A road-vehicle intelligent fusion perception method considering driving risk and communication load in coordination

Abstract

The application discloses a kind of road vehicle intelligent fusion sensing methods considering driving risk and communication load synergistically, comprising the following steps: S1, point cloud data with time and pose label is collected by sensing unit;S2, communication unit obtains cooperative sensing data;S3, computing unit fuses point cloud data and carries out three-dimensional target detection;S4, communication unit receives the cooperative sensing data request of other system;S5, computing unit assesses the driving risk in the sensing area based on safety field theory;S6, computing unit allocates the communication resources available to system between received data request;S7, the point cloud collected to sensing unit is optimized representation;S8, cooperative sensing data is sent to each system requesting cooperative sensing data.According to the application, for the sensing task under different communication resource constraints, efficient sharing between vehicle and road sensing resources is realized, and reliable environmental sensing data is provided for the behavior decision of autonomous vehicle.

Description

Technical Field

[0001] This invention relates to the technical field of vehicle-road cooperative environment perception, and in particular to a road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load. Background Technology

[0002] Environmental perception is crucial for autonomous driving, serving as one of the core capabilities of any autonomous driving system. By perceiving obstacles in the surrounding environment, it enables autonomous vehicles to understand their environment and make appropriate decisions. To meet the needs of autonomous driving decision-making, numerous sensor technologies have been applied to the environmental perception systems of autonomous vehicles, including radar, cameras, and lidar. Lidar, in particular, generates point cloud data by emitting laser beams and measuring their time-of-flight, offering advantages such as high data resolution, wide detection area, and low light dependence. Furthermore, the increasing maturity of 3D object detection and point cloud segmentation algorithms in recent years allows for the accurate extraction of obstacle 3D pose and contour information from point cloud data. With these advantages, lidar has become one of the most widely used perception devices in autonomous driving environmental perception systems.

[0003] Currently, autonomous driving environmental perception systems face numerous unresolved challenges. Firstly, high cost: the manufacturing of high-end sensors like LiDAR involves complex optical and mechanical designs, as well as the production of high-precision components, limiting mass production and resulting in relatively high costs. Secondly, limited field of view: due to obstacle obstruction and limitations in sensor perspective and operating principles, blind spots inevitably exist even in areas close to the vehicle, creating safety hazards. Furthermore, each sensor has an upper limit to its perception range; objects beyond this range will be completely undetectable to the vehicle, impacting the safety and reliability of vehicle decisions at high speeds.

[0004] To overcome these problems, vehicle-to-everything (V2X) technology has emerged, deploying sensing and computing devices on the roadside and leveraging V2X (vehicle-to-everything) communication technology to share perception information among multiple traffic participants, thereby solving the problems of limited field of view and high cost per vehicle. Researchers have proposed various data sharing frameworks based on V2V (vehicle-to-vehicle) and V2I (vehicle-to-infrastructure) communication to achieve global environmental perception from multiple perspectives. These studies mainly employ three data fusion strategies: data-level fusion, feature-level fusion, and target-level fusion. In the target-level fusion framework, each vehicle with communication capabilities shares its own perception results in the network, namely the position and attitude of vehicles and pedestrians. This makes the target-level fusion scheme have lower overhead on the communication network, but its improvement in perception performance is limited due to noise interference and false detections. Feature-level fusion shares features extracted by neural networks. This method achieves a good balance between communication consumption and perception accuracy and can realize multi-modal sensor data processing, thus attracting widespread attention. However, feature-level fusion requires vehicles in the network to use a unified perception model, and features extracted by different perception models are difficult to share, which limits the application of feature-level fusion schemes. Data-level fusion, on the other hand, shares raw sensor data. Even if each autonomous vehicle deploys its own different perception model, the sensor data can still be used interchangeably. In addition, because it directly shares raw perception data, this method can achieve optimal perception performance. However, data-level fusion under high communication traffic is difficult to achieve due to bandwidth limitations. With limited bandwidth resources, the communication of a large amount of sensor data will cause severe latency in the communication network, resulting in low-performance perception results that are unsuitable for decision-making. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication loads. This method achieves efficient sharing of vehicle-road perception resources for perception tasks under different communication resource constraints, providing reliable environmental perception data for the behavioral decisions of autonomous vehicles. To achieve the above-mentioned objectives and other advantages of the present invention, a road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication loads is provided. The method includes a perception system comprising a vehicle-side platform, a roadside platform signal-connected to the vehicle-side platform, a cloud platform communicating with the vehicle-side platform or the roadside platform, positioning and attitude determination equipment, and point cloud acquisition equipment. The vehicle-side platform includes a connected vehicle, which includes a perception unit, a communication unit, and a computing unit. The roadside platform includes road infrastructure such as poles and gantries, and deploys the perception unit, communication unit, and computing unit. The cloud platform includes a cloud server and a cloud database, further comprising a computing unit and a storage unit. The method is characterized by the following steps:

[0006] S1. Collect point cloud data with time and pose labels through the sensing unit;

[0007] S2. The communication unit acquires collaborative sensing data from other systems through the road-vehicle communication network;

[0008] S3. The computing unit fuses point cloud data and performs 3D target detection;

[0009] S4. The communication unit receives collaborative sensing data requests from other systems through the road-vehicle communication network.

[0010] S5. The computing unit assesses driving risks within the perception area based on the safety field theory.

[0011] S6. The computing unit comprehensively considers driving risks and communication load, and allocates the available communication resources of the system among the received data requests.

[0012] S7. The computing unit optimizes and represents the point cloud collected by the sensing unit according to the resource allocation scheme in step S6.

[0013] S8. The communication unit sends collaborative sensing data to each system that requests collaborative sensing data according to the resource allocation scheme in step S6.

[0014] Preferably, step S5 specifically includes the following steps:

[0015] S51. Define the reference point, resolution, and range of the safe field calculation in the coordinate system of the sensing unit.

[0016] S52. Define the coordinates (x, y, z) of the sensing unit in the rectangular coordinate system and the coordinates in the ellipsoidal coordinate system. The transformation relationship between them;

[0017] S53. Generate the set of safe field calculation points H in the sensing unit coordinate system according to the safe field calculation reference point, safe field calculation resolution and safe field calculation range defined in S51.

[0018] S54. For each safety field calculation point in the set of safety field calculation points H, calculate the ellipsoidal coordinates corresponding to the rectangular coordinates (x, y, z) of the safety field calculation point according to the coordinate transformation relationship defined in S53.

[0019] S55. For target-level environmental perception datasets For each detected target vehicle χ, a safety field calculation is performed on the safety field calculation point set defined in S53;

[0020] S56. Superimpose the safety fields generated by each detected target vehicle in the target-level environmental perception dataset to obtain the summed safety field;

[0021] S57. Normalize the total security field to obtain the standard total security field matrix F;

[0022] S58. Based on the standard summation safety field matrix F, add a driving risk assessment item to the collaborative perception data request set generated in step S4 to obtain a collaborative perception data request set with driving risk assessment.

[0023] Preferably, step S55 specifically includes the following steps:

[0024] S551. Define the safety field calculation parameters k1 and k2;

[0025] S552. Define and calculate the virtual mass M of the target vehicle χ. χ ;

[0026] S553. Define and calculate the road condition impact R at point χ for the target vehicle. χ Influencing factors include, but are not limited to, the road adhesion coefficient, road slope, road curvature, and air visibility in the vicinity of the target vehicle.

[0027] S554. For the set of coordinate points H for calculating the safety field described in S53 χ The coordinates (x, y, z) of each safety field calculation point in the sensing unit's rectangular coordinate system and its coordinates in the sensing unit's ellipsoidal coordinate system as described in S54. Calculate the field strength value of the safety field generated by the target vehicle χ at that location according to Equations 2 and 3;

[0028]

[0029]

[0030] Where θ represents the velocity vector v of the line connecting object χ and point (x, y) and target vehicle χ. χ The angle between them.

[0031] Preferably, step S58 specifically includes the following steps:

[0032] S581. For the positioning data T in the global coordinate system sent by each system in the collaborative sensing data request set R in step S4, in step S1. i Perform coordinate transformation to obtain the coordinates of each requested data's system location in the sensing unit coordinate system of this system;

[0033]

[0034] S582. Query the index p of the point closest to the location of the requested data system from the security field calculation point set H described in S53. k And take the safe field strength value at that point. As the result of the system's driving risk assessment;

[0035] S583. Combine the driving risk assessment results of all requesting data systems to generate a driving risk assessment set for each requesting collaborative perception data system.

[0036] Preferably, step S6 specifically includes the following steps:

[0037] S61. Communication resources include: the number of available subframes L, the number of subchannels K, the subframe width t, and the subchannel width b; define the available resource sequence q = [c i t|c i = b·log(1+SINR) i ), i = 1, 2, ..., LK], where SINR i Let be the signal-to-noise ratio of channel i;

[0038] S62. Calculate the size of the point cloud data collected by the sensing unit in step S1 |X s |;

[0039] S63. Before system deployment, determine the target point cloud data compression rate k through simulation experiments. t ;

[0040] S64. Define the compression ratio multiplication factor λ, define the resource allocation matrix and assign initial values. Define a point cloud data compression ratio vector and assign initial values. Define the communication resource requirements and assign an initial value s = |X s |k t Define the remaining allocable resource size and initialize it with r = ∑ i q i ;

[0041] S65. Sort the systems in the collaborative perception data request set according to the driving risk items in the collaborative perception data request set with driving risk assessment as described in S58, to obtain the collaborative perception data request sequence R. s Length n E ;

[0042] S66. Allocate communication resources from the resource sequence in step S61 to each system in the collaborative sensing data request sequence in sequence until the resource allocation is completed or r≤0 or all requirements are met.

[0043] S67. Record the compression ratio vector k and resource allocation matrix x of the final point cloud data.

[0044] Preferably, step S63 specifically includes the following steps:

[0045] S631. Establish a simulation platform, including traffic flow simulation, sensor simulation, and vehicle dynamics simulation modules;

[0046] S632. According to application requirements, calibrate the parameters of each model in the simulation platform, establish a simulation scene, and build the perception unit of step S1 in the simulation platform; specifically including: establishing a scene map, configuring vehicle parameters, configuring sensor parameters, and setting scene traffic flow;

[0047] S633. Set a fixed sampling rate and sample the point cloud data collected by the sensing unit in the simulation as described in step S7.

[0048] S634. A method for fusing simulated point cloud data and performing 3D target detection as defined in step S3;

[0049] S635. Run the simulation and collect the true position values ​​of each target vehicle in the scene through the simulation platform, and collect the target vehicle position data detected by the system in S3.

[0050] S636. Based on the fixed sampling rate described in step S634, the size of the point cloud data collected by the simulation sensing unit in step S632, and the data transmission bandwidth limit in the application requirements, calculate the data transmission delay.

[0051] S637. Perform time delay compensation on the true position value of the target vehicle based on the data transmission delay, and compare it with the true position value of each target vehicle in the scene collected by the simulation platform to calculate the positioning error of the average time delay compensation.

[0052] S638. Update the fixed sampling rate. Repeat steps S634-S638 to optimize the average positioning error until the fixed sampling rate or the average positioning error converges.

[0053] Preferably, step S67 specifically includes the following steps:

[0054] S671. Define the target system index i, pointing to the first system in the collaborative sensing data request sequence described in S65;

[0055] S672. Define a communication resource index j, pointing to the first resource block of the available resource sequence described in S61;

[0056] S673. If resource allocation is complete and r≤0, exit the resource allocation procedure.

[0057] S674. If the remaining communication resource size does not meet the requirement r < s, increase the compression ratio s ← s / λ according to the compression ratio multiplication factor described in S64.

[0058] S675. Define the system communication resource that has been allocated to index i and assign it an initial value p = 0;

[0059] S676. Allocate the resource pointed to by index j to the system x pointed to by index i. i，j ←1;

[0060] S677, Update remaining resources r←rq j Update allocated resources p←p+q j ;

[0061] S678, Index j points to the next resource block j←j+1;

[0062] S679. Repeat steps S6706-S6708 until the allocated resources meet the demand q≥s;

[0063] S6710, Calculate the actual point cloud data compression rate k of the system pointed to by index i. i ←(∑ j q j x ij ) / |X s |;

[0064] S6711, Index i points to the next system i←i+1

[0065] S6712. Repeat steps S6703-S6711 until all requests from the target system are satisfied (i>n). R .

[0066] Compared with existing technologies, the advantages of this invention are as follows: It considers driving risks through a point cloud optimization representation method based on a safety field, significantly reducing the communication resource consumption required for data sharing in the fusion perception system while retaining point cloud information crucial for the perception task; it uses simulation experiments to calibrate the target point cloud data compression rate and introduces a delay-compensated positioning error calculation method based on communication delay estimation to evaluate the accuracy of fusion perception; it comprehensively considers driving risks and communication load, achieving a balance between point cloud data size and perception accuracy. The system adaptively configures communication resources and point cloud compression rate for systems requesting data under different bandwidth-limited scenarios, thereby improving the performance of the fusion perception system. Attached Figure Description

[0067] Figure 1 A diagram illustrating the decoupling method for medical image generation in the road-vehicle intelligent fusion perception method according to the present invention, which collaboratively considers driving risks and communication loads.

[0068] Figure 2A diagram illustrating the decoupling method for medical image generation in the road-vehicle intelligent fusion perception method according to the present invention, which collaboratively considers driving risks and communication loads.

[0069] Figure 3 A diagram illustrating the decoupling method for medical image generation in the road-vehicle intelligent fusion perception method according to the present invention, which collaboratively considers driving risks and communication loads.

[0070] Figure 4 This invention provides a decoupling mode for the intelligent fusion perception method for road-vehicle systems that collaboratively considers driving risks and communication load, based on the present invention.

[0071] Figure 5 A schematic diagram of segmentation on a vestibular schwannoma dataset, illustrating an embodiment of the intelligent fusion perception method for road-vehicle systems that collaboratively considers driving risks and communication load according to the present invention.

[0072] Figure 6 A schematic diagram of segmentation on the kidney KiTS19 dataset for an embodiment of the intelligent fusion perception method for road-vehicle co-considering driving risks and communication load according to the present invention;

[0073] Figure 7 This is a schematic diagram of segmentation on the kidney KiTS19 dataset, illustrating an embodiment of the intelligent fusion perception method for road-vehicle systems that collaboratively considers driving risks and communication load according to the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] Reference Figure 1-7 A road-vehicle intelligent fusion perception system that collaboratively considers driving risks and communication load is proposed. This system deploys vehicle-side subsystems and roadside subsystems, and establishes a simulation test environment. The specific steps are as follows:

[0076] The connected vehicle deploys an onboard computing unit, an onboard storage unit, and an onboard C-V2X communication unit; it also installs a lidar and a satellite-inertial navigation combined positioning system to form an onboard sensing unit; all of these together constitute the vehicle-to-vehicle terminal system. The onboard positioning equipment has the following accuracy: x-direction positioning error 0.05m, y-direction positioning error 0.05m, z-direction positioning error 0.05m, and heading angle error 0.085°. The onboard lidar parameters are as follows: 32 radar beams, sensing range 50m, and 1×10^12 point cloud points per frame. 6Data frame rate 20Hz, vertical field of view -30.0 to 10.0°.

[0077] A roadside sensing unit is constructed by erecting a pole, installing a lidar unit, fixing its position and orientation, and calibrating its installation position and orientation. A roadside computing unit and a roadside C-V2X communication unit are also deployed. These together constitute the roadside subsystem. The lidar configuration and calibration parameters are as follows: installation height 6m, pitch angle 0°, roll angle 0°, radar beamwidth 64, sensing range 120m, and point cloud point count per frame 2×10⁻⁶. 6 Data frame rate 20Hz, vertical field of view -25.0 to 2.0°.

[0078] Optionally, a data center can be established, and cloud servers and cloud databases can be deployed to realize log data storage and perception data analysis and calculation; communication with vehicle subsystems or roadside subsystems can realize computing and storage support;

[0079] The subsystems of one implementation of the road-vehicle intelligent fusion perception system that collaboratively considers driving risks and communication load are as follows: Figure 2 The system composition and information interaction between subsystems are illustrated as follows: Figure 3 .

[0080] Furthermore, a simulation test environment was built using Carla, specifically Car Learning to Act, with the Town10 map selected as the simulation test scenario. Figure 4 SUMO (Simulation of Urban Mobility) is used to manage traffic flow in the scenario. A path configuration file is written, background vehicles are generated, and scale traffic parameters are set. The simulation is run, and the ground truth position values ​​of each target vehicle in the scenario are collected through the simulation platform. The position data of the target vehicles detected by the onboard subsystem are also collected. The data transmission latency is calculated based on the size of the point cloud data collected by the roadside LiDAR generated in the simulation and the data transmission bandwidth limitations in the application requirements. Latency compensation is applied to the ground truth position values ​​of the target vehicles based on the data transmission latency, and the results are compared with the ground truth position values ​​of each target vehicle collected by the simulation platform to calculate the average latency-compensated positioning error. The results are illustrated in the figure below. Figure 5 Set several roadside point cloud compression rates within the 0-1 range to obtain the time delay compensation positioning error-point cloud compression rate curve, as shown below. Figure 6 Comparing random sampling, weighted threshold sampling, and weighted random sampling, the weighted random sampling method is used in this embodiment. The target detection average accuracy-point cloud compression ratio curve is shown below. Figure 7 .comprehensive Figure 6 , Figure 7 Set the target point cloud compression rate k t =0.5.

[0081] As another aspect of the present invention, a method for intelligent fusion perception of road and vehicle that collaboratively considers driving risks and communication load is as follows: Figure 1 The specific steps are as follows:

[0082] S1: Point cloud data with time and pose labels are collected by the sensing unit of the roadside subsystem;

[0083] S2: Optional, the roadside communication unit acquires collaborative sensing data from other systems through the road-vehicle communication network; the roadside computing unit fuses the point cloud data;

[0084] S3: Three-dimensional target detection is performed by the roadside computing unit;

[0085] S4: The vehicle-mounted communication unit requests collaborative sensing data from the roadside subsystem via the road-vehicle communication network;

[0086] S5: The roadside communication unit receives collaborative sensing data requests from the vehicle terminal system via the road-vehicle communication network;

[0087] S6: The roadside computing unit assesses the driving risks within the sensing area based on the safety field theory;

[0088] S7: The roadside computing unit comprehensively considers driving risks and communication load, and allocates the communication resources available in the roadside subsystem among the data requests received from the vehicle subsystem;

[0089] S8: The roadside computing unit optimizes and represents the point cloud collected by the roadside sensing unit according to the resource allocation scheme described in S7;

[0090] S9: The roadside communication unit sends collaborative sensing data to the terminal systems of each vehicle that requested collaborative sensing data according to the resource allocation scheme described in S7;

[0091] S10: The vehicle-mounted communication unit obtains cooperative sensing data from the roadside subsystem through the road-vehicle communication network;

[0092] S11: The vehicle-mounted computing unit fuses point cloud data and performs 3D target detection;

[0093] Furthermore, step S1 specifically includes the following steps:

[0094] S1.1: Based on the external parameters of the roadside sensing units calibrated during system deployment, record the system timestamp t at the sampling time. e and the position and orientation T of the sensing unit coordinate system in the WGS-84 coordinate system. e ;

[0095] S1.2: Point cloud data X is collected by lidar or other sensors deployed in the roadside sensing unit.e Record the x, y, and z coordinates of each data point in the local coordinate system of the roadside sensing unit;

[0096] S1.3: Combine the positioning data described in S1.1 and the point cloud data described in S1.2 to obtain point cloud data P with time and pose labels. e ={t e T e X e}

[0097] Furthermore, step S3 specifically includes the following steps:

[0098] S3.1: The roadside subsystem performs spatial transformation on the point cloud data acquired by all other systems in the collaborative sensing data receiving set based on the pose labels of the point cloud collected by the sensing unit and the pose labels of the point cloud data obtained from other systems, to obtain the collaborative sensing data receiving set in the coordinate system of the sensing unit.

[0099] S3.2: Combine all point clouds in the collaborative sensing dataset after the stitching transformation with the point cloud data collected by the sensing units to obtain the fused point cloud in the sensing unit coordinate system;

[0100] S3.3: Input the fused point cloud obtained by stitching into the trained PointPillars model to obtain the target-level environmental perception dataset. This includes the x, y, and z coordinates of obstacles around the system in the sensing unit coordinate system, as well as the length, width, and height data of the bounding box, denoted as... Where n Y To detect the number of targets.

[0101] Furthermore, step S5 specifically includes the following steps:

[0102] S5.1: Wait for a collaborative sensing data request from the vehicle terminal system within 15ms, and then receive positioning data T from the other system. i and the other party's system identifier m i ;

[0103] S5.2: In the communication window described in S5.1, a total of n... R The collaborative sensing data requests from other systems are combined to obtain a collaborative sensing data request set R = {(m i T i | i = 1, 2, ..., n R}

[0104] Furthermore, step S6 specifically includes the following steps:

[0105] S6.1: Define the reference points x0 and y0 for safety field calculation in the sensing unit coordinate system; the safety field calculation resolution Δ x Δ y Safety field calculation range n x n y ;

[0106] S6.2: Define the transformation relationship between the rectangular coordinate system coordinates (x, y, z) and the ellipsoidal coordinate system coordinates (ξ, η, ζ) of the roadside sensing unit;

[0107] S6.3: Generate the set of safe field calculation points H in the sensing unit coordinate system according to the safe field calculation reference point, safe field calculation resolution, and safe field calculation range defined in step S6.1;

[0108]

[0109] S6.4: For each safe field calculation point in the set of safe field calculation points H, calculate the ellipsoidal coordinates (ξ, η, ζ) corresponding to the rectangular coordinates (x, y, z) of the safe field calculation point according to the coordinate transformation relationship defined in step S6.3;

[0110] S6.5: For target-level environmental perception datasets For each detected target vehicle χ, a safety field calculation is performed on the safety field calculation point set defined in S6.3;

[0111] S6.6: Superimpose the safety fields generated by each detected target vehicle in the target-level environmental perception dataset to obtain the summed safety field;

[0112] S6.7: Normalize the total security field to obtain the standard total security field matrix F;

[0113] S6.8: Based on the standard summation safety field matrix F, add a driving risk assessment item to the collaborative perception data request set generated in step S5.2 to obtain a collaborative perception data request set with driving risk assessment.

[0114] Furthermore, step S6.5 specifically includes the following steps:

[0115] S6.5.1: Define the safety field calculation parameters k1 = 10, k2 = 1, R χ =1;

[0116] S6.5.2: Define and calculate the virtual mass M of the target vehicle χ. χ As shown in equation 9, where |v χ |Vehicle speed, vehicle mass (kg) m χ =3000, Model Correction T χ=1, model coefficient a = 1.566 × 10 -14 b = 0.3345, d = 6.687;

[0117]

[0118] S6.5.3: For the set of coordinate points H for calculating the safety field described in step S6.3 χ The coordinates (x, y, z) of each safety field calculation point in the sensing unit's rectangular coordinate system and its coordinates in the sensing unit's ellipsoidal coordinate system as described in step S6.4. Calculate the field strength value of the safety field generated by the target vehicle χ at that location according to Equations 2 and 3.

[0119] Furthermore, step S6.8 specifically includes the following steps:

[0120] S6.8.1: The positioning data T in the global coordinate system sent by each system in the collaborative sensing data request set R in step S1.1. i Perform coordinate transformation to obtain the coordinates of each requested data's system location in the sensing unit coordinate system of this system;

[0121] S6.8.2: Query the index p of the point in the safe field calculation point set H that is close to the location of the requested data system. k And take the safe field strength value at that point. As the result of the system's driving risk assessment;

[0122] S6.8.3: Combine the driving risk assessment results of all requesting data systems to generate a driving risk assessment set for each requesting collaborative perception data system.

[0123] Further step S7 specifically includes the following steps:

[0124] S7.1: Detect the available communication resources of the roadside subsystem, including: the number of available subframes L, the number of subchannels K, the subframe width t, the subchannel width b, and the signal-to-noise ratio (SINR) of the subchannel. i Define the available resource sequence q = [c i t|c i = b·log(1+SINR) i ), i = 1, 2, ..., LK];

[0125] S7.2: Calculate the size of the point cloud data collected by the sensing unit in step S1.2 |X s |;

[0126] S7.3: Define the compression ratio multiplier λ = 2;

[0127] S7.4: Sort the systems in the collaborative sensing data request set according to the driving risk items in the collaborative sensing data request set with driving risk assessment to obtain the collaborative sensing data request sequence R. s Length n R ;

[0128] S7.5: Define the resource allocation matrix and assign initial values. Define a point cloud data compression ratio vector and assign initial values.

[0129] S7.6: Define communication resource requirements and assign initial values ​​s = |X s |k t Define the remaining allocable resource size and initialize it with r = ∑ i q i ;

[0130] S7.7: Allocate communication resources from the resource sequence to each system in the collaborative sensing data request sequence in sequence until the resources are allocated or all requirements are met;

[0131] S7.8: Record the compression ratio vector k and resource allocation matrix x of the final point cloud data.

[0132] Furthermore, step S7.7 specifically includes the following steps:

[0133] S7.7.1: Define the target system index i, which points to the first system in the collaborative sensing data request sequence;

[0134] S7.7.2: Define communication resource index j, which points to the first resource block in the available resource sequence;

[0135] S7.7.3: If resource allocation is complete and r≤0, exit the resource allocation procedure;

[0136] S7.7.4: If the remaining communication resource size does not meet the requirement r < s, increase the compression ratio s←s / λ according to the compression ratio multiplication factor described in step S64;

[0137] S7.7.5: Define the system communication resource allocated to index i and assign it an initial value p = 0;

[0138] S7.7.6: Allocate the resource pointed to by index j to the system x pointed to by index i. i，j ←1;

[0139] S7.7.7: Update remaining resources r←rq j Update allocated resources p←p+q j ;

[0140] S7.7.8: Index j points to the next resource block j←j+1;

[0141] S7.7.9: Repeat steps S7.7.6-S7.7.8 until the allocated resources meet the demand q≥s;

[0142] S7.7.10: Calculate the actual point cloud data compression rate k of the system pointed to by index i. i ←(∑ j q j x ij ) / |X s |;

[0143] S7.7.11: Index i points to the next system i←i+1;

[0144] S7.7.12: Repeat steps S7.7.3-S7.7.11 until all requests from the target system are satisfied (i>n). R .

[0145] Furthermore, step S8 specifically includes the following steps:

[0146] S8.1: Collect point cloud data X from roadside sensing units e At each point in the set H, a security field calculation is performed to obtain weighted point cloud data. Where x i y i Let ω be the coordinates of the sampling point in the sensing unit coordinate system. i For the weight, n P This represents the total number of sampling points;

[0147] S8.2: The weights of the weighted point cloud data are normalized;

[0148] S8.3: Adjust the weights of the weighted point cloud data according to the point cloud compression ratio vector k, as shown in Equation 5, where ξ = 1;

[0149] S8.4: Use safe field weighted probability sampling, as shown in Equation 7;

[0150] Further step S9 specifically includes the following steps:

[0151] S9.1: Establish communication with the vehicle-side system within the cooperative sensing data request set on the channel allocated by the resource allocation matrix x;

[0152] S9.2: Send compressed point cloud to each vehicle-end system within the data request set.

[0153] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention, and applications, modifications and variations thereof will be apparent to those skilled in the art.

[0154] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load, comprising a perception system, the system including a vehicle-end platform, a roadside platform connected to the vehicle-end platform via signals, a cloud platform communicating with the vehicle-end platform or the roadside platform, positioning and attitude determination equipment, and point cloud acquisition equipment; the vehicle-end platform includes connected vehicles, which include perception units, communication units, and computing units; the roadside platform includes road infrastructure including poles and gantries, deploying perception units, communication units, and computing units; the cloud platform includes cloud servers and cloud databases, further including computing units and storage units, characterized in that... The method includes the following steps: S1. Collect point cloud data with time and pose labels through the sensing unit; S2. The communication unit acquires collaborative sensing data from other systems through the road-vehicle communication network; S3. The computing unit fuses point cloud data and performs 3D target detection; S4. The communication unit receives collaborative sensing data requests from other systems through the road-vehicle communication network. S5. The computing unit assesses driving risks within the perception area based on the safety field theory. S6. The computing unit comprehensively considers driving risks and communication load, and allocates the available communication resources of the system among the received data requests. S7. The computing unit optimizes and represents the point cloud collected by the sensing unit according to the resource allocation scheme in step S6. S8. The communication unit sends collaborative sensing data to each system that requests collaborative sensing data according to the resource allocation scheme in step S6.

2. The road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load as described in claim 1, characterized in that, Step S5 specifically includes the following steps: S51. Define the reference point, resolution, and range of the safe field calculation in the coordinate system of the sensing unit. S52. Define the rectangular coordinate system of the sensing unit. ellipsoidal coordinate system coordinates Transformation relationships between ) S53. Generate a set of safe field calculation points in the sensing unit coordinate system according to the safe field calculation reference point, safe field calculation resolution, and safe field calculation range defined in S51. ; S54, For the set of calculation points in the safety field For each safety field calculation point, the rectangular coordinates of the safety field calculation point are calculated according to the coordinate transformation relationship defined in S53. Corresponding ellipsoidal coordinates ( ); S55. For target-level environmental perception datasets Each detected target vehicle Safety field calculations are performed on the safety field calculation point set defined in S53. S56. Superimpose the safety fields generated by each detected target vehicle in the target-level environmental perception dataset to obtain the summed safety field; S57. Normalize the total security field to obtain the standard total security field matrix. ; S58. According to the standard summation safety field matrix In step S4, a driving risk assessment item is added to the generated collaborative perception data request set to obtain a collaborative perception data request set with driving risk assessment. .

3. The road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load as described in claim 2, characterized in that, Step S55 specifically includes the following steps: S551, Define safety field calculation parameters , ; S552. Define and calculate the target vehicle. virtual quality ; S553, Define and calculate the target vehicle The road conditions at the location Influencing factors include the road adhesion coefficient, road slope, road curvature, and air visibility in the vicinity of the target vehicle. S554. For the set of coordinate points for calculating the safety field described in S53 The coordinates of each safety field calculation point in the sensing unit's rectangular coordinate system are... And its coordinates in the sensing unit ellipsoidal coordinate system in S54 ( ), calculate the target vehicle according to formula (2) and formula (3) The electric field strength value of the generated safety field at that location; （2） （3） in, Representative object With point Connection to target vehicle velocity vector The angle between them.

4. The road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load as described in claim 3, characterized in that, Step S58 specifically includes the following steps: S581, Request the collaborative sensing data set in step S4 The positioning data in the global coordinate system sent by each system in step S1 Perform coordinate transformation to obtain the coordinates of each requested data's system location in the sensing unit coordinate system of this system; ， （4） S582, the set of safety field calculation points described in S53 The query finds the index of points close to the location of the requested data system. And take the safe field strength value at that point. As the result of the system's driving risk assessment; S583. Combine the driving risk assessment results of all requesting data systems to generate a driving risk assessment set for each requesting collaborative perception data system.

5. The road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load as described in claim 4, characterized in that, Step S6 specifically includes the following steps: S61. Communication resources include: the number of available subframes. Number of sub-channels Subframe width Sub-channel width Define the available resource sequence ,in For channel Signal-to-noise ratio; S62. Calculate the size of the point cloud data collected by the sensing unit in step S1. ; S63. Before system deployment, determine the target point cloud data compression rate through simulation experiments. ; S64. Define the compression ratio multiplier. Define the resource allocation matrix and assign initial values. Define the point cloud data compression ratio vector and assign initial values. Define communication resource requirements and assign initial values. Define the remaining allocable resource size and assign an initial value. ; S65. Sort the systems in the collaborative perception data request set according to the driving risk items in the collaborative perception data request set with driving risk assessment as described in S58, to obtain the collaborative perception data request sequence. ,length ; S66. Allocate communication resources from the resource sequence in step S61 to each system in the collaborative sensing data request sequence, allocating resources sequentially until resource allocation is complete. Or all needs are met; S67. Record the compression ratio vector of the final point cloud data. and resource allocation matrix .

6. The road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load as described in claim 5, characterized in that, Step S63 specifically includes the following steps: S631. Establish a simulation platform, including traffic flow simulation, sensor simulation, and vehicle dynamics simulation modules; S632. According to application requirements, calibrate the parameters of each model in the simulation platform, establish a simulation scene, and build the perception unit of step S1 in the simulation platform; specifically including: establishing a scene map, configuring vehicle parameters, configuring sensor parameters, and setting scene traffic flow; S633. Set a fixed sampling rate and sample the point cloud data collected by the sensing unit in the simulation as described in step S7. S634. A method for fusing simulated point cloud data and performing 3D target detection as defined in step S3; S635. Run the simulation and collect the true position values ​​of each target vehicle in the scene through the simulation platform, and collect the target vehicle position data detected by the system in S3. S636. Based on the fixed sampling rate described in step S634, the data transmission delay is calculated using the size of the point cloud data collected by the sensing unit of the simulation platform in step S632 and the data transmission bandwidth limit in the application requirements. S637. Perform time delay compensation on the true position value of the target vehicle based on the data transmission delay, and compare it with the true position value of each target vehicle in the scene collected by the simulation platform to calculate the positioning error of the average time delay compensation. S638. Update the fixed sampling rate. Repeat steps S634-S638 to optimize the average positioning error until the fixed sampling rate or the average positioning error converges.

7. The road-vehicle intelligent fusion perception method that collaboratively considers driving risks and communication load as described in claim 6, characterized in that, Step S67 specifically includes the following steps: S671, Define the target system index , refers to the first system in the collaborative sensing data request sequence described in S65; S672, Define Communication Resource Index , which points to the first resource block of the available resource sequence described in S61; S673, If resource allocation is complete Exit the resource allocation process; S674. If the remaining communication resources do not meet the requirements. Increase the compression ratio according to the compression ratio multiplier described in S64. ; S675, Definition already assigned to index The system communication resources are specified and initialized. ; S676, Index The resource is allocated to the index. The system referred to ; S677, Update remaining resources Update allocated resources ; S678, Index Point to the next resource block ; S679. Repeat steps S676-S678 until the allocated resources meet the demand. ; S6710, Calculate the index The actual point cloud data compression rate of the system referred to ; S6711, Index Pointing to the next system ; S6712. Repeat steps S673-S6711 until all requests from the target system are satisfied. .

Images