Local dynamic map synchronization and multi-scale updating method based on vehicle-road cooperation under high latency

Abstract

The application discloses a local dynamic map synchronization and multi-scale updating method based on vehicle-road cooperation under high time delay, which comprises the following steps: a local dynamic map synchronization method under high time delay, when the communication module is normal and has no time delay, based on the Particle Filter framework considering time sequence time delay synchronization, multi-source observation information is fused into the vehicle-road cooperative perception system, the fused dynamic data is mapped to the associated scene map, and the local dynamic map of a road intersection is updated; when the communication module has time delay, the latest observation data is not received in time, based on the time sequence, the observation value information at the required time is predicted, then the observation value information is added into the vehicle-road cooperative perception system updating equation, and the local dynamic map synchronization is carried out; in the local dynamic map, a multi-scale state transition method is proposed, and the method is integrated into the prediction of the Particle Filter framework. The application can effectively reduce the target jitter and missing phenomenon of the dynamic map caused by communication delay.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation systems, and in particular to a method for local dynamic map synchronization and multi-scale update based on vehicle-road cooperation under high latency. Background Technology

[0002] Currently, dynamic data layers are widely used in autonomous driving. The main sources of dynamic data are: ① environmental information directly acquired by onboard sensors such as cameras and radar, i.e., active dynamic information; ② information provided by intelligent transportation systems or similar external systems, i.e., passive dynamic information, primarily V2X information from road users, including GNSS data, heading, and speed. As research on dynamic data continues to advance, experts have proposed the concept of dynamic maps. Based on a vehicle-road cooperative framework, this involves integrating multiple road segments to increase the field of view for perception. However, communication between roadside RSUs and vehicle-mounted OBUs is unstable and subject to latency. How to utilize the correlation data between the onboard and roadside units to update and synchronize dynamic maps remains a pressing issue.

[0003] The existing technology has the following problems: In the application of local dynamic maps to various dynamic data, the real-time update of the local dynamic map cannot be accurately guaranteed due to interference from network and environmental factors during the communication between the roadside RSU device and the vehicle-mounted OBU device and the edge computing unit via V2X. This means it may be affected by random latency and random packet loss due to unstable communication signals. Furthermore, on the concrete local dynamic map, moving targets may appear to stagnate or jitter. When the latency is too large, the moving target may remain stuck in the area of ​​the previous moment on the dynamic map and then instantly refresh to the area of ​​the new moment, which has a significant impact on the application layer of the local dynamic map. In summary, the following problems exist:

[0004] 1. In adverse weather conditions such as rain, or when road infrastructure creates blind spots for vehicle perception, the perception and positioning performance of a single vehicle sensor will be severely affected.

[0005] 2. Road information is dynamic, and a single vehicle cannot fully perceive all the dynamic information. Therefore, it is necessary to integrate the information from roadside sensors to supplement the perception information.

[0006] 3. Unstable communication and latency between the roadside RSU and the vehicle-mounted OBU can cause stuttering and freezing in the generated dynamic map.

[0007] 4. For the fusion of multi-source information, more accurate perception and positioning and dynamic map updates can be achieved. Therefore, the method of local dynamic map update for vehicle-road cooperative systems needs further research.

[0008] In summary, research on local dynamic map update methods under vehicle-road cooperation is insufficient, especially on local dynamic map update technology in high-latency environments. No relevant research has been conducted on local dynamic map updates in vehicle-road cooperation under high-latency environments. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a method for local dynamic map synchronization and multi-scale update based on vehicle-road cooperation under high latency, which addresses the shortcomings of the prior art.

[0010] The technical solution adopted by this invention to solve its technical problem is:

[0011] This invention provides a method for local dynamic map synchronization and multi-scale update based on vehicle-road cooperation under high latency, the method comprising:

[0012] The vehicle-road cooperative perception system is built by installing a vehicle-mounted hardware perception system on the vehicle and a roadside hardware perception system on the roadside. Multi-source observation information is collected through the vehicle-mounted hardware perception system and the roadside hardware perception system. Both the vehicle-mounted hardware perception system and the roadside hardware perception system communicate with the edge computing unit through a communication module. The edge computing unit is equipped with a local dynamic map for vehicle-road cooperation.

[0013] A local dynamic map synchronization method under high latency is proposed. When the communication module is functioning normally without latency, multi-source observation information is fused into the vehicle-road cooperative perception system based on the Particle Filter framework, which considers time-series latency synchronization. When the communication module experiences latency and fails to receive the latest observation data on time, the required observation value information is predicted based on the time series and then added to the update equation of the vehicle-road cooperative perception system for local dynamic map synchronization. In the local dynamic map, a multi-scale state transition method is proposed and integrated into the prediction of the Particle Filter framework.

[0014] The method for fusing and updating local dynamic maps based on scene maps uses a communication module to associate and fuse dynamic data of the same moving target collected by different hardware sensing devices. The fused dynamic data is then mapped to the associated scene map to update the local dynamic map of a certain intersection.

[0015] Furthermore, the method for obtaining the scene map in the method of the present invention is as follows: on the road segment to be located, at each collection point in an equidistant manner, the data collection vehicle is equipped with a vehicle-mounted lidar and an inertial navigation system combined with a carrier differential system to continuously collect lidar point cloud data and inertial navigation data to create a high-precision three-dimensional point cloud map.

[0016] The method for obtaining lane lines also includes: extracting features from a high-precision 3D point cloud map, removing ground points, clustering lane lines, constructing lane line equations, and calculating the lateral distance between the vehicle and the lane line, using the distance as a lateral constraint.

[0017] Furthermore, the method for local dynamic map synchronization in the method of the present invention is as follows:

[0018] The communication module of the edge computing unit records the time information of transmission and reception. When the difference between the transmission and reception times is lower than the set absolute transmission difference threshold, the reception time information is directly used as the synchronization time of the local dynamic map. When the difference between the transmission and reception times is higher than the set absolute transmission difference threshold, the measurement position data of the vehicle at the current time is predicted based on the time series and the transmission and reception time information.

[0019] Furthermore, the specific steps of local dynamic map synchronization in the method of the present invention are as follows:

[0020] After acquiring dynamic data from both the vehicle and roadside, the communication devices on both sides are used to transmit all the dynamic data to the edge computing unit for synchronous processing.

[0021] S2.1 First, obtain the time information of the two frames preceding the transmission time. , Furthermore, the transmitted information consists of the vehicle's coordinates obtained after processing within a high-precision map coordinate system. , ;

[0022] S2.2, then obtain the time information of the receiving time. , Time information from the previous frame ;

[0023] S2.3, based on the time information of the first two frames , With receiving time information The vehicle's location information at the time of reception was calculated. The location of the vehicle at any time :

[0024]

[0025] S2.4, based on the solution The coordinates were calculated Coordinates of time :

[0026] .

[0027] Furthermore, in this method of the present invention, the communication module utilizes Cellular-V2X wireless communication technology for communication. The PC5 interface of the on-board unit (OBU) in the vehicle-side hardware system is directly connected to the Uu interface of the roadside unit (RSU) in the roadside hardware system. The specific communication process of the communication module is as follows:

[0028] S2.5.1, the RSU in the communication module of the edge computing unit collects roadside dynamic information through roadside sensing devices, and the on-board unit (OBU) in the vehicle-side hardware system collects vehicle-side dynamic information through vehicle-side sensors.

[0029] S2.5.2, the collected roadside information is sent to the edge computing unit through the Uu interface, and the collected vehicle-side information is sent to the edge computing unit through the PC5 interface;

[0030] S2.5.3 The edge computing unit processes the obtained roadside and vehicle-side information and updates the local dynamic map.

[0031] Furthermore, the method for fusing and updating a local dynamic map based on a scene map in this invention specifically includes:

[0032] Based on the scene map coordinate system, and using the Particle Filter framework that considers time-series delay synchronization, the dynamic perception data (RSU detection and tracking data) sent from the roadside to the edge computing unit based on the scene map coordinate system is correlated and fused with the dynamic perception data (vehicle sub-localization data and lane line lateral constraint data) sent from the vehicle to the edge computing unit. After fusion, the dynamic perception data based on the fusion of multi-source data in the scene map coordinate system is output to the application layer of the local dynamic map based on the scene map coordinate system through the communication module.

[0033] Furthermore, the specific steps of the method for fusing and updating local dynamic maps based on scene maps in this invention are as follows:

[0034] S3.1, Initialization operations of the multi-source data Particle Filter framework;

[0035] S3.1.1, Particle Initialization: Specifies the number of particles. The particles are evenly distributed within the planned area. Initial position coordinates are obtained from dynamic data tracked along the roadside. Then, within a certain error range around the initial coordinates, a Gaussian distribution is obtained using Gaussian variance sampling, and a certain number of particles are selected. The weight of each particle is uniformly set to Initialize particles;

[0036] S3.1.2, when using historical positioning information to predict the current state, the vehicle's motion state is approximated as uniform motion in a short period of time. Therefore, based on the uniform motion model, the vehicle's position at the current moment is predicted using the vehicle's position information from the previous two moments:

[0037]

[0038] Selecting the importance density function:

[0039]

[0040] Based on the geometric center, the state of each particle is estimated to obtain a set of particles with an important density function sampled.

[0041] S3.2, the Particle Filter framework is used for multi-source information fusion, where the particle weights are updated as follows:

[0042]

[0043] S3.2.1, Particle weight update of vehicle-side dynamic observation data: After obtaining the particle set through importance density function sampling, the particle weights need to be evaluated; based on the particle state transition model, the predicted vehicle position is obtained, i.e., the particles distributed around that position; after obtaining the RSU observation data after delayed synchronization processing, considering the observation accuracy of RSU, a two-dimensional Gaussian distribution with the mean and standard deviation of 1m is used; each particle is substituted into this Gaussian distribution, the probability obtained by each particle under this distribution is calculated, the probability is multiplied by the original weight, and then normalization is performed to complete the particle weight update of the fused RSU observation information;

[0044] S3.2.2 After completing the particle weight update of the fused RSU observation information, based on the vehicle localization data after time delay synchronization, a two-dimensional Gaussian distribution with the mean of the data and a certain standard deviation is set; each particle is substituted into the Gaussian distribution, the probability obtained by each particle under the distribution is calculated, the probability is multiplied by the original weight, and then normalization is performed to complete the particle weight update of the fused vehicle localization observation information.

[0045] S3.2.3 After completing the particle weight update of the fusion vehicle-side self-localization observation information, based on the lane line lateral constraint, i.e. the lateral distance between the intelligent vehicle and the lane line, the distance is set to the mean of a Gaussian distribution and obtained with a certain standard deviation. Based on the lane line mathematical expression of the lane-level high-precision map, the distance between each particle and the line segment is calculated. The distance obtained by each particle is substituted into the Gaussian distribution to calculate the relevant probability, then multiplied by the original weight, and then normalized to complete the particle weight update of the fusion vehicle-side lateral constraint observation information.

[0046] S3.3, for each particle, the particle's weight represents its confidence level. A Monte Carlo sampling method is used to selectively retain particles with high weights and discard those with low weights. To avoid particle filtering performance degradation due to particle degradation, the particles are redistributed after several iterations. The effective number of samples is used to determine whether particle degradation has occurred.

[0047]

[0048] When Neff is less than n / 2, the particle filter needs to be resampled, where n is the total number of particles.

[0049] Furthermore, the multi-scale update method in the method of the present invention specifically includes:

[0050] A higher frequency filtering update algorithm is incorporated into the prediction stage of the Particle Filter framework. For update frequencies exceeding a certain threshold, The update frequency is below a certain threshold;

[0051] S4.1, when no vehicle-end observation data is obtained at time m, the frequency is used. Perform a state transition and update the state using the lane line lateral constraint observation information obtained in the previous time step;

[0052] S4.2, when the vehicle-end observation data is obtained at time m, the frequency is... Perform a state transition and update the state using the obtained observation data.

[0053] The beneficial effects of this invention are as follows: This invention provides a method for local dynamic map synchronization and multi-scale updating under high latency based on vehicle-road cooperation. After obtaining dynamic data including roadside vehicle tracking data, vehicle self-localization data, and lane lateral constraints, the data is sent to the edge computing unit via a communication module. The edge computing unit then uses a Particle Filter framework that considers time-series latency synchronization to fuse multi-source dynamic observation data and calculate the maximum a posteriori estimate of the dynamic target for local dynamic map updating. Finally, the updated local dynamic map is mapped to the scene map to construct the local dynamic map. Compared with existing technologies, this invention utilizes the fusion of multiple dynamic observation data to update the local dynamic map and employs Cellular-V2X technology for local dynamic map synchronization and multi-scale updates, effectively reducing target jitter and missing data caused by communication latency in the dynamic map. Attached Figure Description

[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0055] Figure 1 This is a sensor layout diagram of the method proposed in this invention;

[0056] Figure 1 The labels in the text are: 1- Intelligent connected vehicles equipped with LiDAR and high-speed vision sensors; 2- Communication modules; 3- Moving and static targets within the roadside area; 4- Sensor equipment installed on the roadside; 5- Blind spots; 6- Edge computing units; 7- Local dynamic maps for vehicle-road cooperation.

[0057] Figure 2 This is a flowchart of a method according to an embodiment of the present invention;

[0058] Figure 3 This is a schematic diagram of the time-series-based delay synchronization method of the present invention;

[0059] Figure 4 This is a schematic diagram of the multi-scale state transition method of the present invention;

[0060] Figure 5 This is a schematic diagram of the vehicle-side hardware setup of the present invention;

[0061] Figure 5 The markings in the text are: 51-Roof, 52-Vehicle-side sensor 1, 53-Vehicle-side sensor 2, 54-Vehicle-side sensor 3, 55-Vehicle-side sensor 4, 56-Vehicle-side data communication device 1, 57-Vehicle-side data communication device 2, 58-Vehicle-side data communication device 3, 59-Vehicle-side data communication device 4, 510-Onboard communication OBU device;

[0062] Figure 6 This is a schematic diagram of the roadside hardware setup of the present invention;

[0063] Figure 6 The markings in the text are: 61-Roadside equipment mobile base, 62-Road vehicle equipment support rod, 63-Roadside sensing equipment support frame, 64-Roadside sensing sensor 1, 65-Roadside sensing sensor 2, 66-Roadside sensing sensor 3, 67-Roadside sensing sensor 4, 68-Roadside communication equipment. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0065] Example 1

[0066] like Figure 1 As shown in the embodiment of the present invention, a method for local dynamic map synchronization and multi-scale update based on vehicle-road cooperation under high latency is provided. The method includes:

[0067] The vehicle-road cooperative perception system is built by installing a vehicle-mounted hardware perception system on the vehicle and a roadside hardware perception system on the roadside. Multi-source observation information is collected through the vehicle-mounted hardware perception system and the roadside hardware perception system. Both the vehicle-mounted hardware perception system and the roadside hardware perception system communicate with the edge computing unit through a communication module. The edge computing unit is equipped with a local dynamic map for vehicle-road cooperation.

[0068] The communication module is mainly responsible for communication in the proposed local dynamic map based on vehicle-road cooperation, and is responsible for sending the data received by the roadside and the vehicle to the edge computing module.

[0069] The edge computing module is mainly responsible for receiving data from the communication module and performing local dynamic map fusion and updates.

[0070] A local dynamic map synchronization method under high latency is proposed. When the communication module is functioning normally without latency, multi-source observation information is fused into the vehicle-road cooperative perception system based on the Particle Filter framework, which considers time-series latency synchronization. When the communication module experiences latency and fails to receive the latest observation data on time, the required observation value information is predicted based on the time series and then added to the update equation of the vehicle-road cooperative perception system for local dynamic map synchronization. In the local dynamic map, a multi-scale state transition method is proposed and integrated into the prediction of the Particle Filter framework.

[0071] The method for fusing and updating local dynamic maps based on scene maps uses a communication module to associate and fuse dynamic data of the same moving target collected by different hardware sensing devices. The fused dynamic data is then mapped to the associated scene map to update the local dynamic map of a certain intersection.

[0072] Example 2

[0073] like Figure 2 As shown, Embodiment 2 of the present invention provides a method for multi-scale synchronization and updating of local dynamic maps based on vehicle-road cooperation under high latency, including a method for local dynamic map synchronization under high latency, a multi-scale particle filter state transition method, and a method for local dynamic map fusion and updating based on the scene map coordinate system; wherein:

[0074] In the high-latency local dynamic map synchronization method, based on the ParticleFilter framework which considers time-series latency synchronization, two different synchronization methods are employed. When the communication module in the local dynamic map can obtain the current observation value on time without latency, its state is corrected based on the observation information. When the communication module in the local dynamic map experiences latency and cannot obtain the observation value on time, the current observation value information is calculated based on the time series, using the sending and receiving times of the communication module. The communication module utilizes Cellular-V2X wireless communication technology. Cellular-V2X technology mainly includes two communication ports: the PC5 interface and the Uu interface. The PC5 interface is the communication interface between terminals, i.e., a short-range direct communication interface between vehicles, people, and road infrastructure, providing low latency, high capacity, and high reliability through direct connection, broadcasting, and network scheduling. The Uu interface is the communication interface between terminals and base stations, enabling reliable communication over long distances and wider ranges. The communication module uses the PC5 interface and Uu interface of Cellular-V2X wireless communication technology for communication.

[0075] In the multi-scale particle filter state transition method, a higher-frequency filtering update algorithm is incorporated into the prediction stage of the particle filter framework. Considering that the lane line lateral constraint information changes relatively little, the lane line lateral constraint information obtained from the previous frame is directly used as the input for observation update. For higher update frequency, It has a relatively low update frequency. ,when If no roadside observation value is received from the vehicle, the system uses the lane line lateral constraint information obtained in the previous frame and... For frequency, perform state transition update when When the system receives the roadside observation values ​​from the vehicle end, it uses Update the frequency fusion vehicle-road multi-source observation data.

[0076] In the local dynamic map fusion and update method based on the scene map coordinate system, the proposed Particle Filter framework, which considers time-series latency synchronization, is used. Dynamic perception data in the scene map coordinate system, sent from the roadside portion of the communication module to the edge computing unit, is fused with dynamic perception data sent from the vehicle-side portion of the communication module to the edge computing unit using the same framework. The fused dynamic perception data, obtained through multi-source data fusion in the scene map coordinate system, is then output through the communication module. A static base map of the local dynamic map is created from the previously constructed local high-precision point cloud map of the current area. Utilizing the extrinsic parameter calibration relationship between the scene map and the roadside equipment, the obtained dynamic perception data obtained through multi-source data fusion in the scene map coordinate system is mapped back onto the static base map of the local dynamic map using a mapping matrix, thus completing the local dynamic map update.

[0077] Example 3

[0078] The method for local dynamic map synchronization and multi-scale update under high latency according to embodiments of the present invention includes the following steps:

[0079] S1. A method for building a local dynamic map system based on vehicle-road cooperation under high latency, which builds a vehicle-road cooperative perception hardware system, including a vehicle-side hardware perception system and a roadside hardware perception system.

[0080] S2. Data synchronization method for high-latency scenarios based on time series. When there is a communication delay in the communication module of a local dynamic map, i.e., the observation value cannot be obtained on time, the observation value information at the current moment is calculated based on the time series.

[0081] S3. A multi-source information fusion method based on vehicle-road cooperation. This method utilizes the proposed Particle Filter framework, which considers time-series delay synchronization, to fuse dynamic perception data from both the vehicle and road sides.

[0082] S4. A multi-scale update method based on vehicle-road cooperation.

[0083] S1 requires building a vehicle-road cooperative sensing hardware system, including a vehicle-side sensing hardware system, a roadside sensing hardware system, and an edge computing unit. For example... Figure 5 As shown, the specific steps include:

[0084] S1.1, Build the vehicle-side hardware perception system. The vehicle-side hardware perception system consists of the roof, vehicle-side sensor 1, vehicle-side sensor 2, vehicle-side sensor 3, vehicle-side sensor 4, vehicle-side data communication device 1, vehicle-side data communication device 2, vehicle-side data communication device 3, vehicle-side data communication device 4, and on-board unit (OBU) communication device.

[0085] S1.1.1, Vehicle-side sensor 1, vehicle-side sensor 2, vehicle-side sensor 3, and vehicle-side sensor 4 are arranged around the roof of the vehicle.

[0086] S1.1.2, Vehicle-side sensor 1, Vehicle-side sensor 2, Vehicle-side sensor 3, and Vehicle-side sensor 4 are sensors and sensor combinations that can observe perception information around the vehicle, such as lidar and vision sensors. The number and layout of the sensors are related to the observation coverage area.

[0087] S1.1.3, Vehicle-side data communication device 1, Vehicle-side data communication device 2, Vehicle-side data communication device 3, and Vehicle-side data communication device 4 are devices capable of establishing wired communication within the vehicle. The number of vehicle-side data communication devices is related to the number of on-board sensors. The on-board communication OBU device is the device that sends communication data from the vehicle to the roadside equipment. It should be ensured that the vehicle can transmit data with the edge computing unit through the on-board communication OBU device.

[0088] S1.2, Roadside hardware sensing system. For example... Figure 6 As shown, the roadside hardware sensing system consists of a roadside equipment mobile base, a road vehicle equipment support rod, a roadside sensing equipment support frame, roadside sensing sensor 1, roadside sensing sensor 2, roadside sensing sensor 3, roadside sensing sensor 4, and roadside communication equipment.

[0089] S1.2.1, the roadside hardware sensing system should be mobile, meaning it can be placed in any suitable area at the required intersection.

[0090] S1.2.2, the roadside equipment mobile base, the road vehicle equipment support rod, and the roadside sensing equipment support structure form the foundation of the roadside sensing hardware system, on which all sensors and communication devices are selected and added.

[0091] S1.2.3, Roadside sensing sensor 1, Roadside sensing sensor 2, Roadside sensing sensor 3, and Roadside sensing sensor 4 are sensors and sensor combinations that can observe the surrounding information of the roadside, such as lidar and vision sensors. The number and layout of the sensors are related to the observation coverage.

[0092] S1.2.4, The roadside communication device receives data sent by roadside sensing sensor 1, roadside sensing sensor 2, roadside sensing sensor 3, and roadside sensing sensor 4, and sends it to the edge computing unit.

[0093] S2 is a data synchronization method for high-latency scenarios based on time series. After acquiring dynamic data from both the vehicle and roadside, the dynamic data from both are transmitted to the edge computing unit using communication devices on both sides. Synchronization is then performed at the edge computing unit. The specific process is as follows: Figure 3 As shown.

[0094] S2.1 First, obtain the time information of the two frames preceding the transmission time. , Furthermore, the transmitted information consists of the vehicle's coordinates obtained after processing within a high-precision map coordinate system. , .

[0095] S2.2, Obtaining the time information of the receiving moment. , Time information from the previous frame That is, the vehicle position coordinates they represent respectively. , .

[0096] S2.3, based on the time information of the first two frames , With receiving time information The vehicle's location information at the time of reception was calculated. The location of the vehicle at any time :

[0097]

[0098] S2.4, based on the solution The coordinates were calculated Coordinates of time :

[0099]

[0100] S2.5 utilizes Cellular-V2X wireless communication technology for communication. The PC5 interface of the on-board unit (OBU) in the vehicle-side hardware system and the Uu interface of the roadside unit (RSU) in the roadside hardware system are directly connected. The specific communication process is as follows:

[0101] In S2.5.1, the RSU in the communication module of the local dynamic map collects roadside dynamic information through roadside sensing devices, and the OBU in the vehicle-side hardware system collects vehicle-side dynamic information through vehicle-side sensors.

[0102] S2.5.2, the roadside information collected by the communication module in the local dynamic map is sent to the edge computing unit through the Uu interface, and the vehicle-side information collected by the communication module in the local dynamic map is sent to the edge computing unit through the PC5 interface.

[0103] S2.5.3 The edge computing unit processes the obtained roadside and vehicle-side information and updates the local dynamic map.

[0104] In S2.5.4, the edge computing unit sends the updated local dynamic map to the application layer.

[0105] S3 is a multi-source information fusion method based on vehicle-road cooperation. It utilizes the synchronized observation data obtained through a time-series-based delay synchronization method, and employs the proposed Particle Filter framework. Based on the scene map coordinate system, it combines dynamic perception data sent from the roadside to the edge computing unit with dynamic perception data sent from the vehicle to the edge computing unit, outputting dynamic perception data fused from multiple sources in the scene map coordinate system.

[0106] S3.1 Initialization operation using the proposed multi-source data Particle Filter framework.

[0107] S3.1.1, Particle Initialization: Specifies the number of particles. The particles are evenly distributed across the planned area. Generally, a larger number of particles can more accurately represent the Bayesian posterior distribution, improving the reliability of the state, but it will increase the algorithm's running time. The initial position coordinates are obtained from the dynamic data of the roadside tracking section. Then, a Gaussian distribution is obtained by sampling within a 1-meter error range of the initial coordinates, using a Gaussian variance. The number of particles is then determined. The weight of each particle is uniformly set to Initialize particles.

[0108] S3.1.2, When using historical positioning information to predict the current state, the vehicle's motion state can be approximated as uniform motion in a short period of time. Therefore, based on the uniform motion model, the vehicle's position at the current moment can be predicted based on the vehicle's position information from the previous two moments.

[0109]

[0110] Based on the prediction model, the importance density function is selected:

[0111]

[0112] Based on the geometric center, the state of each particle is estimated to obtain a set of particles with an important density function sample.

[0113] S3.2, the proposed particle filter framework is used for multi-source information fusion, where the particle weights are updated as follows:

[0114]

[0115] S3.2.1, Particle Weight Update of Vehicle-End Dynamic Observation Data: After obtaining the particle set through importance density function sampling, the particle weights need to be evaluated. Based on the particle state transition model, the predicted vehicle position is obtained, which represents the particles distributed around that position. After obtaining the RSU observation data after delayed synchronization processing, considering the observation accuracy of RSU, a two-dimensional Gaussian distribution with a mean of 1m and a standard deviation of 1m is used. Each particle is substituted into this Gaussian distribution, and the probability obtained by each particle under this distribution is calculated. This probability is multiplied by the original weight, and then normalized to complete the particle weight update of the fused RSU observation information.

[0116] In step S3.2.2, after updating the particle weights of the fused RSU observation information, a two-dimensional Gaussian distribution with a mean of 0.2m and a standard deviation of 0.2m is set based on the vehicle localization data after time-delay synchronization. Each particle is substituted into this Gaussian distribution, and the probability of each particle under this distribution is calculated. This probability is multiplied by the original weight, and then normalized to complete the particle weight update of the fused vehicle localization observation information.

[0117] In step S3.2.3, after updating the particle weights based on the fusion of vehicle-side self-localization observation information, the lateral distance between the intelligent vehicle and the lane line, i.e., the lateral distance between the intelligent vehicle and the lane line, is set to the mean of a Gaussian distribution with a standard deviation of 0.1m to obtain a one-dimensional Gaussian distribution. Based on the lane-level high-precision map's lane line mathematical expression, the distance between each particle and the line segment is calculated. The distance obtained for each particle is then substituted into the Gaussian distribution to calculate the relevant probability, multiplied by the original weight, and then normalized to complete the particle weight update fusion of vehicle-side lateral constraint observation information.

[0118] In step S3.3, for each particle, its weight represents its reliability. Generally, particles with larger weights have a greater filtering effect. Therefore, the Monte Carlo sampling method selectively retains particles with large weights and discards those with small weights. However, as the number of iterations increases, the weights of most particles approach zero, while a small number of particles acquire large weights. This phenomenon is called particle degradation. To avoid the decline in particle filtering performance caused by particle degradation, the particles need to be redistributed after several iterations. Redistribution increases particle diversity. The effective sampling number is used to determine whether particle degradation has occurred.

[0119]

[0120] When Neff is less than n / 2, the particle filter needs to be resampled, where n is the total number of particles.

[0121] S4 is a multi-scale update method based on vehicle-road cooperation. It incorporates a higher-frequency filtering update algorithm into the prediction stage of the Particle Filter framework. For higher update frequency, It has a relatively low update frequency.

[0122] S4.1, when no vehicle-end observation data is obtained at time m, the frequency is used. The state transition is performed and updated using the lane line lateral constraint observation information obtained in the previous time step.

[0123] S4.2, when the vehicle-end observation data is obtained at time m, the frequency is... Perform a state transition and update the state using the obtained observation data.

[0124] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for local dynamic map synchronization and multi-scale update based on vehicle-road cooperation under high latency, characterized in that, The method comprises: The vehicle-road cooperative perception system is built, including loading a vehicle-end hardware perception system on a vehicle, setting a roadside hardware perception system at a roadside, collecting multi-source observation information through the vehicle-end hardware perception system and the roadside hardware perception system; the vehicle-end hardware perception system and the roadside hardware perception system both communicate with an edge computing unit through a communication module, and the edge computing unit is provided with a local dynamic map for vehicle-road cooperation; A local dynamic map synchronization method under high time delay, when the communication module has no time delay in communication, multi-source observation information is fused into the vehicle-road cooperative perception system based on a Particle Filter framework considering time sequence time delay synchronization; when the communication module has time delay in communication, the latest observation data is not received in time, and observation value information at a current required time is predicted based on a time sequence, and then added to a vehicle-road cooperative perception system update equation to perform local dynamic map synchronization; in the local dynamic map, a multi-scale state transition method is proposed and integrated into prediction of the Particle Filter framework; A fusion and update method of the local dynamic map based on a scene map, dynamic data of a same moving target collected by different hardware perception devices are associated and fused through the communication module, and the fused dynamic data are mapped to the associated scene map to update the local dynamic map of a certain intersection; The local dynamic map synchronization method in the method is as follows: The sending time information and the receiving time information of the communication module of the edge computing unit are recorded, when the difference between the sending time information and the receiving time information is lower than a set transmission absolute difference threshold, the receiving time information is directly used as the synchronization time of the local dynamic map; when the difference between the sending time information and the receiving time information is higher than the set transmission absolute difference threshold, the sending time information and the receiving time information are used to predict the measurement position data of the vehicle at the current time based on a time sequence.

2. The method of claim 1, wherein the local dynamic map is synchronized and updated based on a cooperative vehicle infrastructure system (CVIS) in a high latency environment. The scene map acquisition method in the method is as follows: on a to-be-positioned road section, a data collection vehicle is loaded with a vehicle-mounted laser radar, an inertial navigation system and a combined carrier wave differential system to uninterruptedly collect laser point cloud data and inertial navigation data to make a high-precision three-dimensional point cloud map; The lane line acquisition method further comprises: extracting features in the high-precision three-dimensional point cloud map, performing ground point elimination, performing lane line clustering, constructing a lane line equation, and calculating a lateral distance between the vehicle and the lane line, and taking the distance as a lateral constraint.

3. The method of claim 1, wherein the local dynamic map is synchronized and updated based on a cooperative vehicle infrastructure system (CVIS) in a high latency environment. The specific steps of the local dynamic map synchronization in the method are as follows: After the dynamic data of the vehicle end and the roadside are acquired respectively, the dynamic data of the vehicle end and the roadside are all transmitted to the edge computing unit by using the communication devices of the vehicle end and the roadside for synchronous processing in the edge computing unit: S2.1, first acquire the time information of the two frames before the sending time , , and the transmitted information is the coordinate of the vehicle in the high-precision map coordinate system after processing , ; S2.2, time information of the reception time is acquired again , time information of the previous frame ; S2.3, based on time information of the previous two frames , with the reception time information to calculate the position information of the vehicle at the reception time position of the vehicle at the reception time : S2.4, based on the solved coordinates of the coordinates of the moment 。 4. The method of claim 1, wherein the local dynamic map synchronization and multi-scale update based on vehicle-road cooperation under high latency is characterized in that, The communication module uses a Cellular-V2X wireless communication technology for communication, a PC5 interface of a vehicle-mounted communication OBU device in the vehicle-end hardware system is directly connected with a Uu interface of a roadside communication RSU device in the roadside hardware system, and the specific communication process of the communication module is as follows: S2.5.1, RSU in the communication module in the edge computing unit collects roadside dynamic information through roadside sensing devices, and the vehicle-mounted communication OBU in the vehicle-end hardware system collects vehicle-end dynamic information through vehicle-end sensors; S2.5.2, the collected roadside information is sent to the edge computing unit through the Uu interface, and the collected vehicle-end information is sent to the edge computing unit through the PC5 interface; S2.5.3, the edge computing unit updates the local dynamic map after processing the obtained roadside and vehicle-end information.

5. The method of claim 3, wherein the local dynamic map is synchronized and updated based on a cooperative vehicle infrastructure system (CVIS) in a high latency environment. The fusion updating method of the local dynamic map based on the scene map in the method is as follows: Based on the scene map coordinate system, through the Particle Filter framework considering time sequence time delay synchronization, the dynamic perception data based on the scene map coordinate system of the roadside part sent to the edge computing unit, i.e. RSU detection tracking data, and the dynamic perception data of the vehicle-end part sent to the edge computing unit, i.e. vehicle sub-positioning data and lane line transverse constraint data, are associated and fused using the proposed Particle Filter framework considering time sequence time delay synchronization, and the dynamic perception data based on the scene map coordinate system after fusion is output to the application layer of the local dynamic map based on the scene map coordinate system through the communication module.

6. The method of claim 5, wherein the local dynamic map is synchronized and updated based on a cooperative vehicle infrastructure system (CVIS) at a low latency. The specific steps of the fusion updating method of the local dynamic map based on the scene map in the method are as follows: S3.1, initialization operation of the fusion multi-source data Particle Filter framework; S3.1.1, particle initialization: specify the number of particles and the particle average distribution in the planning area, take the dynamic data of the roadside part of the track to get the initial initial position coordinate, and then get the Gaussian distribution sampling in a certain error range near the initial coordinate with Gaussian variance, take a certain number of particles Each particle weight is uniformly set to Initialize the particle; S3.1.2, when predicting the current state using historical positioning information, the motion state of the vehicle is approximately regarded as uniform motion state in a short time, so the vehicle object position at the current time is predicted based on the uniform motion model according to the vehicle position information at the previous two times: An important density function is selected: where y 1：k represents the measurement data at the 1st to kth time in Monte Carlo sampling. where x k （i） denotes the state quantity of x at the kth time instant. State estimation is performed on each particle at the current time based on the geometric center to obtain a particle set sampled by the important density function; S3.2, multi-source information fusion is performed using the Particle Filter framework, and the particle weight is updated as follows: wherein, denotes the weight of the particle, R denotes Roadside observation data, V denotes Vehicle observation data, and L denotes Land lane line observation data; S3.2.1, particle weight update of vehicle-end dynamic observation data: after obtaining the particle set by sampling through the important density function, the particle weight needs to be evaluated; based on the particle state transition model, the predicted position of the vehicle is obtained, i.e. the particles distributed around the position; after obtaining the RSU observation data after delay synchronization processing, a two-dimensional Gaussian distribution with the data as the mean value is considered according to the observation accuracy of the RSU; each particle is brought into the Gaussian distribution, the probability obtained by each particle in the distribution is calculated, the probability is multiplied by the original weight, and then normalization is completed to update the particle weight fused with the RSU observation information; S3.2.2, after completing the particle weight update of the fusion RSU observation information, based on the self-vehicle positioning data after time delay synchronization, a two-dimensional Gaussian distribution with the data as the mean and a certain standard deviation is set; each particle is brought into the Gaussian distribution, the probability obtained by each particle in the distribution is calculated, the probability is multiplied by the original weight, and then normalized to complete the particle weight update of the fusion vehicle end self-positioning observation information; S3.2.3, after completing the particle weight update of the fusion vehicle end self-positioning observation information, based on the lane line lateral constraint, i.e., the lateral distance between the intelligent vehicle and the lane line, the distance is set as the mean of the Gaussian distribution, and a one-dimensional Gaussian distribution is obtained with a certain standard deviation; based on the lane-level high-precision map lane line mathematical expression, the distance between each particle and the line segment is calculated; the distance obtained by each particle is brought into the Gaussian distribution to calculate the related probability, and then multiplied by the original weight, and then normalized to complete the particle weight update of the fusion vehicle end lateral constraint observation information; S3.3, for each particle, the size of the weight of the particle represents the credibility of the particle, and the particles with large weights are selectively retained and the particles with small weights are discarded by the Monte Carlo sampling method; in order to avoid the performance degradation of particle filtering caused by particle degradation, after a number of iterations, the particles are redistributed; the effective sampling number is used to judge whether the particles have degenerated: When N eff If n / 2 is less than n, the particle filter needs to be resampled, where n is the total number of particles.

7. The method of claim 1, wherein the method is based on a cooperative vehicle infrastructure system (CVIS) in a high latency environment. The multi-scale updating method in the method is specifically: In the prediction step of the Particle Filter framework, a higher frequency filter update algorithm is incorporated. for update frequencies above a certain threshold, for update frequencies below a certain threshold. S4.1, when the vehicle end observation data is not obtained at time m, update the lane line lateral constraint observation information at a frequency of state transition is performed and the lane line lateral constraint observation information obtained at the previous time is updated; S4.2, when the vehicle end observation data is obtained at time m, update the state with the obtained observation data at a frequency of make a state transition and update with the obtained observation data.

Images