RSU holographic traffic flow analysis method and system based on edge calculation

Abstract

The invention provides an RSU holographic traffic flow analysis method and system based on edge computing, and the method comprises the steps: enabling an RSU to control a visible light collection device, a laser radar device and an ETC read-write device to synchronously carry out data collection through unified clock triggering and interface cooperation, and obtaining an original perception data set; processing a traffic flow change area in the set based on radiation field data of an edge side neural radiation field multiplexing historical stable area to obtain change area radiation field data, extracting image semantic information, point cloud geometric information and ETC time sequence information, and performing space-time dimension association integration on the three types of information and the change area radiation field data to obtain a traffic flow change area radiation field data set; the to-be-corrected radiation field data is obtained, and after filling and correction, complete radiation field data subjected to cross-type information complementary optimization is obtained; and performing three-dimensional scene reconstruction, and outputting real-time holographic traffic flow data. According to the invention, the accuracy and continuity of the traffic flow data can be improved.

Description

Technical Field

[0001] This invention relates to the fields of data processing and edge computing, and more specifically, to an RSU holographic traffic flow analysis method and system based on edge computing. Background Technology

[0002] With the integration of intelligent transportation and edge computing technologies, Roadside Unit (RSU) holographic traffic flow analysis technology has gradually become a core technology supporting dynamic road management and traffic efficiency optimization. It relies on various sensing devices deployed within the RSU to collect traffic data, generating comprehensive traffic flow information covering the morphology, location, and movement trajectories of traffic participants, providing accurate data for traffic management decisions. However, current implementation suffers from spatiotemporal misalignment issues due to inconsistent data collection standards from different sources. This hinders data integration, preventing stable spatial support for 3D scene construction. Furthermore, the full-scene rendering mode demands high computing power, making it difficult to adapt to the limited computing resources of the RSU edge, thus constraining data processing efficiency. Consequently, the output traffic flow data fails to meet the accuracy and consistency requirements of holographic applications. Summary of the Invention

[0003] In view of this, the present invention provides an RSU holographic traffic flow analysis method and system based on edge computing.

[0004] According to one aspect of the present invention, an edge computing-based RSU holographic traffic flow analysis method is provided. The method includes: the RSU, through unified clock triggering and interface coordination, controls visible light acquisition devices, lidar devices, and ETC reader / writer devices to synchronously acquire data, obtaining an original sensing data set containing multiple types of sensing information, wherein each type of data in the original sensing data set carries a unified reference timestamp; based on the radiation field data of historical stable regions reused by the edge-side neural radiation field, the traffic flow change regions in the original sensing data set are processed to obtain change region radiation field data containing only change region information; and images are extracted from the original sensing data set. Semantic information, point cloud geometric information, and ETC time-series information are integrated with the radiation field data of the changing area in a spatiotemporal dimension to obtain radiation field data to be corrected. Among them, point cloud geometric information provides three-dimensional spatial structure support for the radiation field data to be corrected. Image semantic information is used to fill the density gaps in sparse point cloud areas in the radiation field data to be corrected, and ETC time-series information is used to calibrate the continuity of vehicle movement trajectories in the radiation field data to be corrected, resulting in complete radiation field data optimized by cross-type information complementarity. Based on the complete radiation field data, three-dimensional scene reconstruction is performed to output real-time holographic traffic flow data that includes the three-dimensional shape, spatial location association, and continuous movement trajectory of traffic participants.

[0005] According to another aspect of the present invention, an edge computing system is provided, comprising: a processor; and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the method described above.

[0006] This invention utilizes a unified clock trigger and interface-based collaborative control to synchronously collect data from three types of heterogeneous data acquisition devices via RSU (Radiation Utility Unit). This assigns a unified reference timestamp to the raw sensing data, avoiding spatiotemporal misalignment caused by asynchronous multi-source data acquisition and providing a precise and consistent foundation for data collection. By reusing historical stable regional data from edge-side neural radiation fields to process traffic flow change regions, it can focus on generating radiation field data for dynamic areas, adapting to RSU edge computing power and reducing resource consumption from ineffective computations. Extracting multi-source information and integrating it with radiation field data from change regions through spatiotemporal correlation, and providing 3D spatial structure support with point cloud geometric information, avoids the problem of information stacking without division of labor, constructing radiation field data with a stable 3D foundation for correction. Filling data gaps with image semantic information and calibrating the continuity of vehicle movement trajectories with ETC (Electronic Toll Collection) time-series information can specifically optimize local defects in radiation field data, improving data integrity and coherence. Based on complete radiation field data, it reconstructs 3D scenes and outputs real-time holographic traffic flow data. It can generate accurate data reflecting the 3D morphology, spatial location relationships, and continuous movement trajectories of traffic participants based on the optimized data, providing support for traffic flow analysis that fits real-world scenarios. Attached Figure Description

[0007] Figure 1 This is a schematic diagram illustrating an application scenario provided by an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram illustrating the implementation process of an edge computing-based RSU holographic traffic flow analysis method provided in an embodiment of the present invention.

[0009] Figure 3 This is a schematic diagram of the hardware entity of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0010] The edge computing-based RSU holographic traffic flow analysis method provided in this invention can be applied to, for example... Figure 1The application environment is shown. RSU102 communicates with edge computing system 104 via a network. A data storage system can store the data that edge computing system 104 needs to process. The data storage system can be integrated into edge computing system 104. RSU (Road Side Unit) is a device deployed along the roadside in an intelligent transportation system, whose main function is to communicate and interact with vehicles and other devices (such as edge computing system 104, visible light acquisition device 10, lidar device 20, and ETC reader / writer 30). Edge computing system 104 can be implemented using a standalone edge computing server or a server cluster consisting of multiple edge computing servers.

[0011] Please refer to Figure 2 The edge computing-based RSU holographic traffic flow analysis method provided in this embodiment of the invention specifically includes the following steps: Step S100: The RSU coordinates with the interface through a unified clock trigger to control the visible light acquisition device, lidar device and ETC reader / writer to collect data synchronously, and obtains a raw sensing data set containing multiple types of sensing information. Each type of data in the raw sensing data set carries a unified reference timestamp.

[0012] In traffic scene monitoring, the Roadside Unit (RSU) is a device deployed along the roadside, bearing the crucial responsibility of coordinating the work of various data acquisition devices. A unified clock trigger is key to achieving synchronized data acquisition, relying on a high-precision clock source, such as an atomic clock, to provide a precise and consistent time reference for each device. Interface coordination, through a defined communication protocol, such as the CAN (Controller Area Network) bus protocol, establishes stable communication connections between the RSU and visible light acquisition devices, LiDAR devices, and ETC (Electronic Toll Collection) readers, ensuring orderly collaboration among these devices.

[0013] Visible light acquisition devices are typically high-definition cameras that capture road scenes at a set frame rate (e.g., 25 frames per second) to obtain rich image and video information, including vehicle appearance, driving status, and pedestrian activity. LiDAR devices scan the surrounding environment by emitting laser beams and receiving reflected light, generating point cloud data. This point cloud data accurately records the three-dimensional spatial position and shape information of objects, which can be used to accurately identify the precise contours and positions of vehicles, obstacles, etc. ETC reader / writer devices monitor the ETC tag signal on vehicles in real time. When a vehicle passes, they quickly read the information from the tag, including the vehicle's unique identifier, passage time, and travel route, among other important data.

[0014] To achieve synchronous data acquisition, the RSU initializes a unified clock system, which is synchronized and calibrated with a high-precision atomic clock to ensure millisecond-level time accuracy. Then, the RSU sends a synchronization trigger frame to the bus according to the CAN bus protocol. Upon receiving this frame, each device immediately initiates its data acquisition program. During data acquisition, the visible light acquisition device captures images according to preset exposure times and aperture sizes to ensure appropriate image clarity and brightness. The lidar device operates with stable transmission power and scanning angle to ensure the accuracy and integrity of the point cloud data. The ETC reader / writer continuously sends interrogation signals to capture the response signals from the vehicle's ETC tag.

[0015] The collected data is uniformly stored in the RSU's local storage device, and each data point is given a unified baseline timestamp. This ensures that various types of data are consistent in time, facilitating subsequent correlation and analysis of multiple data types. For example, when performing traffic flow statistics, visible light images, LiDAR point cloud data, and ETC passage records at the same time can be matched based on the timestamp, thereby more accurately analyzing vehicle movement and traffic congestion.

[0016] Step S200: Based on the radiation field data of the historical stable region of the edge side neural radiation field reuse, the traffic flow change region in the original sensing data set is processed to obtain change region radiation field data containing only change region information.

[0017] Edge-side neural radiation fields operate on edge computing devices, enabling efficient modeling and rendering of environmental radiation fields. By learning from large amounts of training data, they establish a mapping relationship between 3D spatial coordinates and radiation field values, where radiation field values ​​can represent attributes such as color and transparency at a given point. Historically stable regions refer to areas where the radiation field information corresponding to spatial coordinates has not changed significantly over multiple consecutive time periods. Reusing radiation field data from these regions avoids redundant processing of unchanged information, greatly improving data processing efficiency.

[0018] Processing traffic flow change areas in the raw sensing data set is to extract truly valuable dynamic change information from massive amounts of raw data. Traffic flow change areas are typically regions where radiation field information changes due to factors such as vehicle movement, pedestrian movement, and changes in the state of traffic facilities.

[0019] In one implementation, step S200 may include the following steps S210-S260: Step S210: Retrieve continuous historical radiation field data for multiple time periods stored in the edge-side storage medium, covering the entire spatial range monitored by the RSU. Extract the spatial coordinate information corresponding to each frame frame by frame, compare the inter-frame information with the same spatial coordinates point by point, mark the coordinate clusters that have not changed in multiple time periods, combine the radiation field information corresponding to the coordinate clusters, and obtain the radiation field data of the historical stable area. The coordinate accuracy is aligned with the acquisition accuracy of the original sensing data set.

[0020] Extracting the spatial coordinate information for each frame is the first step in a detailed analysis of historical radiation field data. In this process, each frame of data is processed sequentially according to time, extracting the radiation field information corresponding to each spatial coordinate from each frame. This radiation field information includes attribute values ​​such as color and transparency at that point. Then, the information for the same spatial coordinate is compared point-by-point across different frames. Specifically, for each spatial coordinate, its radiation field information is compared in adjacent frames or multiple consecutive frames. If the radiation field information of a coordinate remains unchanged over multiple time periods, i.e., the difference in radiation field values ​​is within a preset error range (e.g., the change in radiation field value is less than 0.01), then that coordinate is marked as unchanged.

[0021] These unchanged coordinates are grouped into coordinate clusters, each cluster consisting of a continuous set of unchanged coordinates representing a region. These coordinate clusters represent relatively stable regions over multiple consecutive time periods. Then, the radiation field information corresponding to these coordinate clusters is combined to obtain the radiation field data for historically stable regions. To ensure the accuracy of subsequent processing, the coordinate precision of the historically stable region radiation field data must be aligned with the acquisition precision of the original sensing data set. For example, if the acquisition precision of the original sensing data set is at the centimeter level, then the coordinate precision of the historically stable region radiation field data should also be set to the centimeter level. This ensures data consistency and accuracy during subsequent comparisons and analyses.

[0022] Step S220: Connect the radiation field data of the historical stable region with the currently running neural radiation field rendering space on the edge side point by point in the spatial coordinates, embed the radiation field information corresponding to each coordinate into the corresponding position in the rendering space, and obtain the reference radiation field base layer covering the entire spatial range of RSU monitoring. The time stamp is synchronized with the unified reference timestamp start frame of the original sensing data set.

[0023] The currently running neural radiation field rendering space on the edge side is a virtual space used for real-time rendering of radiation field information, with a defined coordinate system and rendering rules. This space is built based on a neural network model and can map coordinates in three-dimensional space to corresponding radiation field values, thereby achieving the visualization rendering of the environment.

[0024] In one implementation, step S220 may specifically include the following steps S221-S226: Step S221: Extract the spatial coordinate system parameters of the currently running neural radiation field rendering space on the edge side, covering the entire spatial range monitored by RSU, compare them with the spatial coordinate system parameters of the radiation field data in the historical stable area, and adjust the coordinate parameters of the radiation field data in the historical stable area to align the two types of parameters.

[0025] The spatial coordinate system parameters of the currently running neural radiation field rendering space on the edge side are the core attributes of this rendering space. They define key information such as the origin, axis directions, and coordinate precision. These parameters determine how objects are located and represented in the rendering space, forming the basis for accurate rendering. These parameters can be extracted by accessing the neural radiation field rendering space's configuration file or by calling the relevant API interface. The configuration file typically records the various parameters of the coordinate system in detail, while the API interface provides a way to dynamically obtain these parameters, facilitating runtime parameter querying and adjustment.

[0026] Radiation field data from historically stable regions also have their own spatial coordinate system parameters, which may differ from those in the neural radiation field rendering space. To achieve point-by-point connection between the two spatial coordinates, it is necessary to compare the two types of parameters. During the comparison process, careful checks will be made to ensure that the coordinate origins are consistent, the coordinate axis directions are the same, and the coordinate precision matches. For example, if the coordinate origin of the neural radiation field rendering space is (0,0,0), while the coordinate origin of the historically stable region radiation field data is (5,5,5), then there is an inconsistency in the origins.

[0027] If discrepancies are found between the coordinate parameters of the historical stable region radiation field data and the parameters of the neural radiation field rendering space, the coordinate parameters of the historical stable region radiation field data are adjusted. This adjustment can be achieved using a coordinate transformation matrix. A coordinate transformation matrix can transform coordinates from one coordinate system to another. For example, the offset of the origin can be adjusted using a translation matrix.

[0028] Each coordinate of the historical stable region's radiation field data is represented as a homogeneous coordinate (x, y, z, 1), and then multiplied by this translation matrix to obtain the adjusted coordinates. In this way, the coordinate system of the historical stable region can be converted to a coordinate system consistent with the neural radiation field rendering space, thereby achieving alignment between the two types of parameters.

[0029] Step S222: Sort the radiation field data of the historical stable region after adjusting the coordinate parameters according to the spatial coordinates. The sorting order corresponds to the coordinate sorting rules of the currently running neural radiation field rendering space on the edge side, so that the radiation field information position of each coordinate corresponds to the coordinate position of the rendering space.

[0030] In one implementation, step S222 may specifically include the following steps S2221-S2226: Step S2221: Extract the coordinate sorting rules of the currently running neural radiation field rendering space on the edge side. The rules cover the entire spatial range monitored by RSU, and record the coordinate dimensions and priority order of the sorting.

[0031] The coordinate sorting rules in the neural radiation field rendering space currently running on the edge side are important rules established to ensure the spatial order and consistency of data. These rules are designed by developers based on specific application scenarios and algorithm requirements, defining how spatial coordinates are sorted to ensure that radiation field data can be correctly embedded into the rendering space.

[0032] The rules cover the entire spatial range monitored by the RSU to ensure that all radiation field data are sorted according to a unified set of rules, avoiding inconsistencies in the sorting of some data. Recording the coordinate dimensions and their priority order for sorting is crucial, as this determines how the data will be split and reassembled subsequently. For example, the coordinate sorting rule might specify sorting first by the X-axis in ascending order, then by the Y-axis in ascending order, and finally by the Z-axis in ascending order. Therefore, these three coordinate dimensions and their priority order must be clearly defined during recording.

[0033] Step S2222: Split the radiation field data of the historical stable region after adjusting the coordinate parameters according to the recorded coordinate dimensions. The radiation field information of each dimension corresponds to the corresponding coordinate dimension, and the accuracy of the split information is aligned with the accuracy of the radiation field data of the historical stable region.

[0034] The radiation field data is split into multiple independent dimensions according to the recorded coordinate dimensions. For example, if the coordinate sorting rule involves three coordinate dimensions (X, Y, and Z), then the radiation field data is split into radiation field information in the X, Y, and Z dimensions. The splitting process involves analyzing each coordinate in the radiation field data and extracting its corresponding radiation field information into the respective dimensions. The radiation field information in each dimension only contains data related to that coordinate dimension.

[0035] Step S2223: Reassemble the split radiation field information of each dimension according to the priority order of the records. The reassembly order corresponds to the coordinate sorting rule of the currently running neural radiation field rendering space on the edge side. The reassembled information covers the entire spatial range monitored by RSU.

[0036] After the radiation field data is split, the radiation field information of each dimension is reorganized according to the priority order of the records in order to recombine the split data of each dimension into an ordered radiation field data set, so that it conforms to the coordinate sorting rules of the neural radiation field rendering space currently running on the edge side.

[0037] Based on the recorded coordinate dimensions and priority order, the radiation field information for each dimension is arranged sequentially. For example, if the priority order is X-axis first, then Y-axis, and finally Z-axis, then the radiation field information for the X-axis is first arranged in ascending order. Sorting algorithms such as quicksort and mergesort can be used to sort the radiation field information for the X-axis. Then, based on the X-axis information, the radiation field information for the Y-axis is arranged. Specifically, for each X-coordinate value, the corresponding Y-axis radiation field information is arranged in ascending order. Finally, the Z-axis radiation field information is arranged in its corresponding position, also in ascending order. The reconstructed information must cover the entire spatial range monitored by the RSU, ensuring that no radiation field data from any area is missed. During the reconstructing process, the correspondence between the information of each dimension is maintained, ensuring that the radiation field information for each coordinate remains accurately corresponding after reconstructing. This can be ensured by establishing indexes or mapping relationships. For example, a unique identifier can be created for each coordinate, and this identifier is used during the splitting and reconstructing process to ensure the correct association of the information for each dimension.

[0038] Step S2224: Verify the recombined radiation field data point by point according to spatial coordinates, check whether the radiation field information position of each coordinate corresponds to the coordinate position of the rendering space, and align the verification results with the preset requirements of the currently running neural radiation field rendering space on the edge side.

[0039] After the radiation field data is reconstructed, the reconstructed radiation field data is verified point by point according to spatial coordinates in order to ensure that the radiation field information of each coordinate can accurately correspond to the corresponding position of the currently running neural radiation field rendering space on the edge side after reconstruction, and meet the preset requirements of the rendering space.

[0040] The point-by-point verification process involves comparing each coordinate in the reconstructed radiation field data with the coordinates in the neural radiation field rendering space. It checks whether the radiation field information at each coordinate corresponds to its position in the rendering space, including whether the coordinate values ​​are consistent and whether the attributes of the radiation field information are correct. For example, it checks whether the radiation field information at coordinates (1,2,3) accurately corresponds to the position (1,2,3) in the rendering space, and whether the radiation field value at that position is within a preset range.

[0041] The verification results are aligned with the preset requirements of the currently running neural radiation field rendering space on the edge side. These preset requirements may include the range of radiation field values, data integrity, and coordinate accuracy. If the radiation field information at a certain coordinate is found to be inconsistent with the preset requirements, the reconstruction process is checked and corrected.

[0042] Step S2225: Arrange the verified radiation field data continuously according to spatial coordinates. The arranged information forms a continuous radiation field layer, and the layer range covers the entire spatial range monitored by RSU.

[0043] After completing point-by-point verification, the verified radiation field data is arranged continuously according to spatial coordinates in order to organize the discrete radiation field data into a continuous layer, so as to facilitate subsequent processing and rendering.

[0044] The radiation field information for each coordinate is arranged sequentially according to spatial coordinates. This arrangement can be either ascending or descending, depending on the requirements of the neural radiation field rendering space. The arranged information forms a continuous radiation field layer that covers the entire spatial range monitored by the RSU. This layer can be viewed as a two-dimensional or three-dimensional image, where each pixel or voxel represents the radiation field information for one coordinate. This continuous arrangement provides a more intuitive view of the radiation field distribution, facilitating subsequent visualization and analysis. For example, when visualizing traffic scenes, this radiation field layer can be used as the base data for rendering, resulting in a more realistic and accurate depiction of the traffic scene.

[0045] Step S2226: The continuously arranged radiation field layers are used as the radiation field data of the sorted historical stable region for subsequent operations that embed the currently running neural radiation field rendering space on the edge side.

[0046] The resulting radiation field layer, after continuous arrangement, exhibits good spatial order and continuity, conforming to the coordinate sorting rules of the currently running neural radiation field rendering space on the edge side. Therefore, this radiation field layer can be used as the radiation field data for the sorted historical stable region.

[0047] The sorted historical stable region radiation field data is used for subsequent embedding operations into the currently running neural radiation field rendering space on the edge side. During the embedding process, the radiation field information of each coordinate in the radiation field layer is accurately embedded into the corresponding position in the neural radiation field rendering space, thereby achieving the fusion of historical stable region radiation field data and the neural radiation field rendering space. The embedding operation can be implemented by calling the relevant API interface of the neural radiation field rendering space, passing the radiation field layer data as input parameters to the interface, which will automatically embed the radiation field information into the corresponding position in the rendering space.

[0048] Step S223: Input the sorted historical stable region radiation field data coordinate by coordinate into the currently running neural radiation field rendering space on the edge side. The radiation field information of each coordinate is embedded into the corresponding coordinate position of the rendering space, and the embedding accuracy is aligned with the coordinate accuracy of the rendering space.

[0049] After sorting the historical stable region radiation field data, inputting it coordinate-by-coordinate into the currently running neural radiation field rendering space on the edge side is a key step in achieving the fusion of historical data and the rendering space. Inputting coordinate-by-coordinate means inputting the radiation field information of each coordinate into the rendering space in the sorted order.

[0050] The radiation field information for each coordinate is embedded into the corresponding coordinate position in the rendering space, ensuring that the radiation field information accurately corresponds to the corresponding position in the rendering space. The embedding precision is aligned with the coordinate precision of the rendering space. For example, if the coordinate precision of the rendering space is at the millimeter level, then the positional precision of the radiation field information must also be at the millimeter level during embedding.

[0051] During the embedding process, relevant API interfaces for the neural radiation field rendering space can be invoked. The sorted historical stable region radiation field data is passed as input to the API interface, which then accurately embeds the radiation field information into the corresponding location in the rendering space based on the coordinate information. Simultaneously, error checking and correction are performed during the embedding process to ensure the precision and accuracy of the embedding.

[0052] Step S224: Perform coordinate-by-coordinate verification on the embedded rendering space, check whether the radiation field information of each coordinate corresponds to the radiation field data of the historical stable region, and align the verification result with the preset requirements of the neural radiation field rendering space currently running on the edge side.

[0053] During the verification process, the radiation field information of each coordinate in the embedded rendering space is compared one by one with the radiation field data of the historical stable region. This checks whether the radiation field values ​​are consistent, whether the data integrity is the same, etc. For example, it checks whether the radiation field value of coordinates (2,3,4) in the embedded rendering space is the same as the radiation field value of that coordinate in the historical stable region radiation field data, and whether the data format and attributes are consistent.

[0054] The verification results are aligned with the preset requirements of the currently running neural radiation field rendering space on the edge side. These preset requirements may include the range of radiation field values, data format, and coordinate accuracy. If the radiation field information at a certain coordinate does not correspond to the radiation field data of a historical stable region or does not meet the preset requirements, the cause is investigated and corrected. Possible causes include data transmission errors during the embedding process and coordinate mapping errors.

[0055] Step S225: The verified rendering space is supplemented with the unified reference timestamp start frame information of the original sensing data set according to the time stamp, and the supplemented timestamp is synchronized with the unified reference timestamp start frame of the original sensing data set.

[0056] After embedding and verification are completed, time stamp information is added to the verified rendering space to ensure that the information in the rendering space is consistent with the original perceptual data set in time, which facilitates subsequent data analysis and processing.

[0057] The unified reference timestamp start frame information of the original sensing data set is supplemented into the verified rendering space. The supplemented timestamp is synchronized with the unified reference timestamp start frame of the original sensing data set, ensuring that the time information in the rendering space is consistent with the acquisition time of the original data. This time stamp supplementation can be achieved by adding a timestamp field to the data structure of the rendering space and assigning the unified reference timestamp start frame information to this field.

[0058] Step S226: Integrate the rendering space after the supplemented time stamp into a complete reference radiation field base layer, covering the entire spatial range of RSU monitoring, and synchronize the time stamp with the unified reference timestamp start frame of the original sensing data set.

[0059] The radiation field information and time stamp information of all coordinates in the rendered space after the time stamp is added are summarized and organized. It is ensured that the radiation field information and time stamp information of each coordinate are correct and conform to the requirements of the unified reference timestamp start frame synchronization of the original sensory dataset. Data accuracy and integrity can be guaranteed through data cleaning and verification methods. For example, it is checked whether the radiation field value of each coordinate is within a reasonable range and whether the time stamp is consistent with the unified reference timestamp start frame.

[0060] The integrated baseline radiation field layer covers the entire spatial range of RSU monitoring, comprehensively reflecting the radiation field conditions of the RSU monitoring area. This baseline layer can serve as a benchmark for subsequent analysis of traffic flow change areas. By comparing it with the original sensing data set, areas of traffic flow change can be identified. For example, in subsequent steps, the real-time collected raw sensing data can be compared with the baseline radiation field layer to discover areas where radiation field information has changed, thereby determining the changes in traffic flow.

[0061] After the above steps S221-S226, the connection between the radiation field data of the historical stable region and the currently running neural radiation field rendering space on the edge side is completed, and a reference radiation field base layer covering the entire spatial range of RSU monitoring is obtained, and the time stamp is synchronized with the unified reference timestamp start frame of the original sensing data set.

[0062] Step S230: For the visible light acquisition data, lidar acquisition data and ETC reading and writing data in the original sensing data set, perform spatial coordinate correspondence comparison with the reference radiation field base layer frame by frame, mark the coordinate clusters with information differences in each frame, combine the difference coordinate clusters that appear in multiple consecutive frames, obtain the spatial range of the traffic flow change area, and align the boundary with the acquisition boundary of the original sensing data set.

[0063] The raw perception dataset contains visible light data, lidar data, and ETC (Electronic Toll Collection) data, which record traffic scene information from different perspectives. Visible light data presents visual information about the road in the form of images or videos, including vehicle appearance, color, driving status, and pedestrian activity. LiDAR data provides precise three-dimensional position and shape information of objects; through large amounts of point cloud data, the outlines and positions of vehicles, obstacles, etc., can be accurately identified. ETC data records vehicle passage information, such as vehicle identification, passage time, and travel route.

[0064] The purpose of comparing these data frame by frame with the baseline reference radiation field is to identify which coordinate information differs from that in each frame. During the comparison, the information for each coordinate in the original sensing data set is compared in detail with the corresponding coordinate information in the baseline reference radiation field. For visible light data, features such as color and texture are compared; for LiDAR data, features such as density and distribution of point cloud data are compared; and for ETC (Electronic Toll Collection) data, the consistency of vehicle passage information is compared.

[0065] If a discrepancy is found between the information of a certain coordinate in the original sensing data set and the baseline radiation field, that coordinate is marked as a discrepancy coordinate. Consecutive discrepancy coordinates in each frame are combined into a coordinate cluster, with each cluster representing a possible area of ​​traffic flow change. By combining discrepancy coordinate clusters appearing in multiple consecutive frames, a more accurate spatial range of the traffic flow change area can be obtained. This is because discrepancies appearing in a single frame may be due to noise or random factors, while discrepancies appearing in multiple consecutive frames are more likely to indicate genuine traffic flow changes.

[0066] Step S240: For the spatial range of the traffic flow change area, extract the visible light acquisition information, lidar acquisition information and ETC reading and writing information corresponding to each frame from the original sensing data set, and generate corresponding radiation field rendering information for each coordinate according to the rendering rules of the neural radiation field. The rendering information of each coordinate corresponds to the three types of acquisition information, and the rendering accuracy is aligned with the base layer of the reference radiation field.

[0067] Visible light acquisition information includes image and video information of the area, such as the appearance and driving status of vehicles, and the activities of pedestrians. LiDAR acquisition information provides precise three-dimensional position and shape information of objects; point cloud data can accurately identify the outlines and positions of vehicles, obstacles, etc. ETC read / write information records vehicle passage information, such as vehicle identification, passage time, and passage route.

[0068] According to the rendering rules of neural radiation fields, the extracted information is processed to generate corresponding radiation field rendering information coordinate by coordinate. The rendering rules of neural radiation fields are based on neural network models, which define how to convert the collected information into radiation field values. For example, color and texture features in visible light acquisition information can be mapped to the color attributes of the radiation field through a neural network model; point cloud data in lidar acquisition information can be converted into density and shape information of the radiation field; and for ETC reading and writing information, vehicle passage information can be correlated with the dynamic changes of the radiation field.

[0069] The rendering information for each coordinate corresponds to three types of collected information. That is, the radiation field value of each coordinate is determined by the visible light collected information, lidar collected information and ETC reading and writing information corresponding to that coordinate. In this way, the advantages of multi-source data can be fully utilized to more accurately reflect the actual situation of traffic flow change areas.

[0070] Step S250: The radiation field rendering information of the traffic flow change area is split coordinate by coordinate with the base layer of the reference radiation field. The unchanged coordinate information in the base layer of the reference radiation field is removed, and the radiation field data fragment containing only the change area is extracted. The spatial range is aligned with the spatial range of the traffic flow change area, and the time stamp is synchronized with the unified reference timestamp of the corresponding frame.

[0071] The purpose of splitting the radiation field rendering information of the traffic flow change area with the baseline radiation field layer coordinate by coordinate is to identify which coordinates in the baseline radiation field layer have remained unchanged and which have changed. During coordinate-by-coordinate splitting, the radiation field rendering information of the traffic flow change area is compared one by one with the radiation field information of each coordinate in the baseline radiation field layer.

[0072] If the radiation field information for a certain coordinate is found to be identical in both the baseline radiation field layer and the radiation field rendering information of the traffic flow change area, it indicates that the information for that coordinate has not changed, and it will be removed from the baseline radiation field layer. Only coordinates whose information has changed are retained; these coordinates constitute radiation field data segments containing only the change area. The spatial extent of the extracted radiation field data segments is aligned with the spatial extent of the traffic flow change area, ensuring that the data segments contain only information from the traffic flow change area. Spatial alignment can be achieved by checking whether the coordinate's position is within the spatial extent of the traffic flow change area. If any coordinate is found to be outside the extent of the traffic flow change area, it will be deleted from the data segment.

[0073] Step S260: Match the unified reference timestamp of the original sensing data set to the radiation field data fragment of the changed region, mark the corresponding time information frame by frame, and synchronize the timestamp of each frame of radiation field data with the timestamp of the corresponding collected data to obtain radiation field data of the changed region containing only the information of the changed region, and the format is aligned with the base layer of the reference radiation field.

[0074] Time stamp matching can be achieved by adding a timestamp field to the data structure of the radiation field data segment in the changing region and assigning the unified reference timestamp of the original sensing data set to this field. For example, if the timestamp of the corresponding collected data is "2026-2-6 10:30:00.456", then a timestamp field can be added to the data structure of the radiation field data segment in the changing region, and this value can be assigned to this field.

[0075] The timestamp of each frame of radiation field data is synchronized with the timestamp of the corresponding acquired data, thus ensuring data consistency over time. The format of the obtained radiation field data of the changed region, which only contains information about the changed region, is aligned with the base layer of the reference radiation field, including the data storage format, coordinate precision, and the representation of radiation field values.

[0076] Step S300: Extract image semantic information, point cloud geometric information, and ETC time series information from the original sensing data set. Integrate the three types of information with the radiation field data of the changed area in a spatiotemporal dimension to obtain the radiation field data to be corrected. Among them, the point cloud geometric information provides three-dimensional spatial structure support for the radiation field data to be corrected.

[0077] In one implementation, step S300 may specifically include the following steps S310-S360: Step S310: Divide the visible light acquisition data in the original perception data set into pixel regions frame by frame. The size of each pixel region corresponds to the point cloud sampling unit of the lidar acquisition data. Identify the type of traffic participants, the layout of road facilities and the current traffic status in each region. Integrate to obtain image semantic information containing regional information. The pixel range is aligned with the acquisition range of each frame.

[0078] The visible light acquisition data in the original sensing dataset is divided into pixel regions frame by frame to divide the image data into multiple smaller regions for more detailed analysis and processing. The size of each pixel region corresponds to the point cloud sampling unit of the LiDAR acquisition data. This is to enable accurate spatial matching of image semantic information and point cloud geometric information in subsequent processing.

[0079] Identifying traffic participant types, road infrastructure layouts, and current traffic status region-by-region can employ advanced image recognition and semantic analysis techniques. For example, convolutional neural networks (CNNs) can be used to classify and identify each pixel region of the image. During the identification process, for traffic participant types, CNNs can distinguish between vehicles, pedestrians, and bicycles; for road infrastructure layouts, they can identify streetlights, traffic signs, and zebra crossings; and for current traffic status, they can determine whether a vehicle is moving, parked, or waiting for a traffic light. The identified information is then integrated to form image semantic information containing regional information. This information includes detailed semantic details for each pixel region, such as traffic participant types, road infrastructure layouts, and traffic status. The pixel range is aligned with the acquisition range of each frame to ensure that the image semantic information completely covers the acquisition area of ​​the visible light data.

[0080] Step S320: Perform spatial coordinate clustering on the lidar acquisition data frame by frame in the original sensing data set. Each cluster corresponds to an entity object. Identify the spatial contour, real-time position distribution and relative position between objects for each cluster. Integrate to obtain point cloud geometric information containing clustering information. Its coordinate accuracy is aligned with the pixel accuracy of the visible light acquisition data, and its spatial range is aligned with the acquisition range of each frame.

[0081] Spatial coordinate clustering is performed frame-by-frame on the LiDAR data collected in the original sensing dataset to group a large amount of discrete point cloud data so that each cluster corresponds to a specific entity. Clustering algorithms such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise) and K-Means can be used.

[0082] Clustering can divide the large amount of discrete point cloud data collected by LiDAR into multiple clusters, each representing an object such as a vehicle, pedestrian, or obstacle. Then, the spatial contours, real-time position distribution, and relative positions between objects are identified cluster by cluster. These identification tasks can be achieved using geometric analysis and 3D modeling techniques. For example, the spatial contours of objects are obtained by calculating the boundary points of all points in a cluster; the real-time positions of objects are obtained by calculating the center points of the clusters; and the relative positions between objects are obtained by comparing the positional relationships between different clusters.

[0083] The integrated identification information forms a point cloud geometric information system containing clustering information. This information includes detailed geometric information for each object, such as spatial contours, real-time position distribution, and relative positions between objects. The coordinate precision is aligned with the pixel precision of the visible light acquisition data to ensure accurate spatial matching between the point cloud geometric information and image semantic information. The spatial extent is aligned with the acquisition range of each frame to ensure that the point cloud geometric information completely covers the acquisition area of ​​the LiDAR data.

[0084] Step S330: Arrange the ETC read and write data in the original sensing data set in chronological order. Each record corresponds to one vehicle passage. Extract the vehicle's passage identifier, passage time, and passage route association information for each record. Integrate to obtain ETC time-series information containing time stamps. Its time stamps are synchronized with the unified reference timestamp of the original sensing data set, and its number of records is aligned with the number of ETC collections.

[0085] Arranging the ETC read / write data in the original sensing data set in chronological order is to organize the vehicle passage records according to the time sequence. Sorting algorithms, such as quicksort and mergesort, can be used to sort the ETC read / write data.

[0086] Each record corresponds to a single vehicle passage. For each record, the vehicle's passage identifier, passage time, and associated route information are extracted. The passage identifier can be the vehicle's license plate number, ETC tag number, etc., used to uniquely identify a vehicle. The passage time records the specific time the vehicle passed through the ETC reader / writer, while the associated route information records the vehicle's travel path, such as the toll stations entered and exited the highway.

[0087] The extracted information is integrated to form ETC time-series information with timestamps. This information includes the passage identifier, passage time, and passage route association information for each vehicle, and the timestamps are synchronized with the unified reference timestamp of the original sensing data set to ensure accurate temporal correspondence between the ETC time-series information and other data. The number of records is aligned with the number of ETC data collections to ensure the integrity of the ETC time-series information.

[0088] Step S340: Convert the pixel coordinates of the image semantic information into spatial coordinates that are consistent with the radiation field data of the changed area through a preset transformation relationship; convert the spatial coordinates of the point cloud geometric information into a spatial coordinate system consistent with the radiation field data of the changed area; align the time stamp of the ETC time sequence information with the time stamp of the radiation field data of the changed area; align each spatial coordinate of each type of information with the corresponding coordinate of the radiation field data of the changed area; and synchronize the time stamp with the unified reference timestamp of the corresponding frame.

[0089] In one implementation, step S340 may specifically include the following steps S341-S346: Step S341: Extract the spatial coordinate system parameters and time stamping rules of the radiation field data in the changed area. The parameters and rules cover the entire spatial range of the radiation field data in the changed area and correspond to the unified reference timestamp of the original sensing data set.

[0090] Extracting the spatial coordinate system parameters and time stamping rules from the radiation field data of the changing region is fundamental for coordinate transformation and time alignment. Spatial coordinate system parameters include information such as the origin, axis directions, and coordinate precision; these parameters define the spatial coordinate system of the radiation field data of the changing region. Time stamping rules specify how to time-stamp the data, such as the format of the timestamp and the time unit.

[0091] These parameters and rules can be extracted by accessing the data structure or configuration file of the changed region radiation field data. It is crucial to ensure that the parameters and rules cover the entire spatial range of the changed region radiation field data and correspond to a unified reference timestamp of the original sensing data set, in order to facilitate accurate correlation and integration later.

[0092] Step S342: Convert the pixel coordinates of the image semantic information into spatial coordinates that are consistent with the radiation field data of the changed area through a preset conversion formula, so that the range of the converted coordinates is aligned with the spatial range of the radiation field data of the changed area, and the time stamp is synchronized with the unified reference timestamp of the original sensing data set.

[0093] In one implementation, step S342 may specifically include the following steps S3421-S3425: Step S3421: Extract the spatial coordinate system parameters of the radiation field data in the changed area. The parameters cover the entire spatial range of the radiation field data in the changed area and correspond to the unified reference timestamp of the original sensing data set.

[0094] Extracting the spatial coordinate system parameters of the changed region radiation field data again ensures that the pixel coordinates of the image semantic information can be accurately converted to a spatial coordinate system consistent with the changed region radiation field data during coordinate transformation. The parameters cover the entire spatial range of the changed region radiation field data, guaranteeing that the transformed coordinates accurately correspond throughout the entire changed region. Corresponding to the unified reference timestamp of the original sensory data set ensures consistency in the temporal dimension, guaranteeing that the transformed coordinate data can be accurately correlated with other data in time.

[0095] Step S3422: Convert the pixel coordinates of the image semantic information into spatial coordinates that are consistent with the radiation field data of the changed region using a preset conversion formula. The parameters of the conversion formula correspond to the spatial coordinate system parameters of the radiation field data of the changed region, and the range of the converted coordinates is aligned with the spatial range of the radiation field data of the changed region.

[0096] When converting the pixel coordinates of image semantic information into spatial coordinates that are consistent with the radiation field data of the changing region, a preset conversion formula based on camera calibration and coordinate transformation principles is required. Assuming the pixel coordinates of the image semantic information are (u,v), and the spatial coordinates to be converted to be consistent with the radiation field data of the changing region are (X,Y,Z), this process involves the camera's intrinsic and extrinsic parameters.

[0097] Camera intrinsic parameters can be represented by the intrinsic parameter matrix K: ; Among them, f x and f y ... t The configuration is used to describe the position and orientation of the camera in the world coordinate system.

[0098] During the transformation, firstly, the pixel coordinates (u,v) are converted to camera coordinates (x,v). c ,y c ,z c The formula is: ;z c This is the depth value in the camera coordinate system, which can be obtained through LiDAR data or other depth estimation methods. Next, the camera coordinates (x... c ,y c ,z c Converting coordinates from (X,Y,Z) to world coordinates (X,Y,Z) uses a formula like this: In practical applications, the parameters of the transformation formula should correspond to the spatial coordinate system parameters of the radiation field data in the changed region. At the same time, it should be ensured that the range of the transformed coordinates is consistent with the spatial range of the radiation field data in the changed region. This can be achieved by checking and adjusting the range of the transformed coordinates.

[0099] Step S3423: Verify the transformed spatial coordinates point by point, check whether the coordinate transformation results meet the preset requirements, and ensure that the verification results are aligned with the preset requirements of the radiation field data of the changed area.

[0100] Verifying the transformed spatial coordinates point by point ensures the accuracy and reliability of the coordinate transformation. Preset requirements may include the range, accuracy, and continuity of the coordinates. By checking the transformed spatial coordinates point by point, it is determined whether they meet these preset requirements. The verification results are aligned with the preset requirements of the radiation field data in the changed region, such as the range of radiation field values ​​and data integrity. If a coordinate transformation result is found to be inconsistent with the preset requirements, the transformation process is checked and corrected until the verification result meets the requirements.

[0101] Step S3424: Sort the verified image semantic information according to spatial coordinates. The sorting order corresponds to the coordinate sorting rules of the radiation field data of the changed region. The sorted information covers the entire spatial range of the radiation field data of the changed region.

[0102] Sort the verified image semantic information by spatial coordinates to ensure a consistent spatial order between the image semantic information and the radiation field data of the changed areas, facilitating subsequent association and integration operations. The sorting order corresponds to the coordinate sorting rules of the radiation field data of the changed areas, for example, sorting in ascending order according to the X-axis, Y-axis, and Z-axis. The sorted information should cover the entire spatial range of the radiation field data of the changed areas, ensuring that no image semantic information from any region is missed. Sorting allows for more efficient data searching and comparison.

[0103] Step S3425: Supplement the sorted image semantic information with a unified reference timestamp of the original sensing data set. The supplemented timestamp is synchronized with the corresponding frame timestamp of the image semantic information, so that the timestamp of each frame information corresponds to the unified reference timestamp of the original sensing data set. The image semantic information with the supplemented timestamp is used as the adjusted image semantic information for subsequent point-by-point alignment with the radiation field data of the changed area.

[0104] Supplementing the sorted image semantic information with a unified reference timestamp from the original perceptual data set ensures consistency with other data in the temporal dimension. The supplemented timestamp is synchronized with the corresponding frame timestamp of the image semantic information, ensuring that the timestamp of each frame corresponds to the unified reference timestamp of the original perceptual data set. This time stamp supplementation can be achieved by adding a timestamp field to the image semantic information data structure and assigning the unified reference timestamp to this field. The image semantic information with the supplemented timestamp is then used as the adjusted image semantic information for subsequent point-by-point alignment with the radiation field data of the changed region.

[0105] Step S343: Transform the spatial coordinates of the point cloud geometric information to a spatial coordinate system that is consistent with the radiation field data of the changed region, so that the transformed coordinate range is aligned with the spatial range of the radiation field data of the changed region, and the time stamp is synchronized with the unified reference timestamp of the original sensing data set.

[0106] Since point cloud geometry and changing region radiation field data may use different spatial coordinate systems, it is necessary to transform the spatial coordinates of the point cloud geometry to a unified spatial coordinate system with the changing region radiation field data. This can be achieved by calculating a transformation matrix between the two coordinate systems.

[0107] The transformed coordinate range is aligned with the spatial range of the radiation field data in the changed region to avoid coordinates exceeding the range or mismatches. The time stamp is synchronized with the unified reference timestamp of the original sensing data set to ensure that the point cloud geometric information is consistent with other data in time.

[0108] Step S344: Align the time stamp of the ETC time series information with the time stamp rules of the radiation field data of the changed area, so that the adjusted time stamp is synchronized with the unified reference timestamp of the original sensing data set, and the recorded content corresponds to the corresponding frame content of the radiation field data of the changed area.

[0109] Aligning the timestamps of ETC time-series information with the timestamp rules of the changed area radiation field data is to achieve consistency between the two in the time dimension. Based on the timestamp rules of the changed area radiation field data, the timestamps of the ETC time-series information are adjusted to synchronize with the unified reference timestamp of the original sensing data set. The recorded content corresponds to the corresponding frame content of the changed area radiation field data, ensuring that vehicle passage records in the ETC time-series information and vehicle information in the changed area radiation field data can be accurately correlated in time and space.

[0110] Step S345: Compare the adjusted image semantic information, point cloud geometric information, and ETC time series information with the radiation field data of the changed area coordinate by coordinate, check whether the information of each coordinate corresponds, and ensure that the comparison results are aligned with the preset requirements of the radiation field data of the changed area.

[0111] The adjusted image semantic information, point cloud geometric information, and ETC time-series information are compared coordinate-by-coordinate with the radiation field data of the changed area to verify the spatial and temporal consistency between the three types of information and the radiation field data of the changed area. The content of each piece of information is compared coordinate-by-coordinate to check for correspondence.

[0112] The comparison results are aligned with the preset requirements for the radiation field data of the changed area, such as the range of radiation field values, data completeness, and coordinate accuracy. If a mismatch is found in the information of a certain coordinate, the adjustment process is checked and corrected until the comparison results meet the requirements.

[0113] Step S346: Associate the image semantic information, point cloud geometric information, and ETC time series information that have passed the comparison with the time stamp of the radiation field data of the changed area, so that the time stamp of each type of information is synchronized with the corresponding frame time stamp of the radiation field data of the changed area, and complete the point-by-point alignment operation of the spatiotemporal dimension.

[0114] Associating the compared image semantic information, point cloud geometric information, and ETC time-series information with the time stamps of the changed area radiation field data is to further enhance the consistency between the three types of information and the changed area radiation field data in the time dimension. This synchronizes the time stamps of each type of information with the corresponding frame time stamps of the changed area radiation field data, ensuring accurate temporal correspondence. After completing the point-by-point alignment operation in the spatiotemporal dimension, the image semantic information, point cloud geometric information, and ETC time-series information are accurately associated and aligned with the changed area radiation field data in both space and time.

[0115] Step S350: Embed the aligned image semantic information coordinate by coordinate into the corresponding position of the changing region radiation field data. The semantic information of each coordinate covers the corresponding rendering content of the changing region radiation field data. Then embed the point cloud geometric information coordinate by coordinate to supplement the spatial structure support of the changing region radiation field data, so that the three-dimensional structure of the changing region radiation field data corresponds to the point cloud geometric information and the spatial range is aligned with the changing region radiation field data.

[0116] Embedding the aligned image semantic information coordinate-by-coordinate into the corresponding position of the changing region radiation field data is to integrate the detailed semantic content of the image semantic information into the changing region radiation field data. The semantic information of each coordinate covers the corresponding rendering content of the changing region radiation field data. For example, information such as the types of traffic participants identified in the image and the layout of road facilities are embedded into the corresponding position of the radiation field data, so that the radiation field data can more accurately reflect the actual traffic scene.

[0117] Then, point cloud geometric information is embedded coordinate by coordinate to supplement the spatial structure support of the radiation field data in the changing region. Point cloud geometric information provides accurate three-dimensional spatial structure information of the object. By embedding point cloud geometric information, the three-dimensional structure of the radiation field data in the changing region can correspond to the point cloud geometric information, thereby enhancing the spatial accuracy of the radiation field data.

[0118] After embedding, the spatial extent is aligned with the radiation field data of the changing region, ensuring that the semantic information of the image and the geometric information of the point cloud can completely cover the spatial extent of the radiation field data of the changing region, without any information omissions or exceeding the range.

[0119] Step S360: Embed the aligned ETC time series information into the corresponding frame of the changed area radiation field data by time marker. The time series information of each time marker corresponds to the corresponding frame content of the changed area radiation field data. Integrate the image semantic information, point cloud geometric information, ETC time series information and changed area radiation field data to obtain the radiation field data to be corrected. The time series is synchronized with the original sensing data set.

[0120] The aligned ETC time-series information is embedded, time-stamped, into the corresponding frames of the changing area radiation field data. This operation further integrates the ETC time-series information and the changing area radiation field data in the time dimension. Since the ETC time-series information records key information such as vehicle travel time and route, combining it with the changing area radiation field data of the corresponding frames adds dynamic time dimension information to the radiation field data, making the data more timely and practical.

[0121] During the embedding process, it is strictly ensured that the temporal information of each time stamp accurately corresponds to the content of the corresponding frame of the change area radiation field data. For example, if a vehicle passes through an ETC detection point at a certain time, its corresponding ETC temporal information (such as vehicle identification, passage time, route, etc.) will be accurately embedded into the change area radiation field data frame corresponding to that time point.

[0122] After embedding the ETC time-series information, image semantic information, point cloud geometric information, ETC time-series information, and radiation field data of changing areas are integrated. During the integration process, the consistency of various types of information in space and time is fully considered to ensure that different types of data complement and corroborate each other. For example, image semantic information provides a semantic description of the traffic scene, point cloud geometric information constructs the three-dimensional spatial structure of objects, and ETC time-series information records the temporal dynamic information of vehicles. Organically combining these elements can comprehensively and accurately reflect the actual situation of the traffic scene.

[0123] The final radiation field data to be corrected has a time series synchronized with the original sensing data set. This means that the radiation field data to be corrected is consistent with the initially collected multi-source data in time, facilitating subsequent unified analysis and processing based on the timeline of the original data. For example, when performing tasks such as traffic flow statistics and congestion analysis, the time series of the radiation field data to be corrected can be directly used in conjunction with the collection time of the original sensing data to accurately grasp the changing trend of traffic conditions over time.

[0124] Step S400: Fill the density gaps in the sparse point cloud regions of the radiation field data to be corrected with image semantic information, and calibrate the continuity of vehicle motion trajectories in the radiation field data to be corrected with ETC time series information to obtain complete radiation field data optimized by cross-type information complementarity.

[0125] The radiation field data to be corrected may exhibit sparse point cloud conditions in certain areas. This is because during the LiDAR scanning process, factors such as object occlusion and scanning angle can affect the density of point cloud data in some areas, making it impossible to accurately reflect the actual situation in those areas. Image semantic information, however, contains rich visual content, such as vehicle appearance and road texture, and can provide additional information to areas with sparse point clouds.

[0126] By analyzing relevant content in the semantic information of an image and matching it with the radiation field data to be corrected, the corresponding semantic information in the image is embedded into sparse areas of the point cloud, thereby filling density gaps in those areas. For example, a vehicle may be clearly visible in an image, but the point cloud data for that area may be sparse. In this case, based on the vehicle's location and appearance information in the image, the corresponding radiation field information can be added to the corresponding area in the radiation field data to be corrected, making the information for that area more complete.

[0127] On the other hand, the vehicle trajectories in the radiation field data to be corrected may be discontinuous, possibly due to data acquisition errors, object obstruction, or other reasons. ETC time-series information records the vehicle's travel time and path, possessing strong temporal continuity and accuracy. Using ETC time-series information, vehicle trajectories can be calibrated.

[0128] In one implementation, step S400 may specifically include the following steps S410-S460: Step S410: Perform spatial coordinate traversal frame by frame of the radiation field data to be corrected, count the distribution density of radiation field information in each coordinate region, mark the coordinate clusters in each frame whose density differs from the average distribution of the surrounding adjacent coordinate regions, and use them as the initial range of the sparse region of the point cloud. The coordinate accuracy is aligned with the accuracy of the radiation field data to be corrected, and the spatial range is aligned with the acquisition range of the radiation field data to be corrected.

[0129] During the traversal, for each coordinate region, the distribution density of radiation field information within it is statistically analyzed. This distribution density can be measured by calculating indicators such as the quantity and intensity of radiation field values ​​within that region. For example, the number of point clouds per unit volume can be counted, or the average radiation field intensity within that region can be calculated.

[0130] Then, the radiation field information distribution density of each coordinate region is compared with the average distribution of its neighboring coordinate regions. If the density of a coordinate region differs significantly from the average distribution of its neighboring regions (e.g., the density is lower than a certain threshold of the average density of the surrounding regions), then that coordinate region is marked. Consecutive coordinate regions with density differences are grouped into coordinate clusters, which constitute the initial extent of the sparse region of the point cloud.

[0131] During the labeling process, it is ensured that the coordinate accuracy is consistent with that of the radiation field data to be corrected. This means that the coordinate information of the labeled sparse regions of the point cloud has the same accuracy and resolution as the radiation field data to be corrected, avoiding errors in subsequent processing due to inconsistencies in accuracy. At the same time, the spatial extent of the sparse regions of the point cloud is aligned with the acquisition range of the radiation field data to be corrected, ensuring that the labeled areas are completely within the acquisition range of the original data, without any areas exceeding the range or being missed.

[0132] Step S420: Perform continuous frame verification on the coordinate clusters of the initial range, retain sparse coordinate clusters that appear in three or more consecutive frames, remove coordinate clusters that appear in only one or two frames, combine the retained coordinate clusters to form the final range of the sparse area of ​​the point cloud, align the boundary with the acquisition boundary of the original sensing data set, and synchronize the time stamp with the unified reference timestamp.

[0133] Performing continuous frame verification on the coordinate clusters within the initial range is to eliminate false sparsity in one or two frames due to data noise, random interference, or other factors. Coordinate clusters appearing in one or two frames may be temporary anomalies during data acquisition and do not represent true sparse point cloud regions. Sparse coordinate clusters appearing in three or more consecutive frames are more likely due to actual object occlusion, blind spots in LiDAR scanning, or other reasons, and therefore have higher reliability.

[0134] During the verification process, the coordinate clusters in the initial range are checked frame by frame to see if they appear continuously across multiple frames. Coordinate clusters that appear only in a single frame or two frames are removed from the initial range. Only sparse coordinate clusters that appear in three or more consecutive frames are retained.

[0135] The retained sparse coordinate clusters are combined to form the final extent of the sparse region of the point cloud. During the combination process, it is ensured that the boundary of this extent is aligned with the acquisition boundary of the original sensing data set. This is to guarantee that the final extent of the sparse region of the point cloud is completely within the acquisition range of the original data, avoiding unreasonable situations where it exceeds the boundary. At the same time, time information is marked for each coordinate cluster of the final extent, so that its time stamp is synchronized with a unified reference timestamp.

[0136] Step S430: Extract semantic information fragments from the image semantic information that correspond to the final range of the sparse region of the point cloud. Convert the pixel coordinates of the semantic information fragments into spatial coordinates that are consistent with the radiation field data to be corrected according to the preset conversion rules. The range of the converted spatial coordinates corresponds to the final range of the sparse region of the point cloud, and the semantic content corresponds to the original image semantic information.

[0137] In one implementation, step S430 may specifically include the following steps S431-S436: Step S431: Extract the spatial coordinate parameters of the final range of the sparse region of the point cloud. The parameters cover the entire space of the final range of the sparse region of the point cloud and correspond to the unified reference timestamp of the original sensing data set.

[0138] Extracting the spatial coordinate parameters of the final range of the sparse region of the point cloud is for coordinate transformation and information extraction. These spatial coordinate parameters include information such as the origin, coordinate axis direction, and coordinate precision, which define the specific location and range of the sparse region of the point cloud in three-dimensional space.

[0139] These parameters can be accurately extracted by accessing the data structure or related configuration information of the final extent of the sparse region in the point cloud. It is ensured that the parameters cover the entire space of the final extent of the sparse region, meaning that the parameters can fully describe the boundary and internal spatial information of the region. Simultaneously, these parameters correspond to a unified reference timestamp of the original sensing data set.

[0140] Step S432: Filter the pixel regions from the image semantic information that correspond to the spatial coordinate parameters of the final range of the sparse region of the point cloud. The filtered regions are aligned with the spatial range of the final range of the sparse region of the point cloud, and the semantic content corresponds to the semantic information of the original image.

[0141] Based on the spatial coordinate parameters of the final range of the sparse region of the point cloud, the corresponding pixel region is selected from the image semantic information. During the selection process, the spatial coordinates of the sparse region of the point cloud are converted into the pixel coordinate range in the image by using a pre-established mapping relationship between image pixel coordinates and three-dimensional spatial coordinates.

[0142] The selected pixel regions are strictly aligned with the final spatial extent of the sparse point cloud region. This means that the actual spatial location of the selected pixel region in the image is exactly the same as the location of the sparse point cloud region in 3D space. Simultaneously, it ensures that the selected region retains the complete semantic content of the original image's semantic information, such as the type of traffic participants and the layout of road facilities, without loss or alteration. For example, if the sparse point cloud region contains a car, then the selected pixel region should accurately contain the semantic description of that car in the image.

[0143] Step S433: Organize the semantic information of the filtered pixel regions into semantic information fragments. The pixel range of the fragments corresponds to the spatial range of the final range of the sparse region of the point cloud, and the time stamp is synchronized with the unified reference timestamp of the original perception data set.

[0144] The semantic information of the filtered pixel regions is organized into a complete semantic information fragment. During this process, it is ensured that the pixel range of the semantic information fragment precisely corresponds to the spatial range of the final sparse region of the point cloud. This means that each pixel in the semantic information fragment corresponds one-to-one with a sparse region of the point cloud, accurately reflecting the semantic situation of that region.

[0145] A time stamp is added to the semantic information fragments to synchronize them with the unified baseline timestamp of the original sensory data set. This time stamp synchronization ensures that the semantic information fragments are temporally consistent with the original data, facilitating subsequent data association and analysis along the time dimension. For example, in dynamic analysis of traffic scenarios, the semantic information fragments can be matched and fused with other data from the same time period based on the time stamp.

[0146] Step S434: Extract the spatial coordinate accuracy parameters of the radiation field data to be corrected. The parameters cover the entire spatial range of the radiation field data to be corrected and correspond to the unified reference timestamp of the original sensing data set.

[0147] Extracting the spatial coordinate accuracy parameters of the radiation field data to be corrected is to determine the accuracy and resolution of the data in spatial coordinate representation. These parameters include information such as the smallest unit of coordinates and the range of measurement error, which determine the spatial accuracy level of the radiation field data to be corrected.

[0148] These parameters are extracted by accessing the data structure or relevant configuration files of the radiation field data to be corrected. It is ensured that the parameters cover the entire spatial range of the radiation field data to be corrected, meaning that these parameters can describe the spatial accuracy of the entire radiation field data. Simultaneously, these parameters correspond to a unified reference timestamp of the original sensing data set, ensuring that the spatial coordinate accuracy of the radiation field data to be corrected is consistent with the original data in time, facilitating accurate matching and integration with other data in the future.

[0149] Step S435: Convert the pixel coordinates of the semantic information fragment into the spatial coordinates of the radiation field data to be corrected according to the preset conversion rules. The parameters of the conversion rules correspond to the spatial coordinate precision parameters of the radiation field data to be corrected, and the precision of the converted coordinates is aligned with the precision of the radiation field data to be corrected.

[0150] According to preset conversion rules, the pixel coordinates of semantic information fragments are converted into spatial coordinates of the radiation field data to be corrected. These preset conversion rules are based on camera calibration and coordinate transformation principles, enabling accurate conversion of two-dimensional image pixel coordinates into three-dimensional spatial coordinates. During the conversion process, it is ensured that the parameters of the conversion rules correspond to the spatial coordinate accuracy parameters of the radiation field data to be corrected. This means that the coordinate transformation formulas, measurement units, and other parameters used in the conversion process are consistent with the accuracy requirements of the radiation field data to be corrected. For example, if the coordinate accuracy of the radiation field data to be corrected is at the centimeter level, then the conversion rules should also ensure that the converted spatial coordinate accuracy reaches the centimeter level.

[0151] The accuracy of the converted coordinates is aligned with the accuracy of the radiation field data to be corrected. This ensures that the semantic information fragments, after being converted into spatial coordinates, can be accurately matched with the radiation field data to be corrected in space, and will not cause information misalignment or inaccuracy due to differences in accuracy.

[0152] Step S436: Perform coordinate-by-coordinate verification on the converted semantic information fragments, check whether the semantic content of each coordinate corresponds to the semantic information of the original image, and ensure that the verification results are aligned with the preset requirements of the radiation field data to be corrected.

[0153] During the verification process, the semantic content of each coordinate in the transformed semantic information fragment is checked coordinate by coordinate to ensure consistency with the semantic information of the original image. For example, the semantic description at a certain coordinate (such as traffic participant type, road facility attributes, etc.) is checked to ensure it matches the description of that location in the original image. The verification results are aligned with the preset requirements of the radiation field data to be corrected. These preset requirements may include standards for the accuracy, completeness, and consistency of the semantic content. If the semantic content of a coordinate is found to be inconsistent with the preset requirements, the transformation process needs to be checked and corrected to ensure that the semantic content of all coordinates is accurate. Through coordinate-by-coordinate verification, it is ensured that the transformed semantic information fragment can be reliably used to fill density gaps in sparse point cloud regions of the radiation field data to be corrected.

[0154] Step S440: Embed the converted semantic information fragments coordinate by coordinate into the corresponding positions of the final range of the sparse region of the point cloud in the radiation field data to be corrected. The semantic information of each coordinate supplements the corresponding missing content of the radiation field data to be corrected, so that the radiation field information distribution density of the region is aligned with the average distribution of the surrounding adjacent coordinate regions, and the spatial range is aligned with the final range of the sparse region of the point cloud.

[0155] In one implementation, step S440 may specifically include the following steps S441-S446: Step S441: Align the transformed semantic information fragments with the spatial coordinates of the final range of the sparse region of the point cloud point cloud point by point, so that the semantic information of each coordinate corresponds to the corresponding coordinate of the final range of the sparse region of the point cloud point cloud point, and the alignment accuracy is aligned with the accuracy of the radiation field data to be corrected.

[0156] Aligning the transformed semantic information fragments point-by-point with the spatial coordinates of the final extent of the sparse region in the point cloud is a preliminary step in the embedding operation, ensuring that the semantic information can be accurately embedded into the target location. During the alignment process, each coordinate of the semantic information fragment is carefully compared with the corresponding coordinates of the final extent of the sparse region in the point cloud.

[0157] This ensures that the semantic information of each coordinate precisely corresponds to the coordinates of the final extent of the sparse region in the point cloud. For example, the semantic description at a certain coordinate in a semantic information segment (such as the location and type of a vehicle) is accurately matched with the radiation field data at the same coordinate location in the sparse region of the point cloud.

[0158] The alignment accuracy must be aligned with the accuracy of the radiation field data to be corrected. This means that during the alignment process, the matching accuracy of the coordinates must meet the accuracy requirements of the radiation field data to be corrected. For example, if the coordinate accuracy of the radiation field data to be corrected is at the centimeter level, then during alignment, the matching error between the coordinates of the semantic information fragment and the coordinates of the sparse region of the point cloud should be controlled within the centimeter level to avoid inaccurate embedding positions due to inconsistent accuracy.

[0159] Step S442: Input the aligned semantic information fragments coordinate by coordinate into the final range of the sparse region of the point cloud of the radiation field data to be corrected. The semantic information of each coordinate replaces the missing content of the corresponding coordinate in the radiation field data to be corrected. The replaced content corresponds to the distribution of radiation field information in the surrounding adjacent coordinate regions.

[0160] The aligned semantic information fragments are input coordinate-by-coordinate into the final range of the sparse region of the point cloud in the radiation field data to be corrected. During the input process, the semantic information of each coordinate replaces the missing content of the corresponding coordinate in the radiation field data to be corrected. For example, if the radiation field information at a certain coordinate in the sparse region of the point cloud is incomplete or missing, but the semantic information fragment has a complete semantic description of the corresponding position, then the original missing content is replaced with that semantic information.

[0161] The replaced content corresponds to the radiation field information distribution of the surrounding coordinate regions. This requires that the radiation field characteristics of the surrounding areas be considered during the replacement process to ensure that the newly embedded semantic information is consistent with the surrounding environment. For example, if the radiation field information of the surrounding areas mainly describes the vehicle's driving state, then the replaced content should also conform to the semantic logic related to vehicle driving to avoid information conflicts or inconsistencies.

[0162] Step S443: Perform coordinate-by-coordinate density statistics on the final range of the sparse region of the replaced point cloud, and calculate the radiation field information distribution density of each coordinate. The statistical results correspond to the average distribution of the surrounding adjacent coordinate regions.

[0163] Performing coordinate-by-coordinate density statistics on the final extent of the sparse region in the replaced point cloud is to evaluate the distribution of radiation field information in that region after embedding semantic information. During the statistics process, the radiation field information distribution density is calculated for each coordinate. The radiation field information distribution density can be measured in various ways, such as the number of radiation field values ​​per unit volume or the average value of the radiation field intensity.

[0164] The statistical results are compared with the average distribution of neighboring coordinate regions. By comparison, it can be determined whether the density of the sparse region in the replaced point cloud is balanced with the surrounding region. If the statistical results show that the density of the region differs significantly from the average distribution of the surrounding region, it indicates that the embedded semantic information may need further adjustment to achieve a better information balance.

[0165] Step S444: Compare the statistical results with the average distribution of the surrounding adjacent coordinate regions, adjust the embedding position of the semantic information fragments, and align the radiation field information distribution density of the final range of the sparse point cloud region with the average distribution of the surrounding adjacent coordinate regions.

[0166] The distribution density of the radiation field information in the sparse region of the point cloud obtained by statistics is compared with the average distribution of the surrounding coordinate regions. If a difference is found, the embedding position of the semantic information fragment needs to be adjusted.

[0167] The purpose of adjusting the embedding position is to align the radiation field information distribution density of the final range of the sparse point cloud region with the average distribution of the surrounding adjacent coordinate regions. For example, if statistical results show that the density of a certain local area in the sparse point cloud region is still low, more semantic information can be embedded into that area; if the density of a certain area is too high, the amount of semantic information embedded in that area can be appropriately reduced or the content of the embedded semantic information can be adjusted. By continuously adjusting the embedding position, the radiation field information distribution of the sparse point cloud region is gradually optimized, making it more consistent with the surrounding areas and improving the quality and accuracy of the entire radiation field data.

[0168] Step S445: Perform frame-by-frame verification on the final range of the adjusted sparse region of the point cloud, check whether the radiation field information of each frame corresponds to the preset requirements of the radiation field data to be corrected, and verify that the verification results correspond to the content of the original sensing data set.

[0169] The final extent of the adjusted sparse region of the point cloud is verified frame by frame to ensure that, after embedding and adjustment, the radiation field information of that region meets the preset requirements of the radiation field data to be corrected in each frame of data. The preset requirements may include standards such as the range of radiation field values, the accuracy of semantic content, and the completeness of information.

[0170] During frame-by-frame verification, the radiation field information of each frame is carefully checked to ensure it meets these preset requirements. For example, it checks whether the radiation field value is within a reasonable range, whether the semantic description is accurate, and whether the information is complete and without omissions.

[0171] The verification results correspond to the content of the original sensing data set. This means that the original sensing data set must be referenced during the verification process to ensure that the radiation field information of the adjusted sparse area of ​​the point cloud is consistent with the content of the original data. For example, if there is no vehicle information in a certain area of ​​the original sensing data set, but a semantic description of a vehicle suddenly appears in the adjusted sparse area of ​​the point cloud without reasonable basis, then the information in that area will be further checked and corrected.

[0172] Step S446: Use the verified radiation field data to be corrected as the radiation field data to be corrected after filling the density gap, for subsequent calibration operations on the continuity of vehicle motion trajectory.

[0173] After frame-by-frame verification and confirming that the results meet the requirements, the radiation field data to be corrected is used as the corrected radiation field data after filling the density gaps. At this point, the distribution density of radiation field information in the sparse areas of the point cloud is effectively improved, and it is basically aligned with the average distribution of the surrounding adjacent coordinate areas, thus improving the integrity and accuracy of the data.

[0174] The corrected radiation field data, after filling in density gaps, is used for subsequent calibration of vehicle trajectory continuity. Because information from sparse point cloud regions is supplemented, more accurate and complete baseline data is provided for subsequent vehicle trajectory calibration, facilitating more precise analysis and repair of trajectory continuity issues.

[0175] Step S450: Traverse the vehicle motion trajectory frame by frame in the radiation field data to be corrected, mark the coordinate clusters in each frame where the trajectory is interrupted, and use them as the initial range of the trajectory interruption area. The coordinate accuracy is aligned with the accuracy of the radiation field data to be corrected, and the time stamp is synchronized with the unified reference timestamp.

[0176] The purpose of traversing the vehicle trajectory frame by frame in the data to be corrected for the radiation field is to comprehensively check the continuity of the vehicle trajectory in each frame. During the traversal, the vehicle trajectory in each frame is carefully analyzed to determine if there are any interruptions.

[0177] Each frame marks coordinate clusters where the trajectory is interrupted. When a discontinuity is detected in the vehicle's trajectory in a certain area, i.e., the trajectory is suddenly interrupted or jumps, the coordinates of that area are marked, and consecutive interrupted coordinates are combined into coordinate clusters. These coordinate clusters constitute the initial range of the trajectory interruption area.

[0178] The coordinate accuracy is aligned with the accuracy of the radiation field data to be corrected. This ensures that the coordinate information of the marked trajectory interruption region has the same accuracy and resolution as the radiation field data to be corrected. For example, if the coordinate accuracy of the radiation field data to be corrected is at the centimeter level, then the coordinate accuracy of the marked trajectory interruption region should also reach the centimeter level to avoid affecting subsequent calibration operations due to inconsistencies in accuracy.

[0179] The timestamps are synchronized with the unified reference timestamp. A timestamp consistent with the unified reference timestamp is added to the coordinate clusters of each trajectory interruption area. In subsequent analysis, the exact time of the trajectory interruption event can be accurately determined, facilitating comprehensive analysis and calibration in conjunction with other time-related data (such as ETC time series information).

[0180] Step S460: Extract the time sequence information segment corresponding to the initial range of the trajectory interruption area from the ETC time sequence information, synchronize the time stamp of the time sequence information segment with the corresponding frame time stamp of the radiation field data to be corrected, embed the time sequence information segment into the corresponding position of the initial range of the trajectory interruption area of ​​the radiation field data to be corrected, connect the vehicle motion trajectory in the radiation field data to be corrected, calibrate the continuity of the vehicle motion trajectory in the radiation field data to be corrected, and obtain the complete radiation field data after cross-type information complementarity optimization.

[0181] Extract time-series information segments corresponding to the initial range of the trajectory interruption area from the ETC time-series information. Since the ETC time-series information records information such as the vehicle's travel time and route, the corresponding part can be filtered out from the ETC time-series information using the coordinates and time information of the trajectory interruption area. For example, if the trajectory interruption area involves the interruption of a vehicle's movement trajectory at a certain time, then the vehicle's travel records at similar times can be extracted from the ETC time-series information to form a time-series information segment.

[0182] Synchronize the timestamps of the timing information segments with the corresponding frame timestamps of the radiation field data to be corrected. Ensure that the time information in the timing information segments accurately matches the frame time corresponding to the trajectory interruption region in the radiation field data to be corrected. For example, if the trajectory interruption region appears in the radiation field data frame to be corrected at a certain set time, then the timestamp of the timing information segment should also be adjusted to that time to ensure consistency between the two in time.

[0183] The time-series information segments are embedded into the corresponding positions of the initial range of the trajectory interruption region in the radiation field data to be corrected. During the embedding process, based on the vehicle passage information (such as location and time) in the time-series information segments, they are accurately inserted into the positions where the trajectory is interrupted in the radiation field data to be corrected. By embedding the time-series information segments, the interrupted vehicle movement trajectories are connected, restoring their continuity.

[0184] After the above operations, the calibration of the continuity of vehicle motion trajectories in the radiation field data to be corrected was completed. Combined with the previous operation of filling density gaps in sparse areas of the point cloud using image semantic information, complementary optimization across different information types was achieved, ultimately resulting in complete radiation field data optimized by complementary optimization across different information types.

[0185] Step S500: Reconstruct the three-dimensional scene based on the complete radiation field data, and output real-time holographic traffic flow data that includes the three-dimensional shape of traffic participants, spatial location correlation and continuous movement trajectory.

[0186] After processing and optimization through the preceding steps, the complete radiation field data possesses the capability to comprehensively and accurately describe traffic scenarios. Reconstructing a 3D scene based on this data is the process of transforming radiation field information into an intuitive 3D visualization, aiming to provide a clearer and more realistic representation of traffic flow for traffic management and analysis.

[0187] In the process of 3D scene reconstruction, various types of information in the complete radiation field data are fully utilized. The 3D morphological information of traffic participants can be extracted and reconstructed from point cloud geometric information and image semantic information. Point cloud geometric information provides the accurate 3D spatial structure of objects, and by processing and analyzing it, 3D models of traffic participants (such as vehicles and pedestrians) can be constructed. Image semantic information can add details and textures to these models, making them more realistic.

[0188] Spatial location correlation information is determined based on the coordinate information of each object in the complete radiation field data. Precise coordinate positioning clarifies the spatial relationships between traffic participants and with road infrastructure. For example, it determines the specific lane position of a vehicle on the road and its distance from other vehicles.

[0189] Continuous motion trajectory information comes from the calibration and optimization of the vehicle's motion trajectory. After calibration of the ETC timing information, the vehicle's motion trajectory is more continuous and accurate, clearly showing the vehicle's driving path and speed changes over a period of time.

[0190] The final output of real-time holographic traffic flow data includes key information such as the three-dimensional morphology, spatial location correlation, and continuous movement trajectories of traffic participants. Presented in a three-dimensional visualization, this data allows traffic managers to intuitively understand the real-time status of traffic scenarios, including vehicle direction, speed, density, and pedestrian activity trajectories. Furthermore, real-time holographic traffic flow data can also be used for traffic simulation, prediction, and decision support, providing strong support for the efficient operation of intelligent transportation systems.

[0191] In one implementation, step S500 may specifically include the following steps S510-S560: Step S510: Perform entity recognition on each frame of the complete radiation field data, identify traffic participants, road facilities and vehicle trajectories in each frame, extract the three-dimensional morphological information of traffic participants, the spatial layout information of road facilities and the continuous sequence information of vehicle trajectories, and align the information accuracy with the accuracy of the complete radiation field data and the spatial range with the collection range of the complete radiation field data.

[0192] For traffic participant identification, point cloud geometric information and image semantic information are combined. Point cloud geometric information provides the three-dimensional spatial structure of objects, and by clustering and analyzing it, different traffic participants (such as vehicles, pedestrians, bicycles, etc.) can be distinguished. Image semantic information can further determine details such as the type and posture of traffic participants. For example, image recognition algorithms can identify the brand and color of vehicles, and the walking direction of pedestrians.

[0193] The identification of road facilities mainly relies on relevant features in point cloud geometric information and image semantic information. Point cloud data can accurately depict the shape and location of road facilities (such as streetlights, traffic signs, guardrails, etc.), while image semantic information can provide specific types and functional descriptions of the facilities.

[0194] Vehicle trajectory identification is achieved through continuous tracking and analysis of vehicle positions within complete radiation field data. By combining previously calibrated vehicle trajectory information, the vehicle's position and direction of motion in each frame are accurately extracted, forming a continuous sequence of trajectories.

[0195] The process extracts three-dimensional morphological information of traffic participants, spatial layout information of road facilities, and continuous sequence information of vehicle movement trajectories. During extraction, it ensures that the accuracy of the information matches that of the complete radiation field data. For example, if the coordinate accuracy of the complete radiation field data is at the centimeter level, the extracted information should maintain the same accuracy. Simultaneously, the spatial extent is aligned with the acquisition range of the complete radiation field data to ensure that the extracted information covers the entire effective area of ​​the traffic scene, preventing information omissions or exceeding the specified range.

[0196] Step S520: Link the three-dimensional morphological information of traffic participants, the spatial layout information of road facilities, and the continuous sequence information of vehicle movement trajectories point by point according to spatial coordinates, so that the three-dimensional morphological information of each traffic participant is linked with the road facility information of the corresponding spatial location, the continuous sequence information of vehicle movement trajectories is linked with the three-dimensional morphological information of the corresponding traffic participant, and the time stamp is synchronized with the unified reference timestamp.

[0197] Associating the 3D morphological information of traffic participants, the spatial layout information of road facilities, and the continuous sequence information of vehicle movement trajectories point by point according to spatial coordinates is to construct a complete and coherent traffic scene model. This links the 3D morphological information of each traffic participant with the road facility information at the corresponding spatial location. For example, when a car is driving on a road, its 3D model is associated with the road facilities at that location (such as lane markings, traffic signs, etc.). Through this association, the specific locations and relationships of traffic participants in the road environment can be clearly understood.

[0198] In one implementation, step S520 includes the following steps S521-S526: Step S521: Extract the spatial coordinate parameters of the three-dimensional morphological information of traffic participants. The parameters cover the entire space of the three-dimensional morphological information of each traffic participant and correspond to the unified reference timestamp of the original sensing data set.

[0199] Extracting the spatial coordinate parameters of the 3D morphological information of traffic participants is fundamental for information association. These parameters include the coordinate origin, coordinate axis direction, and coordinate precision, which define the specific location and range of the traffic participant's 3D morphological information in 3D space. These parameters are accurately extracted by accessing the data structure or related configuration information of the traffic participant's 3D morphological information. It is ensured that the parameters cover the entire space of each traffic participant's 3D morphological information, that is, they can completely describe the position and size of the traffic participant's 3D model in space. Simultaneously, these parameters correspond to a unified reference timestamp of the original sensing data set, ensuring temporal consistency with the original data and facilitating subsequent association and integration with other data.

[0200] Step S522: Extract the spatial coordinate parameters of the spatial layout information of road facilities. The parameters cover the entire space of the spatial layout information of road facilities and correspond to the unified reference timestamp of the original sensing data set.

[0201] Similarly, spatial coordinate parameters of the spatial layout information of road facilities are extracted. These parameters are used to determine the location and extent of road facilities in three-dimensional space. Accurate spatial coordinate parameters are obtained by analyzing and processing the point cloud geometric information and image semantic information of road facilities. It is ensured that the parameters cover the entire spatial layout information of road facilities, that is, they can completely describe the distribution of road facilities (such as streetlights, traffic signs, guardrails, etc.) in the entire traffic scene. Corresponding to a unified reference timestamp of the original sensing data set, the spatial layout information of road facilities is ensured to be consistent with other data in time, facilitating accurate correlation and analysis.

[0202] Step S523: Compare the spatial coordinate parameters of the three-dimensional morphological information of traffic participants with the spatial coordinate parameters of the spatial layout information of road facilities point by point to find the spatial location of the road facility information corresponding to the three-dimensional morphological information of each traffic participant, and the comparison results correspond to the content of the complete radiation field data.

[0203] In one implementation, step S523 may specifically include the following steps S5231-S5236: Step S5231: Group the spatial coordinate parameters of the three-dimensional morphological information of traffic participants according to time stamps. Each group corresponds to a frame of the unified reference timestamp of the original sensing data set. The grouped coordinate parameters cover the entire space of the three-dimensional morphological information of traffic participants in each frame.

[0204] Because traffic scenarios are dynamic, the positions and states of traffic participants may differ at different times. Grouping by time stamp allows for centralized processing of traffic participant information at the same moment. Each group corresponds to a frame with a unified reference timestamp from the original sensing data set. This means that each group contains the spatial coordinate parameters of the three-dimensional morphological information of traffic participants at a specific moment. The coordinate parameters after grouping cover the entire space of the three-dimensional morphological information of traffic participants in each frame, thus fully describing the position and size of all traffic participants in space at that moment.

[0205] Step S5232: Group the spatial coordinate parameters of the spatial layout information of road facilities according to time stamps. Each group corresponds to a frame of the unified reference timestamp of the original sensing data set. The grouped coordinate parameters cover the entire space of the spatial layout information of road facilities in each frame.

[0206] Similarly, the spatial coordinate parameters of the road facility spatial layout information are grouped by time stamp. The purpose of this grouping is similar to that of the traffic participant grouping: to effectively manage and analyze road facility information over time. Each group corresponds to a frame with a unified reference timestamp from the original sensing data set, ensuring that each group contains the spatial coordinate parameters of the road facility spatial layout information at a specific moment. The grouped coordinate parameters cover the entire spatial layout information of the road facilities in each frame, thus providing a complete description of the spatial distribution of all road facilities at that moment.

[0207] Step S5233: Compare the spatial coordinate parameters of the three-dimensional morphological information of traffic participants at the same time marker with the spatial coordinate parameters of the spatial layout information of road facilities point by point, find the overlapping spatial coordinates of the three-dimensional morphological information of each traffic participant and the information of road facilities, and the comparison results correspond to the content of the complete radiation field data.

[0208] During the comparison process, the coordinate parameters of each traffic participant are compared one by one with the coordinate parameters of road facilities. The overlapping spatial coordinates between the three-dimensional morphological information of each traffic participant and the road facility information are identified. For example, it determines which road facilities (such as lane lines, traffic signs, etc.) have overlapping or adjacent coordinate ranges with the coordinate range of a car.

[0209] The comparison results must correspond to the content of the complete radiation field data. This requires ensuring that the found overlap relationships are consistent with the actual information recorded in the complete radiation field data during the comparison process. For example, if the complete radiation field data shows that a vehicle is adjacent to a traffic sign at a specific location, the comparison results should accurately reflect this relationship to avoid information bias or errors.

[0210] Step S5234: Mark the road facility information corresponding to the overlapping spatial coordinates as the road facility information corresponding to the three-dimensional morphological information of the traffic participant. The marked information covers the entire spatial range monitored by the RSU, and the time stamp is synchronized with the unified reference timestamp of the original sensing data set.

[0211] The road facility information corresponding to the overlapping spatial coordinates is marked as the road facility information corresponding to the three-dimensional morphological information of the traffic participant. This marking method clarifies the road facility information corresponding to each traffic participant in space, facilitating subsequent association and analysis.

[0212] The tagged information covers the entire spatial range of RSU monitoring, ensuring that every traffic participant has accurate corresponding road facility information throughout the entire monitoring area. The timestamp is synchronized with the unified reference timestamp of the original sensing data set, ensuring that the tagged information is consistent with other data in time, facilitating dynamic traffic scenario analysis.

[0213] Step S5235: Perform frame-by-frame verification on the marked information to check whether the road facility information corresponding to the three-dimensional morphological information of each traffic participant is correct, and the verification result corresponds to the content of the complete radiation field data.

[0214] Frame-by-frame verification of the marked information is performed to ensure the accuracy and reliability of the markings. During the verification process, the road infrastructure information corresponding to the three-dimensional morphological information of each traffic participant is checked frame by frame to ensure that it matches the actual situation.

[0215] Check whether the road infrastructure information corresponding to the three-dimensional morphological information of each traffic participant is correct. For example, check whether the marked road infrastructure information (such as lane lines, traffic signs, etc.) matches the actual location and driving status of the traffic participants.

[0216] The verification results correspond to the content of the complete radiation field data. This requires that, during the verification process, the complete radiation field data be used as a reference to ensure that the labeling information is consistent with the original data. If errors or deviations are found in the labeling information, they should be corrected promptly to ensure the accuracy of the associated information.

[0217] Step S5236: Use the verified association information as the association result between the three-dimensional morphological information of traffic participants and the road facility information, and use it for subsequent association operations with the continuous sequence information of vehicle movement trajectories.

[0218] Step S524: Associate the three-dimensional morphological information of traffic participants with the road facility information of the corresponding spatial location. The associated information covers the entire spatial range of RSU monitoring, and the time stamp is synchronized with the unified reference timestamp of the original sensing data set.

[0219] Linking the three-dimensional morphological information of traffic participants with the corresponding spatial location information of road facilities is a further integration and refinement of the spatial information of the traffic scene, building upon the previous steps. Through accurate association, the specific locations and relationships of traffic participants within the road environment are clearly defined.

[0220] The correlated information covers the entire spatial range of RSU monitoring, ensuring that every traffic participant has corresponding road facility information within the entire monitoring area, without omissions or gaps. The timestamp is synchronized with the unified reference timestamp of the original sensing data set, ensuring that the correlated information is consistent with other data in time, facilitating dynamic traffic scenario analysis.

[0221] Step S525: Extract the time stamp and spatial coordinate parameters of the continuous sequence information of the vehicle motion trajectory. The parameters cover the entire time and space range of the continuous sequence information of the vehicle motion trajectory and correspond to the unified reference timestamp of the original sensing data set.

[0222] Extracting the time stamp and spatial coordinate parameters of the continuous sequence information of the vehicle's motion trajectory is crucial for accurately describing the vehicle's movement in time and space. The time stamp records the vehicle's position information at different moments, while the spatial coordinate parameters determine the vehicle's specific location in three-dimensional space. These parameters cover the entire temporal and spatial range of the continuous sequence information of the vehicle's motion trajectory, thus providing a complete description of the vehicle's temporal and spatial changes throughout its entire motion. Corresponding to a unified reference timestamp in the original sensing data set ensures that the vehicle's motion trajectory information is temporally consistent with other data, facilitating accurate correlation and analysis.

[0223] Step S526: Compare the spatial coordinate parameters of the continuous sequence information of vehicle motion trajectory with the spatial coordinate parameters of the three-dimensional morphological information of traffic participants point by point, find the three-dimensional morphological information of traffic participants corresponding to the continuous sequence information of each vehicle motion trajectory, associate the two types of information, and synchronize the associated time stamp with the unified reference timestamp.

[0224] The purpose of comparing the spatial coordinate parameters of the continuous sequence information of vehicle movement trajectories with the spatial coordinate parameters of the three-dimensional morphological information of traffic participants point by point is to determine the specific traffic participant corresponding to each vehicle movement trajectory. During the comparison process, the coordinate information of the two is precisely compared to find the matching relationship between them.

[0225] Find the 3D morphological information of traffic participants corresponding to the continuous sequence information of each vehicle's trajectory. For example, determine which specific car or other traffic participant a particular vehicle's trajectory belongs to. Link these two types of information to form a complete set of traffic participant movement information. The linked timestamp is synchronized with a unified reference timestamp to ensure that the linked information is consistent with other data in time.

[0226] Step S530: Arrange the associated information in an orderly manner according to spatial coordinates to obtain a three-dimensional scene frame covering the entire spatial range of RSU monitoring. The spatial range of the three-dimensional scene frame is aligned with the acquisition range of the complete radiation field data, and the time stamp is synchronized with the unified reference timestamp.

[0227] Based on the spatial coordinate order, the 3D morphological information of traffic participants, road facility information, and vehicle trajectory information are reorganized and sorted. This results in a 3D scene framework covering the entire spatial range of the RSU monitoring system. This framework can completely present the traffic scene within the entire monitoring area, including the distribution of traffic participants, the layout of road facilities, and vehicle trajectories. The spatial range of the 3D scene framework is aligned with the acquisition range of the complete radiation field data, ensuring that the spatial area covered by the framework is consistent with the acquisition range of the original data, and preventing information omissions or exceeding the range. The time stamp is synchronized with a unified reference timestamp, ensuring that the 3D scene framework is consistent with other data in time, facilitating dynamic traffic scene analysis.

[0228] Step S540: Embed the three-dimensional morphological information of traffic participants into the corresponding spatial position of the three-dimensional scene frame. Align the three-dimensional morphological information of each traffic participant with the corresponding spatial coordinates of the three-dimensional scene frame, so that the spatial position of the three-dimensional morphological information corresponds to the spatial layout information of the road facilities, and the time stamp is synchronized with the unified reference timestamp of the corresponding frame.

[0229] Embedding the 3D morphological information of traffic participants into their corresponding spatial locations within a 3D scene framework is the process of placing specific models of traffic participants into the 3D scene. During the embedding process, it is crucial to ensure that the 3D morphological information of each traffic participant is precisely aligned with its corresponding spatial coordinates within the 3D scene framework.

[0230] This ensures that the spatial location of three-dimensional morphological information corresponds to the spatial layout information of road facilities. For example, the three-dimensional model of a vehicle is accurately placed on the corresponding lane, matching the spatial location of lane lines, traffic signs, and other road facilities.

[0231] The timestamps are synchronized with the unified reference timestamps of the corresponding frames, ensuring that the 3D morphological information of traffic participants is consistent with other data in time, facilitating dynamic traffic scene analysis. Embedding operations enrich and enhance the 3D scene framework, enabling a more intuitive display of the actual situation of traffic participants in the road environment.

[0232] Step S550: Embed the continuous sequence information of vehicle motion trajectory into the corresponding time frame of the three-dimensional scene frame. The vehicle motion trajectory information of each time frame is associated with the three-dimensional morphological information of the corresponding traffic participant, so that the continuous sequence information of vehicle motion trajectory is synchronized with the time mark of the three-dimensional scene frame and the spatial range is aligned with the acquisition range of the complete radiation field data.

[0233] During the embedding process, the vehicle's motion trajectory is accurately inserted into the corresponding time frame position within the 3D scene framework based on its time stamp. The vehicle's motion trajectory information for each time frame is associated with the 3D morphological information of the corresponding traffic participant. For example, the motion trajectory information of a vehicle at a certain time is bound to the vehicle's 3D model, ensuring that the vehicle's motion trajectory accurately reflects its actual driving situation in the 3D scene.

[0234] Synchronizing the continuous sequence information of vehicle motion trajectories with the time stamp of the 3D scene frame ensures that the vehicle motion trajectories are consistent with the 3D scene frame in time, facilitating dynamic traffic scene analysis. Aligning the spatial extent with the acquisition range of the complete radiation field data ensures that the vehicle motion trajectory information is spatially consistent with the acquisition range of the original data, preventing information omissions or exceeding the range.

[0235] Step S560: Integrate the three-dimensional scene framework, the three-dimensional morphological information of traffic participants, the spatial layout information of road facilities, and the continuous sequence information of vehicle movement trajectories to obtain real-time holographic traffic flow data. The output format is aligned with the preset output requirements, and the information content corresponds to the complete radiation field data.

[0236] The system integrates the three-dimensional scene framework, the three-dimensional morphological information of traffic participants, the spatial layout information of road facilities, and the continuous sequence information of vehicle movement trajectories, thus fusing the information from each part into a unified whole to form the final real-time holographic traffic flow data.

[0237] During the integration process, the consistency and coordination of information from various parts in space and time are crucial. For example, the three-dimensional morphological information of traffic participants is spatially matched with the spatial layout information of road facilities and the vehicle movement trajectory information, and is updated synchronously in time.

[0238] The final real-time holographic traffic flow data output format is aligned with preset output requirements. These requirements may include data format (e.g., XML, JSON), resolution, frame rate, etc. The information content corresponds to the complete radiation field data, ensuring that the output data accurately reflects the actual traffic situation and provides a reliable basis for traffic management and analysis. By outputting real-time holographic traffic flow data, more intuitive and accurate information support can be provided for the decision-making and control of intelligent transportation systems.

[0239] In one embodiment, an edge computing system is provided, the internal structure of which can be as follows: Figure 3As shown, this edge computing system includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements an edge computing-based RSU holographic traffic flow analysis method.

[0240] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the edge computing system to which the present invention is applied. A specific edge computing system may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

Claims

1. An edge computing based RSU holographic traffic flow analysis method, characterized in that, The method includes: The RSU uses a unified clock trigger and interface coordination to control the visible light acquisition device, lidar device and ETC reader / writer to synchronously acquire data, and obtain a raw sensing data set containing multiple types of sensing information. Each type of data in the raw sensing data set carries a unified reference timestamp. Based on the radiation field data of historical stable regions reused by the edge side neural radiation field, the traffic flow change regions in the original sensing data set are processed to obtain change region radiation field data containing only change region information. Image semantic information, point cloud geometric information, and ETC time series information are extracted from the original sensing data set. The three types of information are then integrated with the radiation field data of the changed area in a spatiotemporal dimension to obtain the radiation field data to be corrected. The point cloud geometric information provides three-dimensional spatial structure support for the radiation field data to be corrected. The image semantic information is used to fill the density gaps in the sparse point cloud regions of the radiation field data to be corrected, and the ETC time series information is used to calibrate the continuity of vehicle movement trajectories in the radiation field data to be corrected, so as to obtain complete radiation field data optimized by cross-type information complementarity. Based on the complete radiation field data, a three-dimensional scene is reconstructed, and real-time holographic traffic flow data containing the three-dimensional shape, spatial location correlation and continuous movement trajectory of traffic participants is output.

2. The method of claim 1, wherein, The radiation field data based on the historical stable region of the lateral neural radiation field reuse is used to process the traffic flow change region in the original sensing data set to obtain change region radiation field data containing only change region information, including: Retrieve continuous historical radiation field data from the edge-side storage medium, covering the entire spatial range monitored by RSU. Extract the spatial coordinate information of each frame frame by frame, compare the inter-frame information with the same spatial coordinates point by point, mark the coordinate clusters that have not changed in continuous periods, combine the radiation field information corresponding to the coordinate clusters, and obtain the radiation field data of the historical stable area. The radiation field data of the historical stable region is connected point by point with the neural radiation field rendering space currently running on the edge side. The radiation field information corresponding to each coordinate is embedded into the corresponding position of the rendering space to obtain the baseline radiation field base layer covering the entire spatial range of RSU monitoring. The time stamp is synchronized with the unified baseline timestamp start frame of the original sensing data set. For the visible light acquisition data, lidar acquisition data and ETC reading and writing data in the original sensing data set, the spatial coordinate correspondence is compared with the baseline layer of the reference radiation field frame by frame. The coordinate clusters with differences in information in each frame are marked. The difference coordinate clusters appearing in multiple consecutive frames are combined to obtain the spatial range of the traffic flow change area, and the boundary is aligned with the acquisition boundary of the original sensing data set. For the spatial range of traffic flow change areas, visible light acquisition information, lidar acquisition information and ETC reading and writing information corresponding to each frame are extracted from the original sensing data set. According to the rendering rules of neural radiation field, corresponding radiation field rendering information is generated coordinate by coordinate. The rendering information of each coordinate corresponds to the three types of acquisition information. The radiation field rendering information of the traffic flow change area is split coordinate by coordinate with the base layer of the reference radiation field. The unchanged coordinate information in the base layer of the reference radiation field is removed, and the radiation field data fragment containing only the change area is extracted. The spatial range is aligned with the spatial range of the traffic flow change area, and the time stamp is synchronized with the unified reference timestamp of the corresponding frame. To match the unified reference timestamp of the original sensing data set to the radiation field data fragments of the changed region, the corresponding time information is marked frame by frame. The timestamp of each frame of radiation field data is synchronized with the timestamp of the corresponding collected data, resulting in radiation field data of the changed region that only contains information about the changed region.

3. The method of claim 2, wherein, The process involves point-by-point spatial coordinate connection between the radiation field data of historically stable regions and the currently running neural radiation field rendering space on the edge side, embedding the radiation field information corresponding to each coordinate into the corresponding position in the rendering space, thereby obtaining a baseline radiation field layer covering the entire spatial range of RSU monitoring. The time stamp is synchronized with the unified baseline timestamp start frame of the original sensing data set, including: Extract the spatial coordinate system parameters of the currently running neural radiation field rendering space on the edge side, covering the entire spatial range monitored by RSU, compare them with the spatial coordinate system parameters of the radiation field data in the historical stable area, and adjust the coordinate parameters of the radiation field data in the historical stable area to align the two types of parameters. The radiation field data of the historical stable region after adjusting the coordinate parameters are sorted according to spatial coordinates. The sorting order corresponds to the coordinate sorting rules of the currently running neural radiation field rendering space on the edge side, so that the radiation field information position of each coordinate corresponds to the coordinate position of the rendering space. The sorted historical stable region radiation field data is input coordinate by coordinate into the currently running neural radiation field rendering space on the edge side. The radiation field information of each coordinate is embedded into the corresponding coordinate position in the rendering space, and the embedding accuracy is aligned with the coordinate accuracy of the rendering space. The embedded rendering space is verified coordinate by coordinate to check whether the radiation field information of each coordinate corresponds to the radiation field data of the historical stable region. The verification results are aligned with the preset requirements of the neural radiation field rendering space currently running on the edge side. The verified rendering space is supplemented with the unified reference timestamp start frame information of the original sensing data set according to the time stamp, and the supplemented timestamp is synchronized with the unified reference timestamp start frame of the original sensing data set. The rendering space after the addition of time stamps is integrated into a complete reference radiation field base layer, covering the entire spatial range monitored by RSU, and the time stamps are synchronized with the unified reference timestamp start frame of the original sensing data set.

4. The method of claim 3, wherein, The step of sorting the radiation field data of the historical stable region after adjusting the coordinate parameters according to spatial coordinates, with the sorting order corresponding to the coordinate sorting rules of the currently running neural radiation field rendering space on the edge side, so that the radiation field information position of each coordinate corresponds to the coordinate position in the rendering space, includes: Extract the coordinate sorting rules of the currently running neural radiation field rendering space on the edge side. The rules cover the entire spatial range monitored by RSU and record the coordinate dimensions and priority order of the sorting. The radiation field data of the historical stable region after adjusting the coordinate parameters is split according to the recorded coordinate dimensions. The radiation field information of each dimension corresponds to the corresponding coordinate dimension, and the accuracy of the split information is aligned with the accuracy of the radiation field data of the historical stable region. The split radiation field information of each dimension is reassembled according to the priority order of the records. The reassembly order corresponds to the coordinate sorting rule of the neural radiation field rendering space currently running on the edge side. The reassembled information covers the entire spatial range monitored by RSU. The recombined radiation field data is verified point by point according to spatial coordinates. The position of the radiation field information at each coordinate corresponds to the coordinate position in the rendering space. The verification results are aligned with the preset requirements of the neural radiation field rendering space currently running on the edge side. The verified radiation field data are arranged continuously according to spatial coordinates. The arranged information forms a continuous radiation field layer, and the layer range covers the entire spatial range monitored by RSU. The continuously arranged radiation field layers are used as the radiation field data of the sorted historical stable region for subsequent operations that embed the currently running neural radiation field rendering space at the edge side.

5. The method of claim 1, wherein, The process involves extracting image semantic information, point cloud geometric information, and ETC time-series information from the original sensing data set, and then integrating these three types of information with the radiation field data of the changed region in a spatiotemporal dimension to obtain the radiation field data to be corrected, including: The visible light acquisition data in the original sensing data set is divided into pixel regions frame by frame. The size of each pixel region corresponds to the point cloud sampling unit of the lidar acquisition data. The types of traffic participants, the layout of road facilities and the current traffic status are identified in each region. The image semantic information containing regional information is integrated and the pixel range is aligned with the acquisition range of each frame. Spatial coordinate clustering is performed frame by frame on the LiDAR acquisition data in the original sensing data set. Each cluster corresponds to an entity object. The spatial contour, real-time position distribution and relative position between objects are identified for each cluster. The point cloud geometric information containing clustering information is integrated. Its coordinate accuracy is aligned with the pixel accuracy of the visible light acquisition data, and its spatial range is aligned with the acquisition range of each frame. The ETC read and write data in the original sensing data set are arranged in time sequence. Each record corresponds to one vehicle passage. The vehicle's passage identifier, passage time and passage route association information are extracted for each record and integrated to obtain ETC time sequence information containing time stamps. Its time stamps are synchronized with the unified reference timestamp of the original sensing data set, and its number of records is aligned with the number of ETC collections. The pixel coordinates of the image semantic information are converted into spatial coordinates that are consistent with the change area radiation field data through a preset transformation relationship. The spatial coordinates of the point cloud geometric information are converted into a spatial coordinate system consistent with the change area radiation field data. The time stamp of the ETC time sequence information is aligned with the time stamp of the change area radiation field data, so that each spatial coordinate of each type of information is aligned with the corresponding coordinate of the change area radiation field data, and the time stamp is synchronized with the unified reference timestamp of the corresponding frame. The semantic information of the aligned image is embedded coordinate by coordinate into the corresponding position of the radiation field data of the changed region. The semantic information of each coordinate covers the corresponding rendering content of the radiation field data of the changed region. Then, the geometric information of the point cloud is embedded coordinate by coordinate to supplement the spatial structure support of the radiation field data of the changed region, so that the three-dimensional structure of the radiation field data of the changed region corresponds to the geometric information of the point cloud and the spatial range is aligned with the radiation field data of the changed region. The aligned ETC time-series information is embedded into the corresponding frame of the changed area radiation field data by time-stamping. The time-series information of each time stamp corresponds to the content of the corresponding frame of the changed area radiation field data. By integrating image semantic information, point cloud geometric information, ETC time-series information and changed area radiation field data, the radiation field data to be corrected is obtained. The time series is synchronized with the original sensing data set.

6. The method of claim 5, wherein, The process of converting the pixel coordinates of image semantic information into spatial coordinates unified with the changed region radiation field data through a preset transformation relationship, converting the spatial coordinates of point cloud geometric information to a spatial coordinate system unified with the changed region radiation field data, aligning the time stamps of ETC time series information with the time stamps of the changed region radiation field data, aligning each spatial coordinate of each type of information with the corresponding coordinates of the changed region radiation field data, and synchronizing the time stamps with the unified reference timestamp of the corresponding frame includes: Extract the spatial coordinate system parameters and time stamping rules of the radiation field data in the changed area. The parameters and rules cover the entire spatial range of the radiation field data in the changed area and correspond to the unified reference timestamp of the original sensing data set. The pixel coordinates of the image semantic information are converted into spatial coordinates that are consistent with the radiation field data of the changed area through a preset conversion formula, so that the range of the converted coordinates is aligned with the spatial range of the radiation field data of the changed area, and the time stamp is synchronized with the unified reference timestamp of the original sensing data set. The spatial coordinates of the point cloud geometric information are transformed to a unified spatial coordinate system with the radiation field data of the changed region, so that the transformed coordinate range is aligned with the spatial range of the radiation field data of the changed region, and the time stamp is synchronized with the unified reference timestamp of the original sensing data set. Align the time stamp of ETC time sequence information with the time stamp rules of radiation field data in the changed area, so that the adjusted time stamp is synchronized with the unified reference timestamp of the original sensing data set, and the recorded content corresponds to the corresponding frame content of radiation field data in the changed area. The adjusted image semantic information, point cloud geometric information, and ETC time series information are compared with the radiation field data of the changed area coordinate by coordinate. The information of each coordinate is checked to see if the comparison results are aligned with the preset requirements of the radiation field data of the changed area. By associating the semantic information of the image, the geometric information of the point cloud, and the time-series information of ETC with the time stamp of the radiation field data of the changed area, the time stamp of each type of information is synchronized with the corresponding frame time stamp of the radiation field data of the changed area, thus completing the point-by-point alignment operation of the spatiotemporal dimension.

7. The method as described in claim 6, characterized in that, The process of converting the pixel coordinates of image semantic information into spatial coordinates consistent with the radiation field data of the changed region using a preset conversion formula, aligning the converted coordinate range with the spatial range of the radiation field data of the changed region, and synchronizing the time stamp with the unified reference timestamp of the original sensory data set includes: Extract the spatial coordinate system parameters of the radiation field data in the changed area. The parameters cover the entire spatial range of the radiation field data in the changed area and correspond to the unified reference timestamp of the original sensing data set. The pixel coordinates of the image semantic information are converted into spatial coordinates that are consistent with the radiation field data of the changed region, and the range of the converted coordinates is aligned with the spatial range of the radiation field data of the changed region. Verify the transformed spatial coordinates point by point to check whether the coordinate transformation results meet the preset requirements and whether the verification results are aligned with the preset requirements of the radiation field data of the changed area. The verified image semantic information is sorted according to spatial coordinates. The sorting order corresponds to the coordinate sorting rules of the radiation field data of the changed area. The sorted information covers the entire spatial range of the radiation field data of the changed area. The sorted image semantic information is supplemented with a unified reference timestamp from the original perceptual data set. The supplemented timestamp is synchronized with the corresponding frame timestamp of the image semantic information, so that the timestamp of each frame corresponds to the unified reference timestamp of the original perceptual data set. The image semantic information with the supplemented timestamp is used as the adjusted image semantic information.

8. The method as described in claim 1, characterized in that, The process involves filling density gaps in sparse point cloud regions of the radiation field data to be corrected with the image semantic information, and calibrating the continuity of vehicle trajectories in the radiation field data to be corrected with the ETC time-series information, resulting in complete radiation field data optimized through cross-type information complementarity, including: The spatial coordinates of the radiation field data to be corrected are traversed frame by frame. The distribution density of radiation field information in each coordinate region is counted. The coordinate clusters in each frame whose density differs from the average distribution of the surrounding adjacent coordinate regions are marked as the initial range of the sparse region of the point cloud. The coordinate accuracy is aligned with the accuracy of the radiation field data to be corrected, and the spatial range is aligned with the acquisition range of the radiation field data to be corrected. The coordinate clusters of the initial range are verified in consecutive frames. Sparse coordinate clusters that appear in three or more consecutive frames are retained, while coordinate clusters that appear in only one or two frames are removed. The retained coordinate clusters are combined to form the final range of the sparse area of ​​the point cloud. The boundary is aligned with the acquisition boundary of the original sensing data set, and the time stamp is synchronized with the unified reference timestamp. Extract semantic information fragments from the image semantic information that correspond to the final range of the sparse region of the point cloud, convert the pixel coordinates of the semantic information fragments into spatial coordinates that are consistent with the radiation field data to be corrected, and the range of the converted spatial coordinates corresponds to the final range of the sparse region of the point cloud, and the semantic content corresponds to the semantic information of the original image. The transformed semantic information fragments are embedded coordinate by coordinate into the corresponding position of the final range of the sparse region of the point cloud in the radiation field data to be corrected. The semantic information of each coordinate supplements the corresponding missing content of the radiation field data to be corrected, so that the radiation field information distribution density of the region is aligned with the average distribution of the surrounding adjacent coordinate regions, and the spatial range is aligned with the final range of the sparse region of the point cloud. The vehicle motion trajectory is traversed frame by frame in the radiation field data to be corrected. The coordinate clusters in each frame where the trajectory is interrupted are marked as the initial range of the trajectory interruption area. The coordinate accuracy is aligned with the accuracy of the radiation field data to be corrected, and the time stamp is synchronized with the unified reference timestamp. Extract time-series information segments corresponding to the initial range of the trajectory interruption area from the ETC time-series information. Synchronize the time stamp of the time-series information segments with the corresponding frame time stamp of the radiation field data to be corrected. Embed the time-series information segments into the corresponding positions of the initial range of the trajectory interruption area of ​​the radiation field data to be corrected. Connect the vehicle motion trajectory in the radiation field data to be corrected and calibrate the continuity of the vehicle motion trajectory in the radiation field data to be corrected. Obtain complete radiation field data after cross-type information complementarity optimization.

9. The method as described in claim 8, characterized in that, The process involves extracting semantic information fragments from the image semantic information that correspond to the final range of the sparse region of the point cloud, converting the pixel coordinates of these semantic information fragments into spatial coordinates consistent with the radiation field data to be corrected, wherein the range of the converted spatial coordinates corresponds to the final range of the sparse region of the point cloud, and the semantic content corresponds to the original image semantic information, including: Extract the spatial coordinate parameters of the final range of the sparse region of the point cloud. The parameters cover the entire space of the final range of the sparse region of the point cloud and correspond to the unified reference timestamp of the original sensing data set. Pixel regions corresponding to the spatial coordinate parameters of the final range of the sparse region of the point cloud are selected from the semantic information of the image. The selected regions are aligned with the spatial range of the final range of the sparse region of the point cloud, and their semantic content corresponds to the semantic information of the original image. The semantic information of the filtered pixel regions is organized into semantic information fragments. The pixel range of the fragments corresponds to the spatial range of the final range of the sparse region of the point cloud. The time stamp is synchronized with the unified reference timestamp of the original perception data set. Extract the spatial coordinate accuracy parameters of the radiation field data to be corrected. The parameters cover the entire spatial range of the radiation field data to be corrected and correspond to the unified reference timestamp of the original sensing data set. The pixel coordinates of the semantic information fragment are converted into the spatial coordinates of the radiation field data to be corrected according to the preset conversion rules. The parameters of the conversion rules correspond to the spatial coordinate precision parameters of the radiation field data to be corrected, and the precision of the converted coordinates is aligned with the precision of the radiation field data to be corrected. The transformed semantic information fragments are verified coordinate by coordinate to check whether the semantic content of each coordinate corresponds to the semantic information of the original image. The verification results are aligned with the preset requirements of the radiation field data to be corrected, thus obtaining embeddable semantic information fragments.

10. An edge computing system, characterized in that, include: processor; And a memory, wherein the memory stores a computer program that, when run by the processor, causes the processor to perform the method as described in any one of claims 1 to 9.

Images