Data processing method, device and equipment, medium and vehicle

Abstract

The invention provides a data processing method, device and equipment, a medium and a vehicle, and relates to the technical field of automobile environment perception. The method comprises the following steps: acquiring current Internet of Vehicles data and various vehicle-mounted sensor data of a vehicle; performing credibility evaluation on the current Internet of Vehicles data to obtain a credibility score of the current Internet of Vehicles data, and performing credibility evaluation on the various vehicle-mounted sensor data to obtain respective credibility scores of the various vehicle-mounted sensor data; and fusing the current Internet of Vehicles data and the various vehicle-mounted sensor data by taking the respective credibility scores of the current Internet of Vehicles data and the various vehicle-mounted sensor data as fusion weights to obtain fused data. According to the invention, collaborative sensing fusion of the V2X and the vehicle-mounted sensor can be realized based on the dynamically adjusted fusion weight, so that the environmental adaptability of V2X collaborative sensing is enhanced, and the detection precision of the driving environment around the vehicle is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive environment perception, and particularly relates to a data processing method, device, equipment, medium and vehicle. Background Art

[0002] With the accelerating trend of intelligentization and networking in the automotive industry, the integration of V2X (Vehicle to Everything) technology and vehicle-mounted intelligence has become an irreversible trend. Cooperative perception based on V2X can break through the limitations of single-vehicle perception, enhance safety redundancy, and optimize group decision-making, which is a key technology for achieving high-level autonomous driving. However, the current cooperative perception technology based on V2X and vehicle-mounted sensors performs cooperative perception fusion based on preset fixed weights, resulting in technical defects such as poor environmental adaptability and low detection accuracy for the driving environment around the vehicle in the related V2X cooperative perception system. Summary of the Invention

[0003] The present invention provides a data processing method, device, equipment, medium and vehicle, aiming to overcome or at least partially solve the above problems.

[0004] In the first aspect of the present invention, a data processing method is provided, and the method includes: Obtain the current vehicle networking data of the vehicle itself and a variety of vehicle-mounted sensor data, where the current vehicle networking data represents the detection results of other vehicles around the vehicle itself on the driving environment around the vehicle itself, and the variety of vehicle-mounted sensor data represents the detection results of a variety of vehicle-mounted sensors of the vehicle itself on the driving environment around the vehicle itself; Perform a credibility evaluation on the current vehicle networking data to obtain a credibility score of the current vehicle networking data, and perform a credibility evaluation on the variety of vehicle-mounted sensor data respectively to obtain the credibility scores of the variety of vehicle-mounted sensor data; Use the credibility scores of the current vehicle networking data and the variety of vehicle-mounted sensor data as fusion weights to fuse the current vehicle networking data and the variety of vehicle-mounted sensor data, and obtain fused data, where the fused data represents the final detection result of the driving environment around the vehicle itself.

[0005] In the second aspect of the present invention, a data processing device is provided, and the device includes: A data acquisition module, configured to obtain the current vehicle networking data of the vehicle itself and a variety of vehicle-mounted sensor data, where the current vehicle networking data represents the detection results of other vehicles around the vehicle itself on the driving environment around the vehicle itself, and the variety of vehicle-mounted sensor data represents the detection results of a variety of vehicle-mounted sensors of the vehicle itself on the driving environment around the vehicle itself; A scoring determination module for performing credibility evaluation on the current vehicle networking data to obtain a credibility score of the current vehicle networking data, and separately performing credibility evaluation on the multiple vehicle-mounted sensor data to obtain credibility scores of the multiple vehicle-mounted sensor data respectively; A data fusion module for fusing the current vehicle networking data and the multiple vehicle-mounted sensor data by using the credibility scores of the current vehicle networking data and the multiple vehicle-mounted sensor data respectively as fusion weights to obtain fused data, where the fused data represents the final detection result of the driving environment around the vehicle.

[0006] A third aspect of the present invention provides an electronic device, which includes: a memory, a processor, and a computer program stored on the memory and running on the processor, where when the computer program is executed by the processor, it implements the data processing method according to the first aspect of the present invention.

[0007] A fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the data processing method according to the first aspect of the present invention.

[0008] A fifth aspect of the present invention provides a vehicle, which at least includes a controller, and when the controller is executed, it implements the data processing method according to the first aspect of the present invention.

[0009] In the data processing method provided by the present invention, credibility evaluation is separately performed on the current vehicle networking data and multiple vehicle-mounted sensor data of the vehicle to obtain credibility scores of the current vehicle networking data and the multiple vehicle-mounted sensor data respectively, and the current vehicle networking data and the multiple vehicle-mounted sensor data are fused by using the credibility scores as fusion weights to obtain fused data that can represent the final detection result of the driving environment around the vehicle. Thus, the present invention performs real-time quantitative evaluation on the credibility scores of multi-modal perception data of the vehicle, and realizes the collaborative perception fusion of V2X and vehicle-mounted sensors based on dynamically adjusted fusion weights, thereby enhancing the environmental adaptability of V2X collaborative perception and improving the detection accuracy of the driving environment around the vehicle. Description of the Drawings

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts.

[0011] Figure 1 is a flowchart of a data processing method shown in an embodiment of the present invention; Figure 2 It is the overall system architecture diagram of a multi-modal data fusion system based on dynamic credibility assessment shown in an embodiment of the present invention; Figure 3 It is the flow chart of a three-level conflict arbitration mechanism shown in an embodiment of the present invention; Figure 4 It is the architecture diagram of a federated learning model optimization shown in an embodiment of the present invention; Figure 5 It is the structural block diagram of a data processing device provided in an embodiment of the present invention; Figure 6 It is the schematic diagram of an electronic device shown in an embodiment of the present invention. Detailed implementation manners

[0012] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand the other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for explaining the present invention, rather than limiting the protection scope of the present invention.

[0013] In one embodiment, referring to Figure 1 , Figure 1 It is the flow chart of a data processing method shown in an embodiment of the present invention. As Figure 1 shown, the data processing method in this embodiment may include the following steps: Step S11: Obtain the current vehicle networking data and various in-vehicle sensor data of the vehicle.

[0014] In this embodiment, the current vehicle networking data of the vehicle itself can be obtained, and various on-vehicle sensor data of the vehicle can also be obtained. Among them, the current vehicle networking data is the vehicle networking data (V2X data) at the current moment, and the vehicle networking data includes at least any one or more of the following: data such as the positions, speeds, and heading angles of surrounding vehicles carried in BSM (Basic Safety Message), CAM (Cooperative Awareness Message), and DENM (Decentralized Environmental Notification Message); the current vehicle networking data represents the detection results of other vehicles around the vehicle on the driving environment around the vehicle. The various on-vehicle sensor data are the sensor data collected in real time by various on-vehicle sensors of the vehicle, and the various on-vehicle sensor data represent the detection results of various on-vehicle sensors of the vehicle on the driving environment around the vehicle.

[0015] Step S12: Evaluate the credibility of the current vehicle networking data to obtain a credibility score of the current vehicle networking data, and evaluate the credibility of the various on-vehicle sensor data respectively to obtain the credibility scores of the various on-vehicle sensor data.

[0016] In this embodiment, independent credibility evaluations can be performed on the current vehicle networking data and the various on-vehicle sensor data of the vehicle respectively to obtain the credibility score of the current vehicle networking data and the credibility scores of the various on-vehicle sensor data respectively, which can be used as the basis for subsequent weight allocation. Among them, the credibility score in this embodiment represents the reliability of the current vehicle networking data and / or the various on-vehicle sensor data, and is a key indicator for quantifying the reliability of the current vehicle networking data and / or the various on-vehicle sensor data, and is used to evaluate the confidence level of communication messages with surrounding vehicles and / or the measurements of on-vehicle sensors of the vehicle. In this way, this embodiment establishes a cross-modal and multi-factor fusion credibility quantification system, breaking through the limitations of traditional single-index evaluations.

[0017] Step S13: Use the credibility scores of the current vehicle networking data and the various on-vehicle sensor data as fusion weights to fuse the current vehicle networking data and the various on-vehicle sensor data to obtain fused data.

[0018] In this embodiment, after obtaining the credibility scores of the current vehicle networking data and the credibility scores of various vehicle-mounted sensor data respectively, the corresponding fusion weights can be assigned to the current vehicle networking data and various vehicle-mounted sensor data respectively based on the credibility scores of the current vehicle networking data and various vehicle-mounted sensor data, so as to fuse the current vehicle networking data and various vehicle-mounted sensor data based on the corresponding fusion weights of the current vehicle networking data and various vehicle-mounted sensor data, and obtain the fused data, which represents the final detection result of the driving environment around the vehicle.

[0019] In this embodiment, by performing real-time quantitative evaluation on the credibility scores of the multi-modal perception data (current vehicle networking data and various vehicle-mounted sensor data) of the vehicle, and fusing the multi-modal perception data based on the dynamically adjusted fusion weights, the collaborative perception fusion of V2X and vehicle-mounted sensors is realized, thereby enhancing the environmental adaptability of V2X collaborative perception and improving the detection accuracy of the driving environment around the vehicle.

[0020] Combined with the above embodiments, in one implementation manner, the above step S11 may specifically include three aspects: data access, data preprocessing and standardization, and preliminary screening of data quality.

[0021] Among them, data access: For the current vehicle networking data of the vehicle, it can support DSRC / C-V2X dual-mode communication, and parse the structured data in the obtained BSM, CAM, DENM and other messages, including vehicle position, speed, heading angle and other information; at the same time, real-time extract the communication quality parameters between the vehicle and the vehicle networking, such as signal strength (RSSI), signal-to-noise ratio (SINR) and packet loss rate; for vehicle-mounted sensor data: in the case where the vehicle-mounted sensor data includes a vehicle-mounted camera, the original image and target detection results (bounding box + classification confidence) collected by the vehicle-mounted camera can be obtained; in the case where the vehicle-mounted sensor data includes a millimeter-wave radar, the millimeter-wave radar point cloud and target tracking list can be obtained; in the case where the vehicle-mounted sensor data includes a lidar, 3D point cloud data can be obtained.

[0022] Data preprocessing and standardization: Format unification: For the current vehicle networking data of the vehicle, convert the current vehicle networking data into a standard JSON format (including timestamp, coordinate system, data source identifier); for vehicle-mounted sensor data, convert the vehicle-mounted sensor data into the Autoware general interface format, for example, point cloud → PCD, image → ROS Image, etc. Perform spatio-temporal alignment on the data: Use the PTP protocol (such as accuracy <1μs) to align the time bases of each data source to achieve time synchronization; unify the coordinate systems of different vehicle-mounted sensors to the vehicle coordinate system through the calibration matrix (conversion error <5cm) to achieve coordinate conversion.

[0023] Initial screening of data quality: Perform anomaly detection on the data. For example, identify illegal values in the current vehicle networking data (such as speed > 200 km / h), identify jumps in in-vehicle sensor data (such as acceleration mutation > 10 m / s²), etc.; perform priority marking. For example, mark the highest priority (Level 0) for emergency events (such as accident warnings in the current vehicle networking data), etc.

[0024] Combined with the above embodiments, in one implementation, the embodiments of the present invention further provide a data processing method. In this method, the "evaluate the credibility of the current vehicle networking data to obtain the credibility score of the current vehicle networking data" in step S12 above may specifically include steps S21 to S24: Step S21: Obtain the communication quality parameter, and obtain the communication quality score of the current vehicle networking data based on the communication quality parameter.

[0025] In this embodiment, obtain the communication quality parameter between the vehicle and the vehicle networking, and may determine the communication quality score of the current vehicle networking data based on this communication quality parameter. This communication quality score characterizes the communication quality of the transmission link of the current vehicle networking data. The higher the communication quality score, the more reliable the transmission link.

[0026] In an optional example, the communication quality parameter includes the signal-to-interference-plus-noise ratio (SINR). The communication quality score of the current vehicle networking data can be determined by the following formula: , where Sigmoid() is the Sigmoid function. Among them, the signal-to-interference-plus-noise ratio reflects the reliability of the transmission link of the current vehicle networking data. The SINR value can be extracted in real time through the physical layer chip (such as C-V2X OBU). In practical applications, the SINR range in vehicle networking data communication is usually between -10 dB and 30 dB. The specific value depends on the communication environment and application scenarios. The Sigmoid function is used to non-linearly map SINR (range -10~30 dB) to [0,1]. For example, an inflection point is set at the critical value of 10 dB (it is generally considered that the communication quality is poor when SINR < 10 dB, and a connection may not be established or communication may not be maintained. The specific inflection point value can be dynamically adjusted according to the actual communication environment). When SINR < 10 dB, the weight decays exponentially to reflect the dynamic impact on the credibility of V2X messages when the communication quality changes.

[0027] Step S22: Obtain the time difference between the sending time and the receiving time of the current vehicle networking data, and obtain the timeliness decay score of the current vehicle networking data based on the time difference and the decay rate.

[0028] In this embodiment, the sending time and receiving time of the current vehicle networking data can be determined, so as to determine the time difference between the sending time and the receiving time of the current vehicle networking data. Then, based on this time difference and a preset attenuation rate (for example, different attenuation rates can be preset based on different scenarios), the aging attenuation score of the current vehicle networking data is determined. This aging attenuation score characterizes the attenuation of the "freshness" of the current vehicle networking data over time, so as to avoid the interference of expired data.

[0029] In an alternative example, the aging attenuation score of the current vehicle networking data can be determined by the following formula: . Where Δt is the time difference between the sending time and the receiving time of the current vehicle networking data (unit: second), and the attenuation rate is controlled by λ, where λ is a control parameter. In a typical scenario, λ = 0.3 can be set (when Δt = 2 seconds → e^{-0.6}≈0.55), then the credibility weight of the message received 2 seconds ago is reduced by 45%. For special scenarios, the value of λ can be dynamically adjusted to reflect the requirements of the special scenario for the timeliness of the current vehicle networking data. For example, in a high-speed scenario, the high-speed movement of the vehicle requires low latency, and λ = 0.5 can be set (attenuation is faster); in a congested scenario, the vehicle moves slowly, and λ = 0.2 can be set (attenuation is slower).

[0030] Step S23: Perform consistency analysis on the current vehicle networking data and the multiple in-vehicle sensor data, and obtain the consistency score of the current vehicle networking data based on the result of the consistency analysis.

[0031] In this embodiment, the current vehicle networking data can be analyzed for consistency by using multiple in-vehicle sensor data, and the result of the consistency analysis is obtained. Then, based on the result of the consistency analysis, the consistency score of the current vehicle networking data is determined.

[0032] In an alternative example, the local in-vehicle sensor data (for example, the point cloud data collected by lidar) can be used to verify the historical consistency of the current vehicle networking data. For example, the consistency analysis is performed on the current vehicle networking data through three consecutive frames of vehicle networking data and its own in-vehicle sensor data (for example, when the average value of the vehicle position error is less than 1.5 meters, the consistency score can be considered as H = 0.9).

[0033] For example, the consistency score of the current vehicle networking data is determined by the following formula : .

[0034] Where N is the number of statistical frames (usually N = 3 can be taken), δ is the error threshold (δ = 1.5 meters can be taken); Θ is an indicator function (1 when the error ≤ δ, otherwise 0); k represents the kth frame of data; is the target position information obtained through the vehicle networking, It is the same target position information obtained by the in-vehicle sensors of the own vehicle. The specific mathematical expression can use the two-dimensional coordinates (x, y) of the target in the vehicle's own coordinate system; The specific calculation method is to calculate the two-dimensional spatial Euclidean distance error between two position points obtained by calculating the same target through the vehicle network and local sensors (in-vehicle sensors of the own vehicle). Assume The corresponding position coordinates are (x1, y1), The corresponding position coordinates are (x2, y2). Then the calculation formula of the above two-dimensional spatial Euclidean distance error can be explicitly transformed into: . If the error of three consecutive frames of data is less than 1.5 meters, then The value of is 1.0 (indicating completely credible); if two of the frames of data meet the standard, then The value of is 0.67.

[0035] Step S24: Obtain the credibility score of the current vehicle network data based on the communication quality score, time-effect attenuation score, and consistency score of the current vehicle network data.

[0036] In this embodiment, the credibility of the current vehicle network data can be evaluated based on the communication quality score of the current vehicle network data, the time-effect attenuation score of the current vehicle network data, and the consistency score of the current vehicle network data, and the credibility score of the current vehicle network data can be determined.

[0037] In an optional embodiment, in order to quantify the dynamic credibility of the current vehicle network data (such as a continuous value from 0 to 1) to accurately reflect its reliability in a specific scenario, different weights can be assigned to the communication quality score of the current vehicle network data, the time-effect attenuation score of the current vehicle network data, and the consistency score of the current vehicle network data, and the communication quality score of the current vehicle network data, the time-effect attenuation score of the current vehicle network data, and the consistency score of the current vehicle network data are weighted and summed to obtain the credibility score of the current vehicle network data As shown in the following formula: ; Among them, α, β, and γ are the weights corresponding to the communication quality score, the time - effect attenuation score, and the consistency score of the current vehicle - networking data respectively. α, β, γ ∈ [0, 1], and α + β + γ = 1. The determination of the weight coefficients α, β, and γ can first be initialized based on experience and application scenarios. For example, in a highway scenario where high timeliness is required, β needs to be increased; in a congestion scenario, communication stability is more critical, so α is prioritized; in vehicles with multi - sensor configurations, γ can take values in [0.3, 0.5], and in vehicles with low - equipped sensors, the weight of γ is reduced to take values in [0.1, 0.2]. For example, for the highway scenario, the following weight coefficients can usually be preset: α = 0.5, β = 0.3, γ = 0.2. On the premise of appropriately increasing the time - effect attenuation weight β, the communication quality is monitored with emphasis, and α = 0.5 is set; for another example, in the urban intersection cooperation scenario, the historical consistency of multi - vehicle trajectories should be emphasized, and it is recommended to set α = 0.3, β = 0.3, γ = 0.5. After setting the initial values for the three weight systems of the communication quality score, the time - effect attenuation score, and the historical consistency score based on expert experience and application scenario characteristics, the final weights α, β, and γ can be further obtained through on - vehicle roadside calibration.

[0038] It should be noted that this embodiment does not limit the execution order between the above - mentioned steps S21 to S23. For example, the execution order can be arbitrarily specified, or they can be executed simultaneously.

[0039] Combined with the above embodiments, in one implementation, a dynamic credibility scoring mechanism is proposed for vehicle - mounted sensors (such as vehicle - mounted cameras, vehicle - mounted millimeter - wave radars, vehicle - mounted lidars). A multi - dimensional quantization model is constructed by combining physical characteristics and environmental interference to ensure the accuracy and robustness of weight assignment during multi - modal fusion.

[0040] Combined with the above embodiments, in one implementation, the embodiment of the present invention also provides a data - processing method. In this method, the multiple vehicle - mounted sensor data at least includes image data collected by a vehicle - mounted camera; the step of "respectively performing credibility evaluation on the multiple vehicle - mounted sensor data to obtain the credibility scores of the multiple vehicle - mounted sensor data" in the above - mentioned step S12 can at least include steps S31 to S34: Step S31: Perform structural similarity analysis on the image data collected by the vehicle - mounted camera and the reference image data to obtain the structural similarity score of the vehicle - mounted camera.

[0041] For the credibility assessment of the image data collected by in-vehicle cameras, it is necessary to comprehensively consider the image quality and environmental interference. Based on this, in this embodiment, the structural similarity analysis can be performed on the image data collected by in-vehicle cameras and the reference image data to quantify the degree of image distortion and obtain the structural similarity score of the in-vehicle camera. Among them, the reference image data can be the reference background image data, and the reference image data can be set with different update mechanisms according to the weather conditions. For example, in sunny scenarios, the reference image data is updated every 5 minutes; in rainy and foggy scenarios, the interference areas are dynamically removed through motion target segmentation (such as optical flow method) to update the reference image data.

[0042] In an alternative embodiment, the structural similarity score of the in-vehicle camera can be determined by the following formula: ; where, is the mean of the local window of the image, reflecting the brightness; is the variance, reflecting the contrast; is the covariance, reflecting the structural similarity; is a constant to prevent the denominator from being zero. Let the image data collected by the in-vehicle camera be x, and let the reference image data be y, then the structural similarity score of the in-vehicle camera can be obtained .

[0043] Furthermore, for the constants and , an in-vehicle scene adaptive adjustment mechanism based on illumination and motion state can be added on the basis of the constant settings of the classic SSIM algorithm to achieve the physical consistency of credibility assessment. In a specific example: =(k1 * L)^2, =(k2 * L)^2; where, L: the dynamic range of pixel values (usually 8 bits for in-vehicle cameras, L = 255); k1, k2: empirical coefficients (by default, k1 = 0.01, k2 = 0.03).

[0044] In-vehicle scene adaptive adjustment strategy: Optimize the constants according to the special optical environment of the vehicle; for example, in tunnels and night scenes, with sudden illumination changes, increase c1 (adjust k1 = 0.02); in rainy and foggy weather, with low contrast, increase c2 (such as k2 = 0.05); in strong glare environments, with local overexposure, decrease c2 (such as k2 = 0.02).

[0045] Step S32: Perform edge detection on the image data collected by the in-vehicle camera, and obtain the clarity score of the in-vehicle camera according to the edge detection result.

[0046] In this embodiment, for the image data collected by the in-vehicle camera, the image focusing quality is also quantified to prevent motion blur from affecting the detection result. Based on this, edge detection is performed on the image data collected by the in-vehicle camera to obtain an edge detection result, and thus the clarity score of the in-vehicle camera is obtained according to the edge detection result.

[0047] In an alternative embodiment, considering that in image processing, the Laplacian operator is widely used for edge detection and image enhancement, and it detects edges by calculating the square of the gradient of each pixel point in the image. Therefore, the clarity score of the in-vehicle camera can be determined by the following formula : ; where, is the image data collected by the in-vehicle camera; Laplacian(): the Laplacian operator, which calculates the second derivative of the image, and the edge response of a clear image is higher; τ is the normalization threshold: the empirical value τ = 120 (if the Laplacian variance > 120, ≈ 1; if the Laplacian variance < 50, ≤ 0.4).

[0048] Step S33: Obtain the ambient light brightness of the driving environment around the vehicle, and obtain the light illumination score of the in-vehicle camera according to the ambient light brightness and the headlight on state.

[0049] In this embodiment, the ambient light brightness of the driving environment around the vehicle can be obtained, and the light illumination score of the in-vehicle camera is obtained according to the ambient light brightness and the headlight on state, so as to realize the adaptive adjustment of the credibility weight of the in-vehicle camera according to the ambient brightness.

[0050] In an alternative implementation manner, the light illumination score of the in-vehicle camera can be determined by the following calculation formula : ; where, the current illumination and illumination in the formula are the ambient light brightness of the driving environment around the vehicle.

[0051] Step S34: Based on the structural similarity score, clarity score, and light illumination score of the in-vehicle camera, obtain the credibility score of the image data collected by the in-vehicle camera.

[0052] In this embodiment, based on the structural similarity score of the in-vehicle camera, the clarity score of the in-vehicle camera, and the light illumination score of the in-vehicle camera, the credibility of the image data collected by the in-vehicle camera can be evaluated, and the credibility score of the image data collected by the in-vehicle camera is determined.

[0053] In an alternative embodiment, the confidence score of the image data collected by the vehicle-mounted camera can be obtained through the following formula , as shown in the following formula: ; Wherein, is the structural similarity score of the vehicle-mounted camera, is the illumination score of the vehicle-mounted camera, is the clarity score of the vehicle-mounted camera.

[0054] It should be noted that this embodiment does not limit the execution order between the above steps S31 to S33. For example: the execution order can be arbitrarily specified, or, executed simultaneously.

[0055] Combined with the above embodiments, in one implementation manner, the embodiment of the present invention further provides a data processing method. In this method, the multiple vehicle-mounted sensor data at least includes the current point cloud data collected by the vehicle-mounted millimeter-wave radar; and, "performing a confidence evaluation on the multiple vehicle-mounted sensor data respectively to obtain the confidence scores of the multiple vehicle-mounted sensor data" in the above step S12 can at least include steps S41 to S45: Step S41: Obtain the point cloud density score of the vehicle-mounted millimeter-wave radar according to the target point cloud number represented by the current point cloud data collected by the vehicle-mounted millimeter-wave radar and the average point cloud number represented by multiple first historical point cloud data.

[0056] In this embodiment, considering that the vehicle-mounted millimeter-wave radar detects targets by transmitting and receiving high-frequency electromagnetic waves, based on this, the confidence evaluation of the vehicle-mounted millimeter-wave radar needs to focus on the consistency of signal quality, environmental interference and physical laws. Specifically, the integrity of target detection can be evaluated first according to the target point cloud number represented by the current point cloud data collected by the vehicle-mounted millimeter-wave radar and the average point cloud number represented by multiple first historical point cloud data to obtain the point cloud density score of the vehicle-mounted millimeter-wave radar.

[0057] In an alternative embodiment, the integrity of target detection can be evaluated by statistically calculating the ratio of the target point cloud number detected in the current frame collected by the vehicle-mounted millimeter-wave radar (that is, the target point cloud number represented by the current point cloud data collected by the vehicle-mounted millimeter-wave radar) to the historical average point cloud number of the sliding window (that is, the average point cloud number represented by the first historical point cloud data). A low density may be due to occlusion or a decrease in detection ability. The point cloud density score of the vehicle-mounted millimeter-wave radar is specifically shown in the following formula: ; wherein, , if a sudden increase in point cloud is detected, such as a sudden increase in point cloud caused by multipath interference, an anomaly detection should be triggered.

[0058] Step S42: Obtain the velocity consistency score of the vehicle-mounted millimeter-wave radar according to the Doppler velocity variances of the current point cloud data collected by the vehicle-mounted millimeter-wave radar and the multiple first historical point cloud data for the same target.

[0059] In this embodiment, the velocity consistency score of the vehicle-mounted millimeter-wave radar can be calculated according to the Doppler velocity variances of the current point cloud data collected by the vehicle-mounted millimeter-wave radar and the multiple first historical point cloud data for the same target.

[0060] In an alternative embodiment, the rationality of the data is verified by combining Doppler velocity measurement and kinematic laws, and the velocity consistency score of the vehicle-mounted millimeter-wave radar is calculated , specifically as shown in the following formula: ; where , is the Doppler velocity variance, obtained by statistically analyzing the variance of the Doppler velocities of the same target within the current frame; β is an empirical value, which can be taken as 0.1 here to support sharp weight reduction under high variances.

[0061] Step S43: Obtain the millimeter-wave radar signal intensity, and obtain the signal-to-noise ratio score of the vehicle-mounted millimeter-wave radar based on the magnitude relationship between the millimeter-wave radar signal intensity and the millimeter-wave radar signal intensity threshold.

[0062] In this embodiment, the millimeter-wave radar signal intensity of the vehicle-mounted millimeter-wave radar can also be obtained to obtain the signal-to-noise ratio score of the vehicle-mounted millimeter-wave radar based on the magnitude relationship between the millimeter-wave radar signal intensity and the preset millimeter-wave radar signal intensity threshold.

[0063] In an alternative implementation, the credibility can be dynamically adjusted based on the radar signal intensity. Specifically, the signal-to-noise ratio score of the vehicle-mounted millimeter-wave radar can be calculated by the following formula: ; where ; is the millimeter-wave radar signal intensity, i.e., the preset millimeter-wave radar signal intensity threshold, which can be set to 15 dB in typical cases. Specifically, when reaches or exceeds 30 dB (i.e., the strong signal state), the value is taken as 1; while when is lower than 20 dB (i.e., the weak signal state), the value is less than 0.5.

[0064] Step S44: Obtain the multipath suppression score of the vehicle-mounted millimeter-wave radar according to the proportion of invalid point clouds in the currently acquired point cloud data of the vehicle-mounted millimeter-wave radar.

[0065] In this embodiment, the invalid point clouds in the currently acquired point cloud data of the vehicle-mounted millimeter-wave radar can be determined, and the proportion of the invalid point clouds in the currently acquired point cloud data of the vehicle-mounted millimeter-wave radar can be determined, so as to obtain the multipath suppression score of the vehicle-mounted millimeter-wave radar according to the proportion of the invalid point clouds in the currently acquired point cloud data of the vehicle-mounted millimeter-wave radar.

[0066] In an alternative embodiment, false echoes can be identified through spatial topology analysis. The algorithm strategy is as follows: after constructing the three-dimensional distribution map of the point cloud of the current frame, if it is observed that the point cloud distribution in a certain area shows mirror symmetry with the surrounding environment (for example, due to the reflection of the tunnel wall), then this area is marked as a multipath interference area, the point cloud in this area is determined as invalid point cloud, and the weight of the corresponding invalid point cloud is suppressed accordingly. Specifically, the multipath suppression score of the vehicle-mounted millimeter-wave radar can be calculated by the following formula: ; where is the proportion of invalid point clouds, and the calculation formula is as follows: ; where is the total number of point clouds detected in the current frame, is the number of "invalid point clouds generated by multipath interference"; among them, this embodiment does not limit the specific method for determining "invalid point clouds generated by multipath interference", and any method for determining "invalid point clouds generated by multipath interference" can implement the content of this embodiment.

[0067] The above multipath suppression score The calculation formula of describes the inverse relationship between the interference intensity and the credibility in an exponential form. The coefficient 3 is an empirical value used to adjust the attenuation rate and control the sensitivity of the credibility when the invalid point clouds increase. Through the exponential decay mechanism, while ensuring physical consistency, the value is limited according to the engineering safety requirements as: , realizing the quantifiable suppression of multipath interference.

[0068] Step S45: Obtain the credibility score of the currently acquired point cloud data of the vehicle-mounted millimeter-wave radar based on the point cloud density score, speed consistency score, signal-to-noise ratio score, and multipath suppression score of the vehicle-mounted millimeter-wave radar.

[0069] In this embodiment, the credibility of the current point cloud data collected by the vehicle-mounted millimeter-wave radar can be evaluated based on the point cloud density score of the vehicle-mounted millimeter-wave radar, the speed consistency score of the vehicle-mounted millimeter-wave radar, the signal-to-noise ratio score of the vehicle-mounted millimeter-wave radar, and the multipath suppression score of the vehicle-mounted millimeter-wave radar, and the credibility score of the current point cloud data collected by the vehicle-mounted millimeter-wave radar can be determined.

[0070] In an alternative embodiment, the credibility score of the current point cloud data collected by the vehicle-mounted millimeter-wave radar can be obtained by the following formula , as shown in the following formula: ; Wherein, is the point cloud density score of the vehicle-mounted millimeter-wave radar, is the speed consistency score of the vehicle-mounted millimeter-wave radar, is the signal-to-noise ratio score of the vehicle-mounted millimeter-wave radar, and is the multipath suppression score of the vehicle-mounted millimeter-wave radar.

[0071] It should be noted that this embodiment does not limit the execution order between the above steps S41 to S44. For example, the execution order can be arbitrarily specified, or they can be executed simultaneously.

[0072] Combined with the above embodiments, in one implementation, the embodiments of the present invention further provide a data processing method. In this method, the multiple vehicle-mounted sensor data at least includes the current point cloud data collected by the vehicle-mounted lidar; and, the "respectively evaluating the credibility of the multiple vehicle-mounted sensor data to obtain the credibility scores of the multiple vehicle-mounted sensor data" in the above step S12 at least includes steps S51 to S54: Step S51: Obtain the reflectivity score of the vehicle-mounted lidar according to the target reflectivity and the reflectivity threshold of the current point cloud data collected by the vehicle-mounted lidar.

[0073] Considering that the accuracy of the vehicle-mounted lidar is significantly affected by weather and target reflectivity, based on this, in this embodiment, when evaluating the reliability of the vehicle-mounted lidar, the dynamic environment and target characteristics need to be comprehensively considered. In this embodiment, the reflectivity score of the vehicle-mounted lidar can be determined according to the target reflectivity of the current point cloud data collected by the vehicle-mounted lidar and the preset reflectivity threshold.

[0074] In an alternative implementation, the reflectivity score of the vehicle-mounted lidar can be obtained by the following formula , to correct the problem of missed detection of low-reflectivity targets (such as black vehicles): ; Wherein, , the normalized reflectivity threshold in the formula is the preset reflectivity threshold: for example, the average reflectivity of a white vehicle body is set to 0.8; for example, when a target with a detected target reflectivity < 0.2 (dark object) is detected, = 0.5, and compensate for missed detection targets by associating camera data.

[0075] Step S52: Obtain the visibility and rainfall intensity of the driving environment around the vehicle to obtain the environmental attenuation score of the vehicle-mounted lidar.

[0076] In this embodiment, the visibility and rainfall intensity of the driving environment around the vehicle can be determined, and based on the visibility and rainfall intensity of the driving environment around the vehicle, the environmental attenuation score of the vehicle-mounted lidar can be determined. In an alternative embodiment, the environmental attenuation score of the vehicle-mounted lidar can be determined by the following formula : ; where ; the visibility unit is m, which can be obtained by the inversion algorithm of the lidar controller or through other means, or the real-time visibility can be obtained through other means; the rainfall intensity unit is mm / h, which can be obtained by a dedicated sensor or through other means, k is an empirical value coefficient, which is preset, and in a typical case, k = 0.05 can be taken. The smaller the k value, the smoother the attenuation response. In practical applications, it can be calibrated according to the hardware performance differences of different lidars.

[0077] Step S53: Obtain the trajectory continuity score of the vehicle-mounted lidar based on the target position detected in the current point cloud data collected by the vehicle-mounted lidar for the same target and the target position predicted based on the historical trajectory of the target in multiple second historical point cloud data.

[0078] In this embodiment, the credibility of the target trajectory can be verified through the spatio-temporal continuity of the lidar to avoid false detection caused by sudden noise or dynamic obstacle interference. Based on this, the target position detected in the current point cloud data collected by the vehicle-mounted lidar for the same target can be determined, and, based on the historical trajectory of the same target in multiple second historical point cloud data, the target position can be predicted. Thus, based on the target position detected in the current point cloud data collected by the vehicle-mounted lidar for the same target and the target position predicted based on the historical trajectory of the target in multiple second historical point cloud data, the trajectory continuity score of the vehicle-mounted lidar can be determined.

[0079] In an alternative embodiment, the trajectory continuity score of the vehicle-mounted lidar can be determined by the following formula : ; where ; The predicted position based on the historical trajectory (dynamically modeled using the Extended Kalman Filter, considering acceleration and yaw rate), i.e., the target position predicted according to the historical trajectory of the target in multiple second historical point cloud data; The target position detected in the current frame, i.e., the target position detected in the current point cloud data collected by the vehicle-mounted lidar for the same target; The predicted speed based on the historical trajectory; The target speed detected in the current frame. = 0.8m: The position tolerance threshold (e.g., in an urban road scenario, the allowed positioning error); = 2m / s: The speed tolerance threshold (set based on the target type: vehicle / pedestrian).

[0080] In an alternative embodiment, an anomaly detection mechanism can be based on and to set up. When the prediction (i.e., ) and the detection (i.e., ) have a deviation: When this occurs, out-of-order correction is triggered: If the target is static (speed variance < 0.1m / s), increase the value to 1.5m, allowing slight positioning jitter; If the target is a turning vehicle (e.g., turn signal activated), relax the lateral threshold to 2 .

[0081] Step S54: Based on the reflectivity score, environmental attenuation score, and trajectory continuity score of the vehicle-mounted lidar, obtain the credibility score of the current point cloud data collected by the vehicle-mounted lidar.

[0082] In this embodiment, the credibility of the current point cloud data collected by the vehicle-mounted lidar can be evaluated based on the reflectivity score of the vehicle-mounted lidar, the environmental attenuation score of the vehicle-mounted lidar, and the trajectory continuity score of the vehicle-mounted lidar, and the credibility score of the current point cloud data collected by the vehicle-mounted lidar can be determined.

[0083] In an alternative embodiment, the credibility score of the current point cloud data collected by the vehicle-mounted lidar can be obtained by the following formula , as shown in the following formula: ; Wherein, is the reflectivity score of the vehicle-mounted lidar, is the environmental attenuation score of the vehicle-mounted lidar, and is the trajectory continuity score of the vehicle-mounted lidar.

[0084] It should be noted that the execution order between the above steps S51 to S53 in this embodiment is not limited. For example, the execution order can be arbitrarily specified, or they can be executed simultaneously.

[0085] Combined with the above embodiments, in one implementation manner, the embodiment of the present invention further provides a data processing method. In this method, after the above step S11, it may further include steps S61 to S62, and the above step S13 may specifically include step S63: Step S61: Detect whether there is a conflict between the detection results represented by pairs of the current vehicle networking data and the multiple vehicle-mounted sensor data.

[0086] In this embodiment, when obtaining the current vehicle networking data of the vehicle itself and the multiple vehicle-mounted sensor data of the vehicle itself, it is possible to detect whether there is a conflict between the detection results represented by pairs of the current vehicle networking data and the multiple vehicle-mounted sensor data. For example, detect whether there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data (such as the image data collected by a vehicle-mounted camera), detect whether there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the second vehicle-mounted sensor data (such as the current point cloud data collected by a vehicle-mounted millimeter-wave radar), and detect whether there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the third vehicle-mounted sensor data (such as the current point cloud data collected by a vehicle-mounted lidar).

[0087] Step S62: In the case where there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data, determine whether to accept the current vehicle networking data according to the vehicle networking data within the first time period and the target vehicle-mounted sensor data within the first time period.

[0088] In this embodiment, in the case where it is determined that there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data, it is possible to trigger the local target vehicle-mounted sensor to perform local verification of the conflict data, obtain the vehicle networking data within the first time period and the target vehicle-mounted sensor data within the first time period, and determine whether to accept the current vehicle networking data according to the vehicle networking data within the first time period and the target vehicle-mounted sensor data within the first time period. In an optional example, it may be determined whether there is a conflict based on whether the difference between the detection results represented by pairs of the current vehicle networking data and the multiple vehicle-mounted sensor data exceeds a difference threshold. The difference threshold can be freely set according to requirements, such as 15%. And the difference threshold between the current vehicle networking data and the detection results represented by pairs of the multiple vehicle-mounted sensor data can be the same threshold, or different thresholds according to different data types, and there is no limitation on this.

[0089] Among them, the first vehicle-mounted sensor data is any one of the multiple vehicle-mounted sensor data of the vehicle itself. The target vehicle-mounted sensor data is the detection result of the target vehicle-mounted sensor of the vehicle itself for the driving environment around the vehicle. The target vehicle-mounted sensor is a high-confidence sensor of the vehicle itself. This high-confidence sensor can be a preset sensor, such as a Frequency Modulated Continuous Wave Radar (FMCW), etc., or it can be a vehicle-mounted sensor with the highest confidence that exceeds the confidence threshold determined according to the real-time confidence of multiple vehicle-mounted sensors. This embodiment does not limit this. The first time period can be a preset time period after it is determined that there is a conflict, such as within 10 ms after it is determined that there is a conflict, and this is not limited.

[0090] For example, in an optional specific example, it can be when it is determined that the current weather is bad weather conditions such as rain and fog. In order to reduce false alarms and improve the reliability of the vehicle safety system, the difference threshold corresponding to the detection results represented pairwise in the current vehicle networking data and multiple vehicle-mounted sensor data is increased. For example, the difference threshold between a local vehicle-mounted sensor (such as a camera, etc.) and the vehicle networking data can be relaxed from the standard 15% to 20%. When it is determined that there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data, activate the high-confidence sensor (such as a Frequency Modulated Continuous Wave Radar (FMCW)) to perform a directional scan on the conflict target area, and the scan range is limited within the horizontal field of view angle of ±30°. And adopt a pre-loaded fault injection detection algorithm to exclude sensor self-abnormalities (such as radar mirror reflection interference). Then, within the next 10 ms, collect 3 frames of radar data (sampling rate of 50 Hz) of the Frequency Modulated Continuous Wave Radar and vehicle networking data, and calculate the target real position through sliding window filtering to determine whether to adopt the current vehicle networking data. For example, the following verification logic can be used to determine whether to adopt the current vehicle networking data (V2X data): ; Among them, is the probability that the detection result represented by the vehicle networking data within the first time period is true, is the probability that the detection result represented by the target vehicle-mounted sensor data (such as the radar data collected by the Frequency Modulated Continuous Wave Radar) within the first time period is true. The purpose of adding 1 to the denominator is to avoid division by zero, and the result takes the normalized error.

[0091] In addition, a timeout fuse mechanism is also set in this embodiment: If the verification is not completed within a certain time (such as 20 ms), forcefully adopt the current vehicle networking data and execute according to the current vehicle networking data (safety first principle).

[0092] Step S63: In the case of adopting the current vehicle networking data, using the credibility scores of the current vehicle networking data and the remaining multiple in-vehicle sensor data other than the first in-vehicle sensor data as fusion weights, fuse the current vehicle networking data and the remaining multiple in-vehicle sensor data to obtain the fused data.

[0093] In this embodiment, in the case of determining to adopt the current vehicle networking data, the first in-vehicle sensor data and its corresponding credibility score are removed. Based on the credibility scores of the current vehicle networking data and the remaining multiple in-vehicle sensor data other than the first in-vehicle sensor data, respective fusion weights are assigned to the current vehicle networking data and the remaining multiple in-vehicle sensor data other than the first in-vehicle sensor data. Thus, based on the respective fusion weights of the current vehicle networking data and the remaining multiple in-vehicle sensor data other than the first in-vehicle sensor data, the current vehicle networking data and the remaining multiple in-vehicle sensor data are fused to obtain the fused data, and the fused data represents the final detection result of the driving environment around the vehicle.

[0094] In this embodiment, in the case of data conflict, first, a local target in-vehicle sensor is used for quick review to achieve primary arbitration, thereby improving vehicle safety.

[0095] Combined with the above embodiments, in one implementation, the embodiment of the present invention further provides a data processing method. In this method, in addition to the above steps, it may further include steps S71 to S74: Step S71: In the case of the failure of the target in-vehicle sensor, or, in the case of the vehicle triggering a collaborative perception request, determine multiple neighboring vehicles based on the communication quality, relative positions of the vehicle and multiple surrounding vehicles, and the respective credit scores of the multiple surrounding vehicles.

[0096] In this embodiment, it is possible to verify whether the target in-vehicle sensor fails, and determine whether the vehicle triggers a collaborative perception request, and the collaborative perception request is used to request the V2X perception data of surrounding vehicles. In an optional example, the protocol design of the collaborative perception request is as follows: Message format: Expand the BSM message field and add a ConflictResolutionRequest part: message ConflictRequest { uint32 conflict_id; / / Conflict unique identifier double timestamp; / / UTC timestamp (microsecond level) GeoPosition target_geo; / / Conflict target WGS84 coordinates ConflictType type; / / Conflict types A / B / C repeated SensorType required_sensors; / / Requested sensor types (radar / camera) } In this embodiment, when it is determined that the target vehicle-mounted sensor fails, or when it is determined that the vehicle triggers a collaborative perception request, multiple neighboring vehicles can be determined based on the communication quality, relative position between the vehicle and multiple surrounding vehicles, and the respective credit scores of the multiple surrounding vehicles. The neighboring vehicle represents the target vehicle that returns a collaborative perception response to the vehicle. Among them, the communication quality is characterized by SINR (Signal-to-Interference-plus-Noise Ratio), which reflects the reliability of the vehicle networking data transmission link. The credit score can be a credit score determined based on the blockchain historical record, and the relative position is the straight-line distance between the vehicle and the surrounding vehicles.

[0097] In an optional example, the neighboring vehicle screening conditions can be set in advance: vehicles with a communication quality greater than the first threshold (such as 15 db), a relative position within the second threshold range (such as 100 m and a heading angle deviation less than 45°), and a credit score higher than the third threshold are determined as neighboring vehicles, and there is no limitation on this.

[0098] In addition, in a preferred implementation manner, multiple initial neighboring vehicles can be determined based on the communication quality, relative position between the vehicle and multiple surrounding vehicles, and the respective credit scores of the multiple surrounding vehicles, and then a greedy algorithm is used to screen out multiple neighboring vehicles (i.e., the optimal neighboring vehicles) from the multiple initial neighboring vehicles to ensure their uniform spatial distribution and thus avoid data redundancy.

[0099] Step S72: Send a collaborative perception request to the multiple neighboring vehicles and receive the collaborative perception responses returned by the multiple neighboring vehicles.

[0100] In this embodiment, a collaborative perception request is sent to the multiple neighboring vehicles, and the collaborative perception responses returned by the multiple neighboring vehicles based on the collaborative perception request are received. Among them, the collaborative perception response returned by a neighboring vehicle represents the detection result of the driving environment around the vehicle by the neighboring vehicle. In an optional example, after receiving the request, the neighboring vehicle can process and return the data according to the priority of the conflict type (response period ≤ 30 ms).

[0101] In one embodiment, considering the possible conflict situations, the priorities of the conflict types are divided into three categories from high to low: Type A conflict (emergency safety event): Involving emergency braking, collision warning, etc. (such as the deceleration reported by V2X > 6 m / s², while the corresponding target is not detected by the camera); corresponding response requirement: arbitration is completed within ≤ 50 ms.

[0102] Type B conflict (status inconsistency): such as traffic light status, lane line recognition differences (the camera shows a red light, while V2X reports a green light); corresponding response requirement: arbitration shall be completed within ≤ 200 ms.

[0103] Type C conflict (position deviation): the position / speed error of the target vehicle exceeds the safety threshold (lateral error > 1 m, longitudinal error > 3 m); corresponding response requirement: arbitration shall be completed within ≤ 100 ms.

[0104] When multiple conflicts occur simultaneously, the arbitration processing priorities corresponding to each conflict can be determined, and arbitration decisions shall be made in descending order of the arbitration processing priorities. In an optional specific example, the method for determining the arbitration processing priorities is as follows: ; among them, the conflict type coefficients of Type A, Type B, and Type C conflicts are set in descending order (the highest is 1), and the scene danger level can be dynamically calculated based on the time to collision (TTC).

[0105] Step S73: Using the communication quality, relative position between the vehicle itself and each neighboring vehicle, and the credit score of each neighboring vehicle as the weight of the neighboring vehicle, perform weighted fusion on the collaborative perception responses returned by multiple neighboring vehicles to obtain the final collaborative perception response.

[0106] In this embodiment, based on the communication quality, relative position between the vehicle itself and each neighboring vehicle, and the credit score of each neighboring vehicle, the weight of the neighboring vehicle can be determined, and then based on the weight of each neighboring vehicle, perform weighted fusion on the collaborative perception responses returned by multiple neighboring vehicles to obtain the final collaborative perception response.

[0107] Step S74: Compare the final collaborative perception response with the current vehicle networking data to determine whether to accept the current vehicle networking data.

[0108] In this embodiment, after obtaining the final collaborative perception response, the final collaborative perception response can be compared with the current vehicle networking data to determine whether to accept the current vehicle networking data.

[0109] In a specific implementation manner, when the primary arbitration fails to resolve conflicts (such as radar failure due to weather) or the conflicts involve multi-target collaboration, secondary arbitration (requesting collaborative perception of V2V) is triggered. For example, broadcast a request for the perception data of the neighboring n vehicles through V2V communication, and use a weighted voting mechanism to fuse and verify these data. After receiving the collaborative perception responses returned by multiple neighboring vehicles, the clock deviation between the neighboring vehicle and the vehicle itself can be compensated through NTP / PTP protocol synchronization, and then the coordinate system of the neighboring vehicle can be converted to the coordinate system of the vehicle itself to achieve spatio-temporal alignment. Then, a multi-source data fusion of target existence / target position is performed through a weighted voting mechanism: ; Wherein, is the credibility weight of an adjacent vehicle, which represents the credibility of the collaborative perception response of the adjacent vehicle. The adjacent vehicle credibility weight = 0.3 * adjacent vehicle credit score + 0.7 * adjacent vehicle sensor quality score, is the probability that the collaborative perception response returned by the adjacent vehicle is true, is the probability that the final collaborative perception response is true.

[0110] In this embodiment, when the first-level arbitration cannot resolve conflicts or the conflicts involve multi-object collaboration, the second-level arbitration is triggered, that is, the collaborative perception of vehicle networking data is triggered to further perform security arbitration on multi-source heterogeneous data, solving the problems of low efficiency and insufficient security of traditional single-level conflict handling.

[0111] Combined with the above embodiments, in one implementation manner, the embodiment of the present invention further provides a data processing method. In this method, in addition to the above steps, it may further include steps S81 to S83: Step S81: When the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data belongs to the first target conflict type, obtain the actual driving environment around the vehicle through the traffic management system.

[0112] In this embodiment, when there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data, it is also necessary to determine the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data, and determine whether the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data belongs to the first target conflict type. When it is determined to belong to the first target conflict type, the actual driving environment around the vehicle can be obtained by sending a request to the traffic management system.

[0113] Exemplarily, the first target conflict type may be a traffic light status conflict type. When it is determined to belong to the traffic light conflict type, the real-time traffic management system can be accessed (such as by the edge node calling the traffic signal API) to obtain the actual driving environment around the vehicle, such as obtaining the real signal light status.

[0114] Step S82: When the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data belongs to the second target conflict type, retrieve and analyze the multi-vehicle trajectory data during the conflict period through the cloud server to obtain the actual driving environment around the vehicle.

[0115] In this embodiment, when there is a conflict between the detection result of the current vehicle networking data representation and the detection result of the first vehicle-mounted sensor data representation, it is also necessary to determine the type of conflict between the detection result of the current vehicle networking data representation and the detection result of the first vehicle-mounted sensor data representation, so as to determine whether the type of conflict between the detection result of the current vehicle networking data representation and the detection result of the first vehicle-mounted sensor data representation belongs to the second target conflict type, where the first target conflict type is different from the second target conflict type. When it is determined that the type of conflict between the detection result of the current vehicle networking data representation and the detection result of the first vehicle-mounted sensor data representation belongs to the second target conflict type, the multi-vehicle trajectory data during the conflict time period can be retrieved and analyzed through the cloud server to obtain the actual driving environment around the vehicle.

[0116] For example, request the cloud server to retrieve the multi-vehicle trajectory data during the conflict time period (such as within 5 seconds before and after the conflict). The range of the multi-vehicle trajectory data can be all the trajectory data of the vehicles within a certain value (such as within 500m) with the conflict as the center, so as to determine the motion consistency of the target and obtain the actual driving environment around the vehicle.

[0117] Step S83: Based on the actual driving environment around the vehicle and the current vehicle networking data, determine whether to adopt the current vehicle networking data.

[0118] In this embodiment, after obtaining the actual driving environment around the vehicle, it is possible to determine whether to adopt the current vehicle networking data based on the actual driving environment around the vehicle and the current vehicle networking data.

[0119] In an example, when there is still a conflict after secondary arbitration, or it is a type B conflict (status inconsistency, such as traffic light status contradiction), at this time, authoritative data verification is required, and tertiary arbitration can be triggered, that is, cloud multi-source verification and trajectory backtracking (such as fusing traffic management data and calling a high-precision map for trajectory backtracking). Among them, when it is determined that the vehicle trajectories of more than the first threshold (such as 80%) point to the same conclusion (such as the detection result of the current vehicle networking data), it can be determined to adopt the current vehicle networking data. In addition, in a preferred implementation manner, historical data matching can also be performed: retrieve historical decisions in a similar scenario (such as the same intersection at the same time period) in the past period of time (such as 30 days), and recommend corresponding solutions when the similarity of the historical decisions exceeds 85%.

[0120] In this embodiment, for the abnormal conflict situation of multi-modal data, a three-level arbitration mechanism based on local sensor quick review -> V2V collaborative perception request -> cloud multi-source verification and trajectory backtracking is proposed. Through this three-level conflict arbitration mechanism, the problems of low efficiency and insufficient security of traditional single-level conflict handling are solved, and the theoretical verification conflict resolution delay is reduced.

[0121] In combination with the above embodiments, in one implementation, the embodiments of the present invention further provide a data processing method. In this method, the credibility evaluation of the current vehicle networking data and the credibility evaluation of various in-vehicle sensor data are respectively implemented through the credibility evaluation model of the vehicle itself. In addition to the above steps, steps S91 to S92 may also be included, and the specific content of "performing the credibility evaluation on the current vehicle networking data and the credibility evaluation on various in-vehicle sensor data respectively" in step S12 above includes step S93: Step S91: Obtain the target geographical grid where the vehicle is currently located.

[0122] In this embodiment, the target geographical grid where the vehicle is currently located can be obtained. Among them, a geographical area includes multiple geographical grids of a target size, and the target geographical grid is the geographical grid where the vehicle is currently located. For example, a city is finely divided into geographical grids (units) of 500 meters × 500 meters, and each geographical grid (unit) independently manages its corresponding credibility evaluation model parameters.

[0123] Step S92: Load the target credibility evaluation model parameters corresponding to the target geographical grid.

[0124] In this embodiment, after determining the target geographical grid, the target credibility evaluation model parameters corresponding to the target geographical grid can be loaded to update the model parameters of the credibility evaluation model, and a credibility evaluation model with the target credibility evaluation model parameters is obtained. Among them, the credibility evaluation model parameters corresponding to each geographical grid can be deployed in the vehicle controller to achieve the loading of the target credibility evaluation model parameters; it can also be that the cloud system deploys the credibility evaluation model parameters corresponding to each geographical grid, and the cloud system can determine the target credibility evaluation model parameters corresponding to the target geographical grid according to the real-time position of the vehicle and send them to the vehicle end, so as to achieve the vehicle end loading the target credibility evaluation model parameters corresponding to the target geographical grid.

[0125] Step S93: Use the credibility evaluation model with the target credibility evaluation model parameters to perform the credibility evaluation on the current vehicle networking data and the credibility evaluation on various in-vehicle sensor data respectively.

[0126] In this embodiment, the credibility evaluation model with the target credibility evaluation model parameters can be used to perform the credibility evaluation on the current vehicle networking data to obtain the credibility score of the current vehicle networking data, and, through the credibility evaluation model with the target credibility evaluation model parameters, the credibility evaluation on various in-vehicle sensor data is respectively performed to obtain the credibility scores of various in-vehicle sensor data.

[0127] In this embodiment, through the embedding mechanism of spatial features (i.e., the update of the geographical rasterization model), the dynamic optimization of the model can be realized to adapt to different scenarios, enhancing the generalization ability of the model. Compared with traditional centralized training, the regional adaptation accuracy is improved.

[0128] Combined with the above embodiments, in one implementation, the embodiment of the present invention also provides a data processing method. The credibility evaluation of the current vehicle networking data is realized through the vehicle networking credibility evaluation model in the credibility evaluation model of the vehicle itself; in addition to the above steps, steps S101 and S102 may further be included: Step S101: When the current moment belongs to the first target time period, the weight corresponding to the communication quality item in the vehicle networking credibility evaluation model is increased from the first weight to the second weight.

[0129] In this embodiment, the vehicle networking credibility evaluation model is used to evaluate the credibility of the current vehicle networking data to obtain the credibility score of the current vehicle networking data. Among them, the vehicle networking credibility evaluation model at least includes: the model parameters of the communication quality item. This embodiment can determine the time period to which the current moment belongs. When it is determined that the current moment belongs to the first target time period, the weight corresponding to the communication quality item in the vehicle networking credibility evaluation model can be increased from the first weight to the second weight. For example, the first target time period may be the time period corresponding to the morning and evening traffic peak periods. At this time, a higher weight can be given to the weight corresponding to the communication quality item in the vehicle networking credibility evaluation model (such as the weight α + 0.2 corresponding to the communication quality item). Among them, the first weight may be the preset weight corresponding to the communication quality item.

[0130] Step S102: When the current moment belongs to the second target time period, the weight corresponding to the communication quality item in the vehicle networking credibility evaluation model is restored to the first weight.

[0131] In this embodiment, the second target time period may be a non-first target time period. When it is determined that the current moment belongs to the second target time period, the weight corresponding to the communication quality item in the vehicle networking credibility evaluation model can be restored to the first weight.

[0132] In addition, in another embodiment, when it is determined that the current moment belongs to the third target time period representing night, after determining the respective fusion weights of various vehicle-mounted sensor data, the respective fusion weights of various vehicle-mounted sensor data are increased to obtain the updated respective fusion weights of various vehicle-mounted sensor data. Thus, according to the fusion weight of the current vehicle networking data and the updated respective fusion weights of various vehicle-mounted sensor data, the current vehicle networking data and various vehicle-mounted sensor data are weighted and fused to obtain the fused data.

[0133] In this embodiment, through the time-sliding window mechanism, dynamic optimization of the model can be achieved to adapt to different scenarios, enhancing the generalization ability of the model. Compared with traditional centralized training, the regional adaptation accuracy is improved.

[0134] Combining the above embodiments, in one implementation, the embodiment of the present invention further provides a data processing method. In this method, step S13 above may specifically include step S111 and step S112: Step S111: When the current vehicle networking data and the multiple in-vehicle sensor data are discrete, using the respective credibility scores of the current vehicle networking data and the multiple in-vehicle sensor data as fusion weights, the current vehicle networking data and the multiple in-vehicle sensor data are fused through a Bayesian network to obtain the fused data.

[0135] In this embodiment, when the current vehicle networking data and the multiple in-vehicle sensor data are discrete, for example, when it is determined that at least one type of data among the current vehicle networking data and the multiple in-vehicle sensor data is discrete, data fusion processing is performed through a Bayesian network, that is, the fusion weights are determined according to the respective credibility scores of the current vehicle networking data and the multiple in-vehicle sensor data, and the current vehicle networking data and the multiple in-vehicle sensor data are fused through a Bayesian network based on the fusion weights to obtain the fused data.

[0136] In this embodiment, for semantic information such as traffic light status recognition and road topology, the Bayesian network is used to model based on probability relationships to solve discrete data conflicts; in this way, not only can non-numerical semantic information be parsed, such as "red light" and "green light" in traffic lights, reflecting symbolic reasoning ability, and the probability form of the output result of the Bayesian network can be directly integrated with the autonomous driving decision-making system to achieve seamless docking; but also it can flexibly integrate V2X rules (such as traffic light phase information from roadside unit RSU) and local sensor data to achieve the scalability of dynamic conditions.

[0137] Step S112: When the current vehicle networking data and the multiple in-vehicle sensor data are continuous, using the respective credibility scores of the current vehicle networking data and the multiple in-vehicle sensor data as fusion weights, the current vehicle networking data and the multiple in-vehicle sensor data are fused through weighted Kalman filtering to obtain the fused data.

[0138] In this embodiment, when the current vehicle networking data and various in-vehicle sensor data are continuous, for example, when it is determined that at least one type of data among the current vehicle networking data and various in-vehicle sensor data is continuous, data fusion processing is performed through an improved weighted Kalman filter (WKF). That is, the fusion weights are determined according to the respective credibility scores of the current vehicle networking data and various in-vehicle sensor data, and the current vehicle networking data and various in-vehicle sensor data are fused through weighted Kalman filtering based on the fusion weights to obtain the fused data.

[0139] Among them, considering that the traditional Kalman filter assumes that all input data has a constant credibility and cannot adapt to the dynamic weight scenarios of multi-source data such as V2X, this embodiment selects an improved weighted Kalman filter (WKF) to fuse the data. It can not only dynamically adjust the weights according to the real-time credibility of data such as V2X and sensors to improve the robustness of state estimation, but also achieve the goal of maintaining high-efficiency calculation without bringing an additional burden to the system computing power by directly correcting the covariance matrix, thereby improving the calculation efficiency (fusion efficiency). In addition, through weighted Kalman filtering, it can also be compatible with traditional autonomous driving perception algorithms, facilitating integration with existing systems and realizing the compatibility of engineering projects.

[0140] In this embodiment, an improved weighted Kalman filter is used to process continuous spatio-temporal state estimations such as vehicle trajectories and speeds, which is suitable for high-frequency data processing scenarios. At the same time, it uses a Bayesian network to efficiently process discrete semantic information such as traffic rules and events. Although the update frequency is low, it plays a crucial role in decision-making. This embodiment forms a "numerical-semantic" dual verification level through WKF and the Bayesian network to solve the problem of single data source dependence in traditional methods.

[0141] In one embodiment, as Figure 2 shown, Figure 2 is the overall system architecture diagram of a multi-modal data fusion system based on dynamic credibility evaluation shown in an embodiment of the present invention. In Figure 2 it, the overall system architecture of this system consists of four layers: Multi-source heterogeneous data input layer: As the basic data entry of the system, it is responsible for accessing, preprocessing, and standardizing multi-source heterogeneous data. The multi-source heterogeneous data includes: (1) the position, speed, and heading angle of surrounding vehicles carried in V2X communication data (such as BSM / CAM / DENM, etc.); (2) the data of in-vehicle sensors (the detection results of cameras, millimeter-wave radars, and lidar). Then, protocol parsing is performed on the V2X communication data, and the data of in-vehicle sensors is standardized and then aligned in time and space (including time synchronization and coordinate transformation, and the error of coordinate transformation < 5 cm) to obtain preprocessed data, which is transmitted to the dynamic credibility evaluation layer.

[0142] Dynamic credibility evaluation layer: Independently calculates real-time credibility scores for each type of data (such as V2X messages, camera detection frames, radar point clouds, etc.), which serves as the basis for weight allocation.

[0143] Among them, the V2X credibility score can be calculated for the V2X messages in the preprocessed data based on multiple aspects such as SINR / aging / history through the V2X credibility model in the credibility evaluation model; the camera credibility score can be calculated for the image data collected by in-vehicle cameras in the preprocessed data based on multiple aspects such as SSIM / clarity / lighting through the camera model (i.e., the in-vehicle camera credibility model) in the credibility evaluation model; the millimeter-wave radar credibility score can be calculated for the current point cloud data collected by the millimeter-wave radar in the preprocessed data based on multiple aspects such as point cloud density / speed consistency / signal-to-noise ratio / multipath suppression through the millimeter-wave radar model (i.e., the millimeter-wave radar credibility model) in the credibility evaluation model; the lidar credibility score can be calculated for the current point cloud data collected by the lidar in the preprocessed data based on multiple aspects such as reflectivity / environmental attenuation / trajectory continuity through the lidar model (i.e., the lidar credibility model) in the credibility evaluation model. Finally, a real-time credibility matrix is determined based on the V2X credibility score, camera credibility score, millimeter-wave radar credibility score, and lidar credibility score, and then the real-time credibility matrix is transmitted to the fusion decision layer. In this way, through the dynamic credibility evaluation layer, multi-factor real-time quantitative evaluation of multi-modal perception data is achieved, reducing errors compared to traditional static fixed weights or preset weight adjustment rules.

[0144] Fusion decision layer: An improved weighted Kalman filter (i.e., improved WKF) + Bayesian network fusion framework is adopted. Trajectory fusion is performed based on dynamic weights through the improved WKF, and semantic reasoning is performed based on the conditional probability table through the Bayesian network, so as to dynamically integrate multi-modal data, and perform confidence threshold judgment to determine whether arbitration triggering is involved. Anomaly data arbitration is performed for the three-level arbitration mechanism (i.e., three-level conflict handling: local -> V2V -> cloud) designed for multi-modal perception abnormal conflicts, and then a 360° environmental perception result is output.

[0145] Model global optimization layer: Continuously optimize the credibility evaluation model based on federated learning, and enhance the scene adaptability in the credibility evaluation model: Each federated client sends local gradients to the edge server, the edge server realizes edge aggregation based on multiple local gradients and sends them to the cloud server, the cloud server performs gradient update based on the processing results of the edge server's secure aggregation, and issues an incremental update package to the federated client, so as to realize the update of the credibility evaluation model in the federated client.

[0146] Thus, the multi-modal data fusion system based on dynamic credibility evaluation in this embodiment provides a multi-modal data fusion method based on dynamic credibility evaluation to solve the problems of rigid multi-modal data fusion weights, poor dynamic environment adaptability, and low conflict arbitration efficiency in the existing V2X collaborative perception system. The core of this embodiment lies in establishing a dynamic credibility quantification evaluation model for multi-modal data, realizing multi-level data conflict arbitration and real-time fusion decision-making, and optimizing and constructing an adaptive credibility model optimization system based on federated learning to solve the above problems existing in the prior art.

[0147] In one embodiment, as Figure 3 shown, Figure 3 is a flowchart of a three-level conflict arbitration mechanism shown in an embodiment of the present invention. In Figure 3 , when a data conflict is detected, the conflict type is determined to determine whether the conflict belongs to type A conflict, type B conflict or type C conflict. In the case where the conflict is determined to be a type A conflict, primary arbitration (quick review of local sensors) is performed: Activate the FMCW radar to collect three frames of data for verification within 10 ms, and determine whether the error is less than 15% based on the three frames of data. In the case where the error is determined to be less than 15%, local decision-making is determined to be executed; in the case where the error is determined to be not less than 15%, secondary arbitration is triggered to perform a V2X collaboration request: Among them, neighboring vehicles are first screened (such as selecting surrounding vehicles with SINR≥15 db), and then the support rate is determined through a weighted voting mechanism. In the case where the support rate is determined to be greater than 80%, the fusion weight is updated to perform new data fusion.

[0148] In the case where the conflict is determined to be a Class B conflict, secondary arbitration (V2V cooperative perception request) is performed: request the RSU signal status to determine whether the API response is successful; in the case where the response is determined to be successful, force the synchronization of the signal phase, and in the case where the response is determined to be unsuccessful, trigger tertiary arbitration and perform trajectory backtracking analysis: request multi-trajectory verification through the cloud to determine whether the historical matching degree of the multi-trajectories is greater than 85%, and in the case where the historical matching degree is greater than 85%, push the solution.

[0149] In the case where the conflict is determined to be a Class C conflict, tertiary arbitration (cloud multi-source verification and trajectory prediction comparison) is performed: request the cloud to retrieve the adjacent vehicle data, perform weighted least squares fitting on the trajectory data of the adjacent vehicle through Hampei filtering, determine the average error, and in the case where the average error < 2m, determine the existence of the target (such as an obstacle).

[0150] In one embodiment, as Figure 4 shown, Figure 4 is an optimized architecture diagram of a federated learning model shown in an embodiment of the present invention. In Figure 4 it, the architecture includes: local clients (such as vehicle / RSU terminals), edge aggregation points (such as MEC servers), and cloud global servers.

[0151] In the local client, a local dataset is stored, and the local dataset includes: local credibility evaluation data (such as historical V2X message conflict records, sensor false alarm rates). And, a lightweight sub-network is deployed, and the lightweight sub-network that can use local data to train the credibility model (such as only updating the λ and α coefficients, i.e., λ / α update). In the local client, during local training, Gaussian noise (σ = 0.01) is added to the gradient through differential privacy (DP) injection, satisfying the privacy budget of (ε = 1.0, δ = 1e-5), and Top-K gradient compression is adopted, that is, Top-K sparsification (retaining the top 10% of the maximum gradient values) + Huffman coding is adopted, and the compression rate reaches 90% to obtain the updated gradient (i.e., encrypted gradient) for gradient uploading to the edge aggregation node. (For example, the vehicle side is allowed to upload the gradient during the idle period (such as when charging the vehicle) to avoid network congestion and realize the asynchronous federated mechanism).

[0152] In the edge aggregation node, the model gradients uploaded by multiple local clients within the receiving area (such as a radius of 1 km) are received and cached (i.e., regional model caching), and preliminary gradient screening is performed (i.e., anomaly detection (3σ principle): detecting outliers based on the L2 norm and removing data deviating from the mean ±3σ), and gradient aggregation is performed through weighted averaging to obtain the aggregated gradient and update the regional model. The edge aggregation node synchronizes the information of the regional model update to the local clients covered by the edge aggregation node in real time, and at the same time uses a secure transmission protocol to transmit the aggregated gradient to the cloud global server.

[0153] In the cloud global server, aggregate gradients from multiple edge aggregation nodes are received. The global model in the global model library is updated based on the aggregate gradients through a secure aggregation protocol, and an incremental update package is sent to local clients (which can be pushed to local clients daily). Among them, spatio-temporal feature embedding can be performed in the cloud global server: Geographical gridification is carried out, and the city is finely divided into grid cells of 500 meters × 500 meters. Each cell independently manages the parameters of its sub-model, and the cloud global server can intelligently load the corresponding parameter configuration (i.e., scenario adaptation parameters) according to the real-time position of the vehicle and / or the scenario; and through the time sliding window mechanism: During the morning and evening traffic peaks, the model will assign a higher weight (α + 0.2) to communication quality, while at night, the weight of the sensor will be increased accordingly (β + 0.15), so as to achieve incremental update of the model (i.e., incremental update generation). In this way, through federated learning and spatio-temporal gridification, global optimization of model parameters is achieved, the generalization ability of the model is enhanced, and the regional adaptation accuracy is improved compared with traditional centralized training.

[0154] In this way, without leaking the original data of vehicles / roadside units, this embodiment continuously optimizes the credibility evaluation model based on distributed training, and enables the model to automatically adapt to local features for different regions (such as cities, highways, tunnels, etc.), environmental conditions (day and night, weather), and traffic densities, enhancing the generalization ability of the model. And the conflict case library will be updated: The solution to this conflict will be stored in the federated learning case library for incremental learning by other vehicles. And dynamic model adjustment is achieved: If a certain type of conflict (such as GPS drift in a tunnel) frequently occurs in a certain area, the edge node model will be fine-tuned (such as adjusting the failure factor).

[0155] In a specific example, multi-vehicle collaborative perception and traffic light state conflict arbitration at urban intersections are realized: The application scenario of this embodiment is: At an intersection of a main road in a certain city, the traffic flow is dense during the morning rush hour, and the following conflicts occur: The RSU sends a green light signal (confidence level 0.92) through V2X, the on-vehicle camera detects a red light (confidence level 0.85), and the lidar point cloud shows that the vehicle in front is braking suddenly (deceleration > 8m / s²).

[0156] The specific steps of this embodiment: 1) Data input and preprocessing: Parse the V2X message: Extract the signal light phase (G), SINR = 18dB, and timestamp alignment error < 1ms; Obtain sensor data: The SSIM of the camera image is 0.88, the light intensity is 120,000 Lux (backlight); The lidar detects 3 suddenly braking vehicles, and the trajectory continuity factor = 0.4; Perform spatio-temporal alignment: The world coordinate system conversion error < 3 cm.

[0157] 2) Dynamic credibility assessment: V2X credibility calculation: ; Camera credibility calculation: ; LiDAR credibility: The environmental attenuation factor = 0.6 (visibility 500 m), and trajectory continuity triggers anomaly detection.

[0158] 3) Fusion decision-making and conflict triggering: Through Bayesian network reasoning: Input evidence: In the field of wireless communication, the signal-to-interference-plus-noise ratio (SINR) is a key indicator for measuring communication quality. For example, when the SINR reaches 18 dB, it usually means high communication quality, which can lead to a vehicle-to-everything (V2X) communication accuracy rate as high as 94%. In addition, changes in light intensity will affect the false alarm rate of the camera. For example, under specific lighting conditions, the false alarm rate of the camera may reach 32%.

[0159] Calculate the posterior probability: P(true green light) = 0.94×(1 - 0.32)×0.7 / [0.94×0.68×0.7 + 0.06×0.32×0.3] = 0.91 Conflict determination: Class A conflict (hard braking event) and Class B conflict (signal light contradiction) are triggered simultaneously.

[0160] 4) Three-level arbitration execution: First-level arbitration: Activate the FMCW radar for directional scanning and confirm the actual existence of 2 hard braking vehicles within 3 frames.

[0161] Second-level arbitration: The V2V requests data from 3 adjacent vehicles (response time 28 ms), and 4 / 5 vehicles report the green light status.

[0162] Third-level arbitration: The cloud retrieves the traffic signal machine log and confirms that the RSU message lags by 2 seconds (the red light is actually on).

[0163] 5) Federated learning optimization: Upload conflict cases: Mark the scenario of "high-light V2X - camera time series inconsistency".

[0164] Model incremental update: Dynamically adjust the time decay factor λ in the signal light CPT from 0.3 → 0.5; the edge node sends a update packet, and the vehicle model update within the area takes < 15 seconds.

[0165] In another specific example, emergency obstacle detection in a highway foggy section is implemented: The application scenario of this embodiment is as follows: sudden occurrence of freeway group fog (visibility < 50m), with the following data conflicts: V2X message reports a stationary obstacle 100m ahead (confidence level 0.82), millimeter-wave radar detects multipath interference (point cloud density suddenly increases by 300%), and the camera cannot identify due to low light (SSIM = 0.45).

[0166] The specific steps of this embodiment are as follows: 1) Preliminary screening of data quality: V2X message verification: historical consistency H = 0.95 (error in the past 5 frames < 1m); Sensor anomaly marking: radar point cloud density score = 0.2 (compared with the sliding window mean), and the camera credibility is automatically reduced to 0.3; 2) Dynamic weight fusion: Improved WKF parameters: The state estimation covariance is inflated by 2 times (in the obstacle area); Trajectory prediction correction: the deviation between the predicted position by the extended Kalman filter and the V2X report is 0.8m, and the dynamic trust threshold is relaxed to 1.2m (severe weather strategy); 3) Multi-level conflict handling: First-level arbitration: the lidar penetrates the fog layer and detects a metal reflector (confidence level 0.78); Second-level arbitration: triggered by a V2V collaboration request, obtain the radar data of the rear vehicle (3 / 5 vehicles confirm the obstacle); Third-level arbitration: match the meteorological data in the cloud and activate the "group fog mode" parameters: the V2X weight is increased to 0.9, and the radar multipath suppression factor is forced to be set to 0.1; 4) Online optimization of the model: The edge node aggregates the group fog scenario gradient data from 10 vehicles; Federated learning update: in the lidar environmental attenuation formula, κ changes from 0.05 to 0.08, and the dynamic trust threshold module adds a visibility-speed correlation term.

[0167] The above 2 specific implementation cases are verified through typical scenarios, demonstrating the benefits of the present invention in terms of dynamic weight accuracy, conflict resolution efficiency, and model generalization ability.

[0168] It should be noted that for the method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of the present invention are not limited by the described action sequence, because according to the embodiments of the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0169] Based on the same inventive concept, an embodiment of the present invention provides a data processing device 500. Refer to Figure 5 , Figure 5 which is a structural block diagram of the data processing device provided by an embodiment of the present invention. As Figure 5 shown, the device includes: A data acquisition module, configured to obtain the current vehicle networking data of the vehicle itself and various in-vehicle sensor data. The current vehicle networking data represents the detection results of other vehicles around the vehicle on the driving environment around the vehicle, and the various in-vehicle sensor data represents the detection results of various in-vehicle sensors of the vehicle itself on the driving environment around the vehicle; A score determination module, configured to perform a credibility evaluation on the current vehicle networking data to obtain a credibility score of the current vehicle networking data, and perform a credibility evaluation on the various in-vehicle sensor data respectively to obtain credibility scores of the various in-vehicle sensor data; A data fusion module, configured to use the credibility scores of the current vehicle networking data and the various in-vehicle sensor data as fusion weights to fuse the current vehicle networking data and the various in-vehicle sensor data to obtain fused data, and the fused data represents the final detection result of the driving environment around the vehicle.

[0170] Optionally, the score determination module includes: A first determination module, configured to obtain a communication quality parameter and obtain a communication quality score of the current vehicle networking data based on the communication quality parameter; A second determination module, configured to obtain the time difference between the sending time and the receiving time of the current vehicle networking data, and obtain an aging attenuation score of the current vehicle networking data based on the time difference and the attenuation rate; A third determination module, configured to perform a consistency analysis on the current vehicle networking data and the various in-vehicle sensor data, and obtain a consistency score of the current vehicle networking data based on the result of the consistency analysis; A first evaluation module, configured to obtain a credibility score of the current vehicle networking data based on the communication quality score, the aging attenuation score, and the consistency score of the current vehicle networking data.

[0171] Optionally, the various in-vehicle sensor data at least includes image data collected by an in-vehicle camera; the score determination module at least includes: A fourth determination module, configured to perform a structural similarity analysis on the image data collected by the in-vehicle camera and reference image data to obtain a structural similarity score of the in-vehicle camera; A fifth determination module, configured to perform edge detection on the image data collected by the in-vehicle camera and obtain a clarity score of the in-vehicle camera according to the edge detection result; The sixth determination module is configured to obtain the ambient light brightness of the driving environment around the vehicle, and obtain the illumination score of the in-vehicle camera according to the ambient light brightness and the headlight on state; The second evaluation module is configured to obtain the credibility score of the image data collected by the in-vehicle camera based on the structural similarity score, clarity score, and illumination score of the in-vehicle camera.

[0172] Optionally, the multiple in-vehicle sensor data at least includes the current point cloud data collected by the in-vehicle millimeter-wave radar; the score determination module at least includes: The seventh determination module is configured to obtain the point cloud density score of the in-vehicle millimeter-wave radar according to the number of target point clouds represented by the current point cloud data collected by the in-vehicle millimeter-wave radar and the average number of point clouds represented by multiple first historical point cloud data; The eighth determination module is configured to obtain the velocity consistency score of the in-vehicle millimeter-wave radar according to the Doppler velocity variance of the same target in the current point cloud data collected by the in-vehicle millimeter-wave radar and the multiple first historical point cloud data; The ninth determination module is configured to obtain the millimeter-wave radar signal strength, and obtain the signal-to-noise ratio score of the in-vehicle millimeter-wave radar based on the magnitude relationship between the millimeter-wave radar signal strength and the millimeter-wave radar signal strength threshold; The tenth determination module is configured to obtain the multipath suppression score of the in-vehicle millimeter-wave radar according to the proportion of invalid point clouds in the current point cloud data collected by the in-vehicle millimeter-wave radar; The third evaluation module is configured to obtain the credibility score of the current point cloud data collected by the in-vehicle millimeter-wave radar based on the point cloud density score, velocity consistency score, signal-to-noise ratio score, and multipath suppression score of the in-vehicle millimeter-wave radar.

[0173] Optionally, the multiple in-vehicle sensor data at least includes the current point cloud data collected by the in-vehicle lidar; the score determination module at least includes: The eleventh determination module is configured to obtain the reflectivity score of the in-vehicle lidar according to the target reflectivity of the current point cloud data collected by the in-vehicle lidar and the reflectivity threshold; The twelfth determination module is configured to obtain the visibility and rainfall intensity of the driving environment around the vehicle, and obtain the environmental attenuation score of the in-vehicle lidar; The thirteenth determination module is configured to obtain the trajectory continuity score of the in-vehicle lidar according to the target position detected in the current point cloud data collected by the in-vehicle lidar for the same target and the target position predicted from the historical trajectory of the target in multiple second historical point cloud data; A fourth evaluation module, configured to obtain a credibility score of the current point cloud data collected by the vehicle-mounted lidar based on the reflectivity score, the environmental attenuation score, and the trajectory continuity score of the vehicle-mounted lidar.

[0174] Optionally, the apparatus further includes: A detection module, configured to detect whether there is a conflict between the detection results represented by any two of the current vehicle networking data and the multiple vehicle-mounted sensor data after obtaining the current vehicle networking data of the vehicle and the multiple vehicle-mounted sensor data. A first determination module, configured to determine whether to adopt the current vehicle networking data according to the vehicle networking data within a first time period and the target vehicle-mounted sensor data within the first time period when there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data; the target vehicle-mounted sensor data is the detection result of the target vehicle-mounted sensor of the vehicle for the driving environment around the vehicle, and the first vehicle-mounted sensor data is any one of the multiple vehicle-mounted sensor data. A data fusion module, including: A data fusion sub-module, configured to, when adopting the current vehicle networking data, use the credibility scores of the current vehicle networking data and the remaining multiple vehicle-mounted sensor data except the first vehicle-mounted sensor data as fusion weights, and fuse the current vehicle networking data and the remaining multiple vehicle-mounted sensor data to obtain the fused data.

[0175] Optionally, the apparatus further includes: An adjacent vehicle determination module, configured to determine multiple adjacent vehicles based on the communication quality, relative position, and respective credit scores of multiple surrounding vehicles of the vehicle when the target vehicle-mounted sensor fails or when the vehicle triggers a cooperative perception request. A result return module, configured to send a cooperative perception request to the multiple adjacent vehicles, and receive cooperative perception responses returned by the multiple adjacent vehicles. A cooperative perception response returned by an adjacent vehicle represents the detection result of the adjacent vehicle for the driving environment around the vehicle. A result fusion module, configured to use the communication quality, relative position, and credit score of the vehicle with each adjacent vehicle as the weight of the adjacent vehicle, and perform weighted fusion on the cooperative perception responses returned by the multiple adjacent vehicles to obtain a final cooperative perception response. A comparison module, configured to compare the final cooperative perception response with the current vehicle networking data to determine whether to adopt the current vehicle networking data.

[0176] Optionally, the apparatus further includes: The first acquisition module is configured to obtain the actual driving environment around the vehicle through the traffic management system when the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data belongs to the first target conflict type; The second acquisition module is configured to retrieve and analyze the multi-vehicle trajectory data during the conflict time period through the cloud server to obtain the actual driving environment around the vehicle when the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data belongs to the second target conflict type; The second determination module is configured to determine whether to trust the current vehicle networking data based on the actual driving environment around the vehicle and the current vehicle networking data.

[0177] Optionally, the credibility evaluation of the current vehicle networking data and the credibility evaluations of the various vehicle-mounted sensor data are implemented through the credibility evaluation model of the vehicle; the apparatus further includes: The third acquisition module is configured to obtain the target geographical grid where the vehicle is currently located, and a geographical area includes multiple geographical grids of a target size; The loading module is configured to load the target credibility evaluation model parameters corresponding to the target geographical grid; The score determination module includes: The score determination sub-module is configured to use the credibility evaluation model with the target credibility evaluation model parameters to perform the credibility evaluation on the current vehicle networking data and the various vehicle-mounted sensor data respectively.

[0178] Optionally, the credibility evaluation of the current vehicle networking data is implemented through the vehicle networking credibility evaluation model in the credibility evaluation model of the vehicle; the apparatus further includes: The first adjustment module is configured to increase the weight corresponding to the communication quality item in the vehicle networking credibility evaluation model from the first weight to the second weight when the current moment belongs to the first target time period; The second adjustment module is configured to restore the weight corresponding to the communication quality item in the vehicle networking credibility evaluation model to the first weight when the current moment belongs to the second target time period.

[0179] Optionally, the data fusion module includes: The first processing module is configured to, when the current vehicle networking data and the various vehicle-mounted sensor data are discrete, use the credibility scores of the current vehicle networking data and the various vehicle-mounted sensor data as fusion weights to fuse the current vehicle networking data and the various vehicle-mounted sensor data through a Bayesian network to obtain the fused data; A second processing module, which is used when the current vehicle networking data and the multiple vehicle-mounted sensor data are of a continuous type, takes the respective credibility scores of the current vehicle networking data and the multiple vehicle-mounted sensor data as fusion weights, and fuses the current vehicle networking data and the multiple vehicle-mounted sensor data through weighted Kalman filtering to obtain the fused data.

[0180] Based on the same inventive concept, another embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps in the data processing method described in any one of the above embodiments of the present invention.

[0181] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, as Figure 6 shown. Figure 6 FIG. is a schematic diagram of an electronic device shown in an embodiment of the present invention. The electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes, it implements the steps in the data processing method described in any one of the above embodiments of the present invention.

[0182] Based on the same inventive concept, another embodiment of the present invention provides a vehicle, which at least includes a controller. When the controller executes, it implements the steps in the data processing method described in any one of the above embodiments of the present invention.

[0183] For the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, please refer to the partial description of the method embodiments.

[0184] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0185] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0186] Embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate a device for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or multiple blocks.

[0187] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or multiple blocks.

[0188] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or multiple blocks.

[0189] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

[0190] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or terminal device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or terminal device comprising the element.

[0191] The above has introduced in detail a data processing method, apparatus, device, medium and vehicle provided by the present invention. Specific examples are used in this text to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A data processing method, characterized in that, The method includes: Obtaining the current vehicle networking data of the vehicle and various in-vehicle sensor data, where the current vehicle networking data represents the detection results of other vehicles around the vehicle on the driving environment around the vehicle, and the various in-vehicle sensor data represents the detection results of various in-vehicle sensors of the vehicle on the driving environment around the vehicle; Performing a credibility assessment on the current vehicle networking data to obtain a credibility score of the current vehicle networking data, and performing a credibility assessment on the various in-vehicle sensor data respectively to obtain the credibility scores of the various in-vehicle sensor data; Using the credibility scores of the current vehicle networking data and the various in-vehicle sensor data as fusion weights to fuse the current vehicle networking data and the various in-vehicle sensor data to obtain fused data, where the fused data represents the final detection result of the driving environment around the vehicle.

2. The data processing method according to claim 1, wherein Performing a credibility assessment on the current vehicle networking data to obtain a credibility score of the current vehicle networking data, including: Obtaining a communication quality parameter and obtaining a communication quality score of the current vehicle networking data based on the communication quality parameter; Obtaining the time difference from the sending time to the receiving time of the current vehicle networking data, and obtaining an aging attenuation score of the current vehicle networking data based on the time difference and the attenuation rate; Performing a consistency analysis on the current vehicle networking data and the various in-vehicle sensor data, and obtaining a consistency score of the current vehicle networking data based on the result of the consistency analysis; Obtaining the credibility score of the current vehicle networking data based on the communication quality score, the aging attenuation score, and the consistency score of the current vehicle networking data.

3. The data processing method according to claim 1, wherein The various in-vehicle sensor data at least includes image data collected by an in-vehicle camera; Performing a credibility assessment on the various in-vehicle sensor data respectively to obtain the credibility scores of the various in-vehicle sensor data, at least including: Performing a structural similarity analysis on the image data collected by the in-vehicle camera and reference image data to obtain a structural similarity score of the in-vehicle camera; Performing edge detection on the image data collected by the in-vehicle camera and obtaining a clarity score of the in-vehicle camera according to the edge detection result; Obtaining the ambient light brightness of the driving environment around the vehicle and obtaining a lighting score of the in-vehicle camera according to the ambient light brightness and the headlight on state; Obtaining the credibility score of the image data collected by the in-vehicle camera based on the structural similarity score, the clarity score, and the lighting score of the in-vehicle camera.

4. The data processing method according to claim 1, wherein The various in-vehicle sensor data at least includes current point cloud data collected by an in-vehicle millimeter-wave radar; Performing a credibility assessment on the various in-vehicle sensor data respectively to obtain the credibility scores of the various in-vehicle sensor data, at least including: Obtaining a point cloud density score of the in-vehicle millimeter-wave radar according to the number of target point clouds represented by the current point cloud data collected by the in-vehicle millimeter-wave radar and the average number of point clouds represented by multiple first historical point cloud data; Obtaining a speed consistency score of the in-vehicle millimeter-wave radar according to the Doppler speed variance of the same target in the current point cloud data collected by the in-vehicle millimeter-wave radar and the multiple first historical point cloud data; Obtain the millimeter-wave radar signal intensity, and based on the magnitude relationship between the millimeter-wave radar signal intensity and the millimeter-wave radar signal intensity threshold, obtain the signal-to-noise ratio score of the vehicle-mounted millimeter-wave radar; Obtain the multipath suppression score of the vehicle-mounted millimeter-wave radar according to the proportion of invalid point clouds in the current point cloud data collected by the vehicle-mounted millimeter-wave radar; Based on the point cloud density score, speed consistency score, signal-to-noise ratio score, and multipath suppression score of the vehicle-mounted millimeter-wave radar, obtain the credibility score of the current point cloud data collected by the vehicle-mounted millimeter-wave radar.

5. The data processing method according to claim 1, characterized in that, The multiple vehicle-mounted sensor data at least includes the current point cloud data collected by the vehicle-mounted lidar; perform credibility evaluation on the multiple vehicle-mounted sensor data respectively to obtain the credibility scores of the multiple vehicle-mounted sensor data respectively, at least including: Obtain the reflectivity score of the vehicle-mounted lidar according to the target reflectivity and the reflectivity threshold of the current point cloud data collected by the vehicle-mounted lidar; Obtain the visibility and rainfall intensity of the driving environment around the vehicle to obtain the environmental attenuation score of the vehicle-mounted lidar; According to the target position detected in the current point cloud data collected by the vehicle-mounted lidar for the same target and the target position predicted based on the historical trajectory of the target in multiple second historical point cloud data, obtain the trajectory continuity score of the vehicle-mounted lidar; Based on the reflectivity score, environmental attenuation score, and trajectory continuity score of the vehicle-mounted lidar, obtain the credibility score of the current point cloud data collected by the vehicle-mounted lidar.

6. The data processing method according to claim 1, wherein After obtaining the current vehicle networking data and multiple vehicle-mounted sensor data of the vehicle, the method further includes: Detect whether there is a conflict between the detection results represented by the current vehicle networking data and the detection results represented by the multiple vehicle-mounted sensor data taken pairwise; In the case where there is a conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first vehicle-mounted sensor data, determine whether to adopt the current vehicle networking data according to the vehicle networking data within the first time period and the target vehicle-mounted sensor data within the first time period; the target vehicle-mounted sensor data is the detection result of the target vehicle-mounted sensor of the vehicle for the driving environment around the vehicle, and the first vehicle-mounted sensor data is any one of the multiple vehicle-mounted sensor data; Using the credibility scores of the current vehicle networking data and the multiple vehicle-mounted sensor data as fusion weights, fuse the current vehicle networking data and the multiple vehicle-mounted sensor data to obtain the fused data, including: In the case of adopting the current vehicle networking data, using the credibility scores of the current vehicle networking data and the remaining multiple vehicle-mounted sensor data except the first vehicle-mounted sensor data as fusion weights, fuse the current vehicle networking data and the remaining multiple vehicle-mounted sensor data to obtain the fused data.

7. The data processing method according to claim 6, wherein The method further includes: In the case where the target vehicle-mounted sensor fails, or, in the case where the vehicle triggers a cooperative perception request, determine multiple neighboring vehicles based on the communication quality, relative position between the vehicle and multiple surrounding vehicles, and the respective credit scores of the multiple surrounding vehicles; Send cooperative perception requests to the multiple neighboring vehicles, and receive the cooperative perception responses returned by the multiple neighboring vehicles. The cooperative perception response returned by one neighboring vehicle represents the detection result of the driving environment around this vehicle. Use the communication quality, relative position between this vehicle and each neighboring vehicle, and the credit score of each neighboring vehicle as the weight of that neighboring vehicle, and perform weighted fusion on the cooperative perception responses returned by the multiple neighboring vehicles to obtain the final cooperative perception response. Compare the final cooperative perception response with the current vehicle networking data to determine whether to adopt the current vehicle networking data.

8. The data processing method according to claim 7, characterized in that, The method further includes: When the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first in-vehicle sensor data belongs to the first target conflict type, obtain the actual driving environment around this vehicle through the traffic management system. When the type of conflict between the detection result represented by the current vehicle networking data and the detection result represented by the first in-vehicle sensor data belongs to the second target conflict type, retrieve and analyze the multi-vehicle trajectory data during the conflict time period through the cloud server to obtain the actual driving environment around this vehicle. Based on the actual driving environment around this vehicle and the current vehicle networking data, determine whether to adopt the current vehicle networking data.

9. The data processing method according to any one of claims 1 to 8, characterized in that The credibility assessment of the current vehicle networking data and the credibility assessments of the various in-vehicle sensor data are implemented through the credibility assessment model of this vehicle. The method further includes: Obtain the target geographical grid where this vehicle is currently located. A geographical area includes multiple geographical grids of the target size. Load the target credibility assessment model parameters corresponding to the target geographical grid. The credibility assessment of the current vehicle networking data and the credibility assessments of the various in-vehicle sensor data respectively include: Use the credibility assessment model with the target credibility assessment model parameters to perform the credibility assessment on the current vehicle networking data and the credibility assessments of the various in-vehicle sensor data respectively.

10. The data processing method according to any one of claims 1 to 8, characterized in that The credibility assessment of the current vehicle networking data is implemented through the vehicle networking credibility assessment model in the credibility assessment model of this vehicle. The method further includes: When the current moment belongs to the first target time period, increase the weight corresponding to the communication quality item in the vehicle networking credibility assessment model from the first weight to the second weight. When the current moment belongs to the second target time period, restore the weight corresponding to the communication quality item in the vehicle networking credibility assessment model to the first weight.

11. The data processing method according to any one of claims 1 to 8, characterized in that, Use the credibility scores of the current vehicle networking data and the various in-vehicle sensor data respectively as the fusion weights to fuse the current vehicle networking data and the various in-vehicle sensor data to obtain the fused data, including: When the current vehicle networking data and the various in-vehicle sensor data are discrete, use the credibility scores of the current vehicle networking data and the various in-vehicle sensor data respectively as the fusion weights, and fuse the current vehicle networking data and the various in-vehicle sensor data through a Bayesian network to obtain the fused data. When the current vehicle networking data and the multiple in-vehicle sensor data are of continuous type, using the credibility scores of the current vehicle networking data and the multiple in-vehicle sensor data respectively as fusion weights, the current vehicle networking data and the multiple in-vehicle sensor data are fused through weighted Kalman filtering to obtain the fused data.

12. A data processing device, characterized in that, The device includes: A data acquisition module, configured to obtain the current vehicle networking data of the vehicle itself and multiple in-vehicle sensor data, where the current vehicle networking data represents the detection results of other vehicles around the vehicle itself on the driving environment around the vehicle itself, and the multiple in-vehicle sensor data represents the detection results of multiple in-vehicle sensors of the vehicle itself on the driving environment around the vehicle itself; A score determination module, configured to perform credibility evaluation on the current vehicle networking data to obtain the credibility score of the current vehicle networking data, and perform credibility evaluation on the multiple in-vehicle sensor data respectively to obtain the credibility scores of the multiple in-vehicle sensor data respectively; A data fusion module, configured to use the credibility scores of the current vehicle networking data and the multiple in-vehicle sensor data respectively as fusion weights to fuse the current vehicle networking data and the multiple in-vehicle sensor data to obtain the fused data, where the fused data represents the final detection results of the driving environment around the vehicle itself.

13. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the data processing method according to any one of claims 1 to 11.

14. A computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by the processor, it implements the data processing method according to any one of claims 1 to 11.

15. A vehicle, comprising at least a controller, characterized in that, When the controller is executed, it implements the data processing method according to any one of claims 1 to 11.

Images