extraction rules. These rules are based on the shape, color, texture, and other features of the vehicle, and are used to guide the construction of a feature point automatic extractor. The feature point automatic extractor is a extraction component that can automatically find the target vehicle from the image and extract key feature points such as the front, rear, and wheel center according to the pre-defined rules. These feature points play an important role in the subsequent identification process, helping the system terminal to more accurately identify the target vehicle and track its motion trajectory in the image.

[0064] Synchronize the first cropped image set to the feature point automatic extractor for feature point extraction, obtaining a first feature point identifier set;

[0065] Perform feature point matching on the first feature point identifier set, and perform three-dimensional pose perception on the target vehicle according to the feature point matching result, obtaining a first instantaneous three-dimensional pose parameter;

[0066] Similarly, based on the chronological order, the multiple synchronized image sets are used to perform three-dimensional pose perception on the target vehicle, obtaining the real-time three-dimensional pose parameter.

[0067] Preferably, after the system terminal completes the tracking and positioning of the target vehicle and obtains the first set of limited images, the next step is to synchronize these limited image sets to the feature point automatic extractor for feature point extraction. The feature point automatic extractor will automatically identify and extract these feature points from the limited image set according to the pre-set feature extraction rules. The extracted feature points will be sorted into the first feature point identification set, which contains the location, type and other related information of the feature points. Then, the system terminal will perform feature point matching on this feature point identification set. Feature point matching is the process of matching the same feature points in different images. By comparing the location, shape, color and other information of the feature points in different images, it can be determined whether they match. Once the feature point matching is completed, the system terminal performs three-dimensional pose perception on the target vehicle according to the matching results. This process combines the internal and external parameters of the image sensor for three-dimensional reconstruction to perceive the three-dimensional pose of the vehicle. By comparing the position differences and relative relationships of feature points in different images, the system terminal can calculate the three-dimensional pose parameters of the vehicle in the actual space, such as position, direction and inclination. This process will be performed in chronological order, i.e. multiple synchronized image sets will be processed in sequence. For each synchronized image set, the system terminal will perform feature point extraction, feature point matching and three-dimensional pose perception to obtain the instantaneous three-dimensional pose parameters of the target vehicle at different times. By combining these instantaneous three-dimensional pose parameters, the system terminal obtains the real-time three-dimensional pose parameters of the target vehicle, which can fully and accurately reflect the actual state and behavior of the vehicle in the traffic scene. In summary, this process is automated and efficient, capable of real-time processing and analysis of large amounts of image data to provide accurate basic data for vehicle violation detection.

[0068] Further, the application provides that the first feature point identification set is subjected to feature point matching, and the target vehicle is subjected to three-dimensional pose perception according to the feature point matching result to obtain first instantaneous three-dimensional pose parameters. The method further comprises:

[0069] Analyzing and obtaining M sets of internal and external parameter information of the M image sensors;

[0070] Performing feature point matching on the first feature point identification set to obtain multiple sets of feature point correspondence relationships, which constitute the feature point matching result;

[0071] Optionally, the system terminal is crucial in processing the three-dimensional pose perception of the target vehicle. The parameters of the image sensor are essential in this process. These parameters include both intrinsic and extrinsic parameters, which together determine how the image sensor captures and represents objects in the real world. Intrinsic parameters, also commonly known as camera intrinsics, describe the characteristics of the camera itself, such as focal length, principal point coordinates, distortion coefficients, and so on. These parameters are necessary to understand how images are captured by the camera and converted into digital format. Extrinsic parameters, or camera extrinsics, describe the position and orientation of the camera in three-dimensional space, i.e., the pose of the camera. They usually include a rotation matrix and a translation vector, which are used to align the camera coordinate system with the world coordinate system. After obtaining these intrinsic and extrinsic parameter information, the system terminal analyzes them to understand how each image sensor affects the captured images and ensures that these differences are taken into account when performing three-dimensional pose perception. Then, the system terminal matches the feature points extracted from the same vehicle in different angle images, and when two feature points are sufficiently similar, they are considered to be matched. Through this process, the system terminal obtains a set of corresponding relationships between feature points, which constitute the feature point matching result. These results are crucial for subsequent three-dimensional pose perception, as they provide matching information of vehicle features under different perspectives, enabling the system to accurately calculate the three-dimensional pose of the vehicle. In summary, analyzing the intrinsic and extrinsic parameter information of the image sensor is to understand how images are captured and represented, while feature point matching is to correspond the same feature points in different images. These two steps together form the basis of three-dimensional pose perception, providing necessary information for accurately calculating the three-dimensional pose of the vehicle.

[0072] According to the feature point matching result and the M sets of intrinsic and extrinsic parameter information, performing three-dimensional reconstruction to obtain a feature point three-dimensional coordinate set;

[0073] According to the feature point three-dimensional coordinate set, performing pose estimation on the target vehicle to obtain the first instantaneous three-dimensional pose parameter.

[0074] Optionally, after obtaining the feature point matching results and the internal and external parameter information of the image sensor, the system terminal proceeds to the process of three-dimensional reconstruction. The goal of three-dimensional reconstruction is to recover the coordinates of the feature points in three-dimensional space according to the feature point matching results in two-dimensional images, combined with the internal and external parameter information of the camera. First, the system terminal establishes a camera model using the internal and external parameter information. The internal parameters define the internal characteristics of the camera, while the external parameters define the position and direction of the camera in the world coordinate system. Through these parameters, the system terminal can accurately map the two-dimensional image points captured by the camera to the three-dimensional space. Then, the system terminal connects the matched feature point pairs according to the feature point matching results. This means that for each matched feature point pair, the system terminal knows their positions in their respective images. Then, using the camera model and the matched feature point pairs, the system terminal calculates the coordinates of these feature points in three-dimensional space through triangulation. After completing the three-dimensional reconstruction, the system terminal obtains a set of three-dimensional coordinates of the feature points, which contains the precise positions of the key feature points on the target vehicle in three-dimensional space. Then, the system terminal uses these three-dimensional coordinates to estimate the pose of the target vehicle. Pose estimation is a process of calculating the direction and position of the target object in three-dimensional space. For the target vehicle, this includes the orientation, position, and possible rotation of the vehicle. Finally, the system terminal obtains the first instantaneous three-dimensional pose parameters based on the results of pose estimation, which describe the three-dimensional pose of the target vehicle at the current time. In summary, this process combines feature point matching results, internal and external parameter information of the image sensor, and three-dimensional reconstruction techniques to ultimately obtain the three-dimensional pose parameters of the target vehicle at the current time. These parameters provide important basis for subsequent vehicle behavior analysis, violation detection, etc.

[0075] Obtain local road geometry parameters as an interactive constraint by using the preset trigger range;

[0076] In one embodiment, when the preset trigger range is set, the system terminal uses it as an interactive constraint to further obtain and analyze the geometry parameters of the local road. These geometry parameters describe the specific shape, width, curvature, and other features of the road, which are crucial for accurately determining whether the vehicle is in violation. Specifically, the system terminal will extract the geometric information of the relevant road area according to the preset trigger range. At the same time, other geometric features of the road, such as lane lines, traffic signs, etc., are also included in the analysis. Finally, these parameters are used to determine whether the vehicle is in violation, providing important data support and analysis basis for subsequent vehicle violation detection.

[0077] Perform violation identification according to the real-time three-dimensional pose parameters and the local road geometry parameters to obtain a vehicle violation identification result.

[0078] In one embodiment, the violation identification based on real-time three-dimensional pose parameters and local road geometry parameters is a key step in the vehicle violation detection system. In this process, the system terminal will analyze the real-time three-dimensional pose parameters of the vehicle with the pre-acquired local road geometry parameters. First, the system terminal will understand the specific position, direction and pose of the vehicle in the road according to the real-time three-dimensional pose parameters. This includes whether the vehicle changes lanes, whether it exceeds the speed limit, whether it drives in the opposite direction, etc. At the same time, the system terminal will also consider the traffic environment around the vehicle, such as lane width, traffic sign position, etc., which are all derived from the previously extracted local road geometry parameters. Then, the real-time pose of the vehicle is matched with the road geometry parameters. For example, if the vehicle is identified as driving in the opposite direction, it will check whether the current position and direction of the vehicle are opposite to the direction of the road; if the vehicle exceeds the speed limit, it will compare the actual speed of the vehicle with the speed limit of the road. Finally, according to the results of the analysis, the system terminal will obtain the identification results of whether the vehicle violates the rules. These identification results are crucial for traffic management and safety monitoring, which can help maintain traffic order and ensure road safety.

[0079] Further, the application provides a method for obtaining vehicle violation identification results by identifying violations based on real-time three-dimensional pose parameters and local road geometry parameters, the method further comprising:

[0080] generating a planar frame sequence according to the real-time three-dimensional pose parameters;

[0081] dividing the driving area according to the local road geometry parameters to obtain a plurality of driving area constraints;

[0082] Optionally, generating a sequence of planar frame based on real-time three-dimensional pose parameters is a process of converting vehicle pose information in three-dimensional space into two-dimensional planar image representation. This step is to convert the three-dimensional position and orientation information of the vehicle into a bounding box or frame in a two-dimensional image, so that the position and pose of the vehicle on the planar road can be more intuitively displayed. After obtaining the real-time three-dimensional pose parameters, the system terminal will calculate the projection position and direction of the vehicle on the two-dimensional plane according to these parameters. The projection position is the center point or a certain specific point of the bottom of the vehicle, and the direction reflects the driving direction of the vehicle. By drawing these position and direction information into a planar frame, the system terminal can generate a continuous planar frame sequence to show the moving track of the vehicle on the road. On the other hand, the local road geometry parameters provide specific information about the shape, width, curvature, etc. of the road, which is crucial for the division of the driving area. Driving area constraints refer to the range and restrictions of the vehicle that can be determined according to the road geometry parameters. These constraints are set based on factors such as road width, lane line position, traffic signs, etc. to ensure that the vehicle drives safely and in compliance with the rules on the road. In the process of driving area division, the system terminal will identify different lanes, intersections, sidewalks, etc. according to the local road geometry parameters, and set corresponding driving constraints according to the characteristics and requirements of these areas. For example, in some areas, the speed of the vehicle may be limited, while in other areas, the vehicle may be required to maintain a certain driving direction. By combining the planar frame sequence generated based on real-time three-dimensional pose parameters and the driving area constraints divided based on local road geometry parameters, the system terminal can more accurately determine whether the vehicle has violated traffic rules or driving constraints. This is crucial for traffic management and safety monitoring, and helps to discover and handle potential traffic safety hazards in a timely manner. In summary, generating a sequence of planar frame based on real-time three-dimensional pose parameters is to more intuitively display the position and pose of the vehicle on the planar road, while dividing the driving area based on local road geometry parameters is to determine the range and restrictions of the vehicle that can be driven. These two steps together provide important visual and analytical basis for vehicle violation detection.

[0083] Traverse the planar frame sequence with the plurality of driving area constraints to identify lane change violations and obtain a first violation identification result;

[0084] Perform driving speed analysis based on the planar frame sequence to obtain a real-time driving speed, and generate a second violation identification result based on the real-time driving speed;

[0085] Perform driving direction analysis based on the planar frame sequence to obtain driving direction information, and generate a third violation identification result based on the driving direction information;

[0086] Integrate the first violation identification result, the second violation identification result, and the second violation identification result to obtain the vehicle violation identification result.

[0087] Optionally, the system terminal traverses the planar frame sequence using the driving area constraints to identify lane changing violations. This means that when a vehicle changes lanes without a legal reason, the system terminal identifies it as a violation. By comparing the vehicle planar frame with the road geometry parameters, it can determine whether the lane changing behavior is legal and generate a first violation identification result. Second, the system terminal analyzes the driving speed based on the planar frame sequence. By calculating the position change of the vehicle at different time points, it can estimate the real-time driving speed of the vehicle. Then, compare the real-time driving speed with the road speed limit. If the vehicle is speeding, it will be identified as a violation and a second violation identification result will be generated. In addition, the system terminal also analyzes the driving direction based on the planar frame sequence. By monitoring the orientation change of the vehicle planar frame, it can determine whether the vehicle is driving in the prescribed direction. For example, behaviors such as driving against traffic on a one-way street or turning left at a no-left-turn intersection will be identified as violations and a third violation identification result will be generated. Finally, the system terminal integrates the first violation identification result, the second violation identification result and the third violation identification result to form the final vehicle violation identification result. In this way, it can comprehensively understand the vehicle's violation during driving and take appropriate handling measures.

[0088] To sum up, the embodiments of the present application have at least the following technical effects:

[0089] The embodiments of the present application realize real-time image acquisition and three-dimensional pose perception of the target vehicle through steps such as positioning the target road violation site, pre-deploying violation detection equipment, and setting the trigger range. Further, through digital twin technology to restore the target road and establish a violation site identification network, the detection point identification accuracy is improved. At the same time, considering multiple driving area constraints, the planar frame sequence, driving speed and direction information are comprehensively analyzed, so as to realize accurate and real-time identification of vehicle violation behavior, and provide an efficient and accurate solution for traffic management. These technical effects collectively solve the technical problem that traditional monocular cameras cannot construct a three-dimensional scene, ensuring high-precision identification of vehicle position.

[0090] Embodiment Two

[0091] Based on the same inventive concept as the three-dimensional pose perception-based vehicle violation identification method in the foregoing embodiments, as shown in Figure 2 The present application provides a three-dimensional pose perception-based vehicle violation identification system, which comprises:

[0092] Site positioning module 1: the site positioning module 1 is used for positioning the violation site of the target road, and obtains the vehicle violation detection point;

[0093] The device pre-deployment module 2 is configured to pre-deploy a violation detection device at the vehicle violation detection site, and obtain a high-position image acquisition array.

[0094] The trigger range determination module 3 is configured to preset a trigger distance, and determine a preset trigger range according to the preset trigger distance and the vehicle violation detection site.

[0095] The image acquisition module 4 is configured to activate the high-position image acquisition array to acquire images of the target vehicle when the target vehicle enters the preset trigger range, and obtain a real-time image set.

[0096] The three-dimensional posture perception module 5 is configured to perceive a three-dimensional posture of the target vehicle based on the real-time image set, and obtain a real-time three-dimensional posture parameter.

[0097] The road geometry parameter acquisition module 6 is configured to obtain a local road geometry parameter by taking the preset trigger range as an interaction constraint.

[0098] The violation identification module 7 is configured to identify a violation according to the real-time three-dimensional posture parameter and the local road geometry parameter, and obtain a vehicle violation identification result.

[0099] Further, the site positioning module 1 is configured to perform the following method:

[0100] Obtain road design information of the target road through interaction;

[0101] Take the road design information as a modeling reference, and perform modeling and restoration of the target road by using a digital twin technology, and obtain a road restoration model;

[0102] Preset a geometry segmentation constraint, and perform cutting processing on the road restoration model based on the geometry segmentation constraint, and obtain a restoration sub-block set;

[0103] Synchronize the restoration sub-block set to a pre-constructed violation site identification network, identify a violation site based on the violation site identification network, and obtain a detection point identification set;

[0104] Position and restore the detection point identification set on the road restoration model, and obtain the vehicle violation detection site.

[0105] Further, the site positioning module 1 is configured to perform the following method:

[0106] A first modeling data attribute and a second modeling data attribute are preset, and a terrain-related dataset is obtained by data calling the road design information based on the first modeling data attribute, and a road-related dataset is obtained by data calling the road design information based on the second modeling data attribute;

[0107] The terrain-related dataset is taken as a terrain modeling reference, and terrain modeling restoration is performed by using a digital twin technology to obtain a terrain model;

[0108] The road-related dataset is taken as a road modeling reference, and modeling restoration is performed by using a digital twin technology based on the terrain model to obtain an initial restoration model;

[0109] Upstream road traffic data sets and downstream road traffic data sets of the target road are obtained interactively, and traffic flow fitting is performed on the initial restoration model by using the upstream road traffic data sets and the downstream road traffic data sets to obtain the road restoration model.

[0110] Further, the site positioning module 1 is used to perform the following method:

[0111] The violation site identification network includes K identification channels, and the K identification channels are respectively built based on K preset operators;

[0112] A plurality of sample restoration atomic blocks are obtained interactively, wherein the plurality of sample restoration atomic blocks have sample detection point identifiers and sample traffic flow identifiers based on artificial semantics;

[0113] The K identification channels are trained by using the plurality of sample restoration atomic blocks until the detection point identifier accuracy of the K identification channels is higher than a preset value;

[0114] A first restoration atomic block is called from the restoration atomic block set, and the first restoration atomic block is synchronized to the K identification channels of the violation site identification network for violation site identification to obtain K groups of candidate detection point identifiers;

[0115] A identifier repetition frequency threshold is preset, and the K groups of candidate detection point identifiers are screened based on the identifier repetition frequency threshold to obtain a first detection point identifier;

[0116] By analogy, the restoration atomic block set is synchronized to the violation site identification network for violation site identification to obtain the detection point identifier set.

[0117] Further, the three-dimensional posture perception module 5 is used to perform the following method:

[0118] The high-position image acquisition array includes M image sensors, and the real-time image set includes M groups of real-time image sequences;

[0119] timestamp synchronization and synchronous image extraction are performed on the real-time image set to obtain a plurality of synchronous image sets, wherein each synchronous image set comprises M real-time images with the same timestamp;

[0120] Based on the time sequence order, a first synchronous image set is extracted from the plurality of synchronous image sets, and a tracking algorithm is used to track and locate the target vehicle in the first synchronous image set to obtain a first limited image set;

[0121] A feature extraction rule is preset, and a feature point automatic extractor is constructed based on the feature extraction rule;

[0122] The first limited image set is synchronized to the feature point automatic extractor for feature point extraction to obtain a first feature point identification set;

[0123] Feature point matching is performed on the first feature point identification set, and three-dimensional pose perception is performed on the target vehicle according to the feature point matching result to obtain a first instantaneous three-dimensional pose parameter;

[0124] By analogy, the plurality of synchronous image sets are used to perform three-dimensional pose perception on the target vehicle based on the time sequence order to obtain the real-time three-dimensional pose parameter.

[0125] Further, the three-dimensional pose perception module 5 is configured to perform the following method:

[0126] M sets of internal and external parameter information of the M image sensors are analyzed and obtained;

[0127] Feature point matching is performed on the first feature point identification set to obtain a plurality of feature point correspondence relationships, which constitute the feature point matching result;

[0128] Three-dimensional reconstruction is performed according to the feature point matching result and the M sets of internal and external parameter information to obtain a feature point three-dimensional coordinate set;

[0129] Pose estimation is performed on the target vehicle according to the feature point three-dimensional coordinate set to obtain the first instantaneous three-dimensional pose parameter.

[0130] Further, the three-dimensional pose perception module 5 is configured to perform the following method:

[0131] A planar frame sequence is generated according to the real-time three-dimensional pose parameter;

[0132] Driving area division is performed according to the local road geometric parameter to obtain a plurality of driving area constraints;

[0133] Lane change violation identification is performed by using the plurality of driving area constraints to traverse the planar frame sequence to obtain a first violation identification result;

[0134] According to the plan view frame sequence, driving speed analysis is performed to obtain real-time driving speed, and a second violation identification result is generated according to the real-time driving speed;

[0135] According to the plan view frame sequence, driving direction analysis is performed to obtain driving direction information, and a third violation identification result is generated according to the driving direction information;

[0136] The first violation identification result, the second violation identification result, and the third violation identification result are integrated to obtain the vehicle violation identification result.

[0137] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes the specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be executed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0138] The above only describes the preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0139] The present application is only an exemplary description of the present application, and is considered to cover any and all modifications, changes, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art can make various modifications and changes to the present application without departing from the scope of the present application. Thus, if these modifications and changes of the present application belong to the scope of the present application and its equivalents, the present application intends to include these modifications and changes.

Claims

1. A method for identifying vehicle violations based on three-dimensional pose perception, characterized in that, The method comprises: violation site positioning is performed on a target road to obtain a vehicle violation detection point, and the method comprises: interactively obtaining road design information of the target road; using the road design information as a modeling reference, performing modeling restoration on the target road by using a digital twin technology to obtain a road restoration model, and the method further comprises: presetting first modeling data attributes and second modeling data attributes, performing data calling on the road design information based on the first modeling data attributes to obtain a terrain-related data set, and performing data calling on the road design information based on the second modeling data attributes to obtain a road-related data set; using the terrain-related data set as a terrain modeling reference, performing terrain modeling restoration by using the digital twin technology to obtain a terrain model; using the road-related data set as a road modeling reference and using the terrain model as a modeling basis, performing modeling restoration by using the digital twin technology to obtain an initial restoration model; interactively obtaining an upstream road traffic data set and a downstream road traffic data set of the target road, and performing traffic flow fitting on the initial restoration model by using the upstream road traffic data set and the downstream road traffic data set to obtain the road restoration model; prearranging a violation detection device at the vehicle violation detection point to obtain a high-position image acquisition array; presetting a trigger distance and determining a preset trigger range according to the preset trigger distance and the vehicle violation detection point; when a target vehicle enters the preset trigger range, activating the high-position image acquisition array to perform image acquisition on the target vehicle to obtain a real-time image set; performing three-dimensional posture perception on the target vehicle based on the real-time image set to obtain real-time three-dimensional posture parameters; using the preset trigger range as an interactive constraint to obtain local road geometric parameters; performing violation recognition according to the real-time three-dimensional posture parameters and the local road geometric parameters to obtain a vehicle violation recognition result, and the method comprises: generating a planar frame sequence according to the real-time three-dimensional posture parameters; performing driving area division according to the local road geometric parameters to obtain a plurality of driving area constraints; performing lane change violation recognition by using the plurality of driving area constraints to traverse the planar frame sequence to obtain a first violation recognition result; performing driving speed analysis according to the planar frame sequence to obtain a real-time driving speed, and generating a second violation recognition result according to the real-time driving speed; performing driving direction analysis according to the planar frame sequence to obtain driving direction information, and generating a third violation recognition result according to the driving direction information; 2. The method of claim 1, wherein, integrating the first violation recognition result, the second violation recognition result, and the third violation recognition result to obtain the vehicle violation recognition result. violation site positioning is performed on a target road to obtain a vehicle violation detection point, and the method further comprises: presetting a geometric segmentation constraint, and performing cutting processing on the road restoration model based on the geometric segmentation constraint to obtain a restoration sub-block set; synchronizing the restoration sub-block set to a pre-constructed violation site recognition network, performing violation site recognition based on the violation site recognition network, and obtaining a detection point identification set; In the road restoration model, the detection point identification set positioning restoration is performed to obtain the vehicle violation detection point.

3. The method of claim 2, wherein, The restoration atom block set is synchronized to a pre-constructed violation site identification network, and violation site identification is performed based on the violation site identification network to obtain a detection point identification set. The method further comprises: The violation site identification network comprises K identification channels, and the K identification channels are respectively built based on K preset operators; A plurality of sample restoration atom blocks are interactively obtained, wherein the plurality of sample restoration atom blocks have sample detection point identifications and sample vehicle flow identifications based on artificial semantics; Training of the K identification channels is performed using the plurality of sample restoration atom blocks until the detection point identification accuracy of the K identification channels is higher than a preset value; A first restoration atom block is called from the restoration atom block set and synchronized to the K identification channels of the violation site identification network for violation site identification to obtain K groups of candidate detection point identifications; A preset identification repetition frequency threshold is set, and screening processing of the K groups of candidate detection point identifications is performed based on the identification repetition frequency threshold to obtain a first detection point identification; In this way, the restoration atom block set is synchronized to the violation site identification network for violation site identification to obtain the detection point identification set.

4. The method of claim 1, wherein, Based on the real-time image set, three-dimensional pose perception of the target vehicle is performed to obtain real-time three-dimensional pose parameters. The method further comprises: The high-position image acquisition array comprises M image sensors, and the real-time image set comprises M groups of real-time image sequences; Time stamp synchronization and synchronous image extraction are performed on the real-time image set to obtain a plurality of synchronous image sets, wherein each synchronous image set comprises M real-time images with the same time stamp; Based on a time sequence order, a first synchronous image set is extracted from the plurality of synchronous image sets, and tracking positioning of the target vehicle is performed on the first synchronous image set using a tracking algorithm to obtain a first limited image set; A preset feature extraction rule is set, and a feature point automatic extractor is constructed based on the feature extraction rule; The first limited image set is synchronized to the feature point automatic extractor for feature point extraction to obtain a first feature point identification set; Feature point matching is performed on the first feature point identification set, and three-dimensional pose perception of the target vehicle is performed according to the feature point matching result to obtain a first instantaneous three-dimensional pose parameter; In this way, the plurality of synchronous image sets are used to perform three-dimensional pose perception of the target vehicle based on a time sequence order to obtain the real-time three-dimensional pose parameters.

5. The method of claim 4, wherein, Feature point matching is performed on the first feature point identification set, and three-dimensional pose perception of the target vehicle is performed according to the feature point matching result to obtain a first instantaneous three-dimensional pose parameter. The method further comprises: M sets of internal and external parameter information of the M image sensors are analyzed and obtained; Feature point matching is performed on the first feature point identification set to obtain a plurality of feature point correspondence relationships, and the plurality of feature point correspondence relationships constitute the feature point matching result; Three-dimensional reconstruction is performed according to the feature point matching result and the M sets of internal and external parameter information to obtain a feature point three-dimensional coordinate set; According to the feature point three-dimensional coordinate set, a pose of the target vehicle is estimated to obtain a first instantaneous three-dimensional pose parameter.

6. A vehicle violation identification system based on three-dimensional pose perception, characterized in that, The system comprises: A site positioning module is configured to position a violation site on a target road to obtain a vehicle violation detection point. The site positioning module is further configured to interactively obtain road design information of the target road. The site positioning module is further configured to use the road design information as a modeling reference to model and restore the target road by using a digital twin technology to obtain a road restoration model. Specifically, a first modeling data attribute and a second modeling data attribute are preset, the road design information is called based on the first modeling data attribute to obtain a terrain-related data set, and the road design information is called based on the second modeling data attribute to obtain a road-related data set. The terrain-related data set is used as a terrain modeling reference to model and restore the terrain by using the digital twin technology to obtain a terrain model. The road-related data set is used as a road modeling reference to model and restore the road based on the terrain model by using the digital twin technology to obtain an initial restoration model. Upstream road traffic data and downstream road traffic data of the target road are interactively obtained, and the upstream road traffic data and the downstream road traffic data are used to fit the vehicle flow in the initial restoration model to obtain the road restoration model. A device pre-disposition module is configured to pre-dispose a violation detection device at the vehicle violation detection site to obtain a high-position image acquisition array. A trigger range determination module is configured to preset a trigger distance and determine a preset trigger range based on the preset trigger distance and the vehicle violation detection point. An image acquisition module is configured to activate the high-position image acquisition array to acquire images of the target vehicle when the target vehicle enters the preset trigger range to obtain a real-time image set. A three-dimensional pose perception module is configured to perceive a three-dimensional pose of the target vehicle based on the real-time image set to obtain a real-time three-dimensional pose parameter. A road geometry parameter acquisition module is configured to obtain a local road geometry parameter by using the preset trigger range as an interactive constraint. A violation identification module is configured to identify a violation based on the real-time three-dimensional pose parameter and the local road geometry parameter to obtain a vehicle violation identification result. The violation identification module is further configured to generate a planar frame sequence based on the real-time three-dimensional pose parameter. Specifically, a driving area is divided based on the local road geometry parameter to obtain a plurality of driving area constraints. The plurality of driving area constraints are used to traverse the planar frame sequence to identify a lane-changing violation to obtain a first violation identification result. A driving speed is analyzed based on the planar frame sequence to obtain a real-time driving speed, and a second violation identification result is generated based on the real-time driving speed. A driving direction is analyzed based on the planar frame sequence to obtain driving direction information, and a third violation identification result is generated based on the driving direction information. The first violation identification result, the second violation identification result, and the third violation identification result are integrated to obtain the vehicle violation identification result.

Images