Multi-source heterogeneous data processing device and method for vehicle

Abstract

The invention discloses a multi-source heterogeneous data processing device and method for a vehicle, and the device is applied to a target vehicle, and the device comprises an interface adaption module which is integrated with a first interface, a second interface and a third interface, the heterogeneous original data of the plurality of data sources comprises vehicle state data of a target vehicle received through a first interface, roadside sensing data received through a second interface and traffic event data received through a third interface; a central processing unit and a time division multiplexing bus, the central processing unit is connected with the interface adaptation module through the time division multiplexing bus, and the central processing unit is used for obtaining fused basic probability distribution according to the vehicle state data, the roadside sensing data and the traffic event number; and controlling the target vehicle according to the fused basic probability distribution. According to the invention, the technical problem that the decision-making efficiency of intelligent driving is affected due to poor transmission real-time performance of multi-source data for intelligent driving control is solved.

Description

Multi-source heterogeneous data processing device and method for vehicles Technical Field

[0001] This invention belongs to the field of intelligent driving technology, and in particular relates to a multi-source heterogeneous data processing device and method for vehicles. Background Technology

[0002] Existing data processing platforms employ a single-interface design, requiring multi-source data to undergo protocol conversion through multiple independent gateways. For example, standard datasets transmitted by a roadside device via a mobile communication module and driving data from the vehicle's CAN (Controller Area Network) bus need to be connected to different modules. This increases system integration complexity by more than 30%, and data synchronization latency typically exceeds 100 milliseconds, failing to meet the real-time requirement of no more than 200 milliseconds for edge cloud processing. The high data synchronization latency results in poor real-time transmission of multi-source data used for intelligent driving control, impacting the decision-making efficiency of intelligent driving. Summary of the Invention

[0003] This invention provides a multi-source heterogeneous data processing device and method for vehicles, which solves the technical problem that the real-time transmission of multi-source data required for intelligent driving control is poor in the prior art, affecting the decision-making efficiency of intelligent driving.

[0004] According to a first aspect of the present invention, a multi-source heterogeneous data processing device for a vehicle is provided, applied to a target vehicle. The multi-source heterogeneous data processing device includes: an interface adaptation module for receiving heterogeneous raw data from multiple data sources; the interface adaptation module integrating a first interface, a second interface, and a third interface; the heterogeneous raw data from the multiple data sources including vehicle status data of the target vehicle received through the first interface, roadside perception data received through the second interface, and traffic event data received through the third interface; a central processing unit (CPU) and a time-division multiplexing bus; the CPU being connected to the interface adaptation module via the time-division multiplexing bus; the CPU controlling the time-division multiplexing bus to receive data according to a preset scheduling cycle, so that the first interface, the second interface, and the third interface... The system receives vehicle status data, roadside perception data, and traffic event data in turn. The central processing unit (CPU) constructs a three-dimensional feature space based on the vehicle status data, roadside perception data, and traffic event data, corresponding to vehicle-mounted data evidence, roadside data evidence, and event data evidence. It then generates basic probability distributions for the vehicle-mounted data evidence, roadside data evidence, and event data evidence for the target driving problem, each basic probability distribution including the support for each decision objective of the target driving problem. The CPU fuses the basic probability distributions to obtain a fused basic probability distribution, which includes new support for each decision objective of the target driving problem. Finally, it controls the target vehicle based on the fused basic probability distribution.

[0005] In conjunction with the first aspect, in some embodiments, the central processing unit is configured to: preprocess the vehicle state data from the vehicle-mounted side to obtain a first standard dataset; preprocess the roadside perception data from the roadside to obtain a second standard dataset; preprocess the traffic event data from the traffic event detector to obtain a third standard dataset; construct a three-dimensional feature space for the first standard dataset to obtain the vehicle-mounted data evidence; construct a three-dimensional feature space for the second standard dataset to obtain the roadside data evidence; and construct a three-dimensional feature space for the third standard dataset to obtain the event data evidence.

[0006] In conjunction with the first aspect, in some embodiments, the first interface is a CAN bus interface, used to receive various vehicle status signals, parse each vehicle status signal according to a structured format and store it to obtain the vehicle status data; the second interface is a mobile communication module, used to parse the received mobile communication signals to extract various roadside sensing signals, store each roadside sensing signal by device identifier to obtain the roadside sensing data; the third interface is an Ethernet interface, used to receive traffic event alarms and cache the received traffic event alarms to obtain the traffic event data.

[0007] In conjunction with the first aspect, in some embodiments, the central processing unit is configured to: remove anomalous data frames from the vehicle state data using an anomaly detection model constructed based on the isolated forest algorithm, and fill in missing values ​​in the vehicle state data using a time-series prediction model to obtain a first standard dataset; remove anomalous data frames from the roadside perception data using the anomaly detection model, and fill in missing values ​​in the roadside perception data using the time-series prediction model to obtain a second standard dataset; and remove anomalous data frames from the traffic event data using the anomaly detection model, and fill in missing values ​​in the traffic event data using the time-series prediction model to obtain the third standard dataset.

[0008] In conjunction with the first aspect, in some embodiments, the central processing unit is configured to: construct a three-dimensional feature space of time dimension, spatial dimension, and attribute dimension for the first standard dataset to obtain the vehicle-mounted data evidence, wherein the vehicle-mounted data evidence includes time dimension features, spatial dimension features, and attribute dimension features; construct a three-dimensional feature space of time dimension, spatial dimension, and attribute dimension for the second standard dataset to obtain the roadside data evidence, wherein the roadside data evidence includes time dimension features, spatial dimension features, and attribute dimension features; and construct a three-dimensional feature space of time dimension, spatial dimension, and attribute dimension for the third standard dataset to obtain the event data evidence, wherein the event data evidence includes time dimension features, spatial dimension features, and attribute dimension features.

[0009] In conjunction with the first aspect, in some embodiments, the central processing unit is configured to: fuse the three basic probability assignments corresponding to the vehicle-mounted data evidence, the roadside data evidence, and the event data evidence to obtain a fused basic probability assignment; determine a conflict coefficient based on the three basic probability assignments; if the conflict coefficient is less than a preset coefficient threshold, obtain the fused basic probability assignment; if the conflict coefficient is greater than or equal to the preset coefficient threshold, correct the three basic probability assignments and re-fuse them to obtain the fused basic probability assignment.

[0010] In conjunction with the first aspect, in some embodiments, the central processing unit is configured to: after obtaining the fused basic probability allocation, obtain a historical basic probability allocation from a historical database as a reference basic probability allocation based on the fused basic probability allocation and traffic state data representing the current traffic state, wherein the historical database stores historical basic probability allocations corresponding to different historical time periods, different spatial grids, and different traffic states; determine the deviation rate between the fused basic probability allocation and the reference basic probability allocation; if the deviation rate is less than a preset deviation threshold, control the target vehicle based on the fused basic probability allocation; if the deviation rate is greater than or equal to the preset deviation threshold, correct the three basic probability allocations and then re-fuse them.

[0011] In conjunction with the first aspect, in some embodiments, the central processing unit is configured to: schedule the CAN bus to receive data in a first time slot within the scheduling period, schedule the mobile communication module to receive data in a second time slot within the scheduling period, and schedule the Ethernet interface to receive data in a third time slot within the scheduling period, wherein the first time slot, the second time slot, and the third time slot are the division results of the scheduling period; the interface adaptation module further includes a hardware timer, which is used to control the switching of the first time slot, the second time slot, and the third time slot; the multi-source heterogeneous data processing device further includes a timing module based on the Global Navigation Satellite System, which is used to add a timestamp to each data frame of the received vehicle status data, the roadside perception data, and the traffic event data.

[0012] In conjunction with the first aspect, in some embodiments, the central processing unit is further configured to: perform format conversion on the received vehicle status data, roadside perception data, and traffic event data based on a predefined standardized data structure, converting them into serialized data of the same format for storage, wherein each piece of serialized data includes a unique data identifier, a timestamp, a data source type, a device identifier, core data fields, and data precision.

[0013] According to a second aspect of the present invention, a multi-source heterogeneous data processing method is provided, comprising: controlling a time-division multiplexing bus to schedule a first interface, a second interface, and a third interface of an interface adaptation module to receive data in turn according to a preset scheduling cycle, such that the first interface receives vehicle status data of a target vehicle, the second interface receives roadside perception data, and the third interface receives traffic event data; constructing a three-dimensional feature space based on the vehicle status data, the roadside perception data, and the traffic event data to obtain corresponding vehicle-mounted data evidence, roadside data evidence, and event data evidence; generating basic probability distributions for the vehicle-mounted data evidence, the roadside data evidence, and the event data evidence for a target driving problem, each basic probability distribution including the support of each decision objective for the target driving problem; fusing the basic probability distributions to obtain a fused basic probability distribution, the fused basic probability distribution including new support of each decision objective for the target driving problem; and controlling the target vehicle according to the fused basic probability distribution.

[0014] The embodiments of this invention provide one or more technical solutions, achieving at least the following technical effects or advantages: By configuring an interface adaptation module integrating a first interface, a second interface, and a third interface, the central processing unit controls the time-division multiplexing bus to receive data according to a preset scheduling cycle, enabling the first, second, and third interfaces to take turns directly receiving vehicle status data, roadside perception data, and traffic event data. This technical solution enables direct access to heterogeneous raw data from three different data sources—vehicle-side, roadside, and traffic event—without requiring protocol conversion through multiple independent gateways. This improves the real-time transmission performance of multi-source heterogeneous data and reduces the synchronization latency. Reducing the synchronization latency of multi-source data enhances the efficiency of intelligent driving decision-making using multi-source data. Based on the heterogeneous raw data from multiple sources (vehicle status data, roadside perception data, and traffic event data), vehicle-side data evidence, roadside data evidence, and event data evidence are generated. Then, based on these three pieces of evidence, basic probability assignments for the target driving problem are generated. Finally, these basic probability assignments are fused, achieving efficient and reliable fusion decision-making. By utilizing the fused basic probability allocation to control the target vehicle, the control reliability of intelligent driving is improved. Simultaneously enhancing the decision-making efficiency and reliability of intelligent driving can improve the safety of intelligent driving control.

[0015] Based on the implementation methods provided in the above aspects, the present invention can be further combined to provide more implementation methods. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 shows a schematic diagram of the structure of a multi-source heterogeneous data processing device according to some embodiments of the present invention; Figure 2 shows the data processing flow of the multi-source heterogeneous data processing device according to some embodiments of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In the embodiments of this specification, the term "multiple" means "two or more", that is, including two or more cases; the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] This invention provides a multi-source heterogeneous data processing device for a vehicle, applied to a target vehicle. Figure 1 shows a schematic diagram of the structure of the multi-source heterogeneous data processing device according to some embodiments of the present invention, and Figure 2 shows the data processing flow of the multi-source heterogeneous data processing device in some embodiments of the present invention. As shown in Figures 1 and 2, the multi-source heterogeneous data processing device includes: an interface adaptation module 110, a central processing unit 130, and a time-division multiplexing bus 120. The central processing unit 130 is connected to the interface adaptation module 110 through the time-division multiplexing bus 120. The interface adaptation module 110 integrates a first interface 111, a second interface 112, and a third interface.

[0021] The interface adaptation module 110 is used to receive heterogeneous raw data from multiple data sources. In some embodiments, the interface adaptation module 110 is used to receive heterogeneous raw data from three data sources. In some embodiments, the three data sources are: vehicle-side, road-side, and cloud-side. The interface adaptation module 110 enables direct access to heterogeneous raw data from the vehicle-side, road-side, and cloud-side through the first interface 111, the second interface 112, and the third interface 113, without the need for multiple independent gateways for protocol conversion.

[0022] In some embodiments, the heterogeneous raw data from the three data sources includes vehicle status data of the target vehicle received through the first interface 111, roadside perception data received through the second interface 112, and traffic event data received through the third interface 113.

[0023] Based on a predefined standardized data structure, the received vehicle status data, roadside perception data, and traffic event data are converted into serialized data of the same format for storage. Each piece of serialized data includes a unique data identifier, timestamp, data source type, device identifier, core data fields, and data precision.

[0024] In some embodiments, the first interface 111 is a CAN (Controller Area Network) bus interface, used to receive various vehicle status signals, parse each vehicle status signal according to a structured format, and store it to obtain vehicle status data. The CAN bus interface is compatible with the ISO11898 standard and supports adaptive switching between 500kbps and 1Mbps transmission rates. It receives vehicle status signals (such as vehicle speed, acceleration, tire pressure, etc.) sent by various on-board ECUs (Electronic Control Units) of the target vehicle in a differential signal manner. The on-board ECUs include engine control, braking control, tire pressure monitoring, etc.

[0025] In some embodiments, the CAN bus interface has a built-in voltage monitoring circuit. This circuit uses a combination of voltage divider resistors and a comparator to monitor the bus voltage of the CAN bus interface in real time. When the bus voltage of the CAN bus interface is detected to be greater than a first voltage threshold (e.g., 4.8V), it is determined that the target vehicle has entered the start-up state, and the acquisition of vehicle status data is activated. When the bus voltage is detected to be less than a second voltage threshold (e.g., 1.2V), it is determined that the target vehicle has entered the off state, and the CAN bus interface enters a low-power sleep mode. In the low-power sleep mode, the power consumption is reduced from 1.2 watts to 0.15 watts, ensuring reduced energy consumption when the target vehicle is off.

[0026] After filtering out invalid and redundant frames from the received vehicle status signals, the system parses and extracts structured data including timestamps, signal names, and signal values, thus reducing parsing latency. For example, the structured data extracted from the vehicle speed signal is: "1620000000 milliseconds_vehicle speed_60 kilometers per hour", and the structured data extracted from the acceleration signal is: "1620000001 milliseconds_acceleration_0.8 meters per square second".

[0027] In some embodiments, the second interface 112 is a mobile communication module, such as a 5G communication module or a more advanced mobile communication module. The mobile communication module is used to parse the received mobile communication signals to extract various roadside sensing signals, and to store each roadside sensing signal by distinguishing it with a device identifier to obtain roadside sensing data.

[0028] Understandably, the coordinates, speed, and other information of the RSU (Road Side Unit) are extracted, and the device ID of the RSU is added before storage, such as: "RSU-001_Target 1_Coordinates_UTM (352000, 483000)". In some embodiments, a frame filtering mechanism is used when parsing mobile communication signals to filter out invalid and redundant frames, thereby reducing parsing latency.

[0029] In some embodiments, the mobile communication module employs a V2X (Vehicle to Everything) chip supporting the 3GPP Release 16 standard, using a PC5 (Proximity Communication Interface 5) interface for direct communication, operating in the 5905-5925MHz frequency band. Various roadside sensing signals include those from multiple roadside devices, such as radar point cloud data, camera video streams, and traffic light status. The mobile communication module receives roadside sensing signals from each device at a rate of 1Gbps. The module incorporates priority judgment logic. When the priority of the roadside sensing signal to be received is greater than a preset priority, it automatically switches to the first receiving mode. If the priority of the roadside sensing signal to be received is less than or equal to the preset priority, it maintains the second receiving mode. The transmission rate of the second receiving mode is lower than that of the first receiving mode to reduce buffering latency for receiving important data.

[0030] It is understandable that roadside sensing signals with higher than preset priority can include: sudden obstacle warnings, sudden changes in traffic light status, etc.

[0031] For example, roadside sensing signals can be divided into five priorities (level 1 is the lowest and level 5 is the highest). When a roadside sensing signal with a priority greater than 3 is detected, it automatically switches to the first receiving mode. In addition to automatically switching to the first receiving mode, the receiving buffer can be increased (for example, from 2MB to 8MB) and unnecessary verification processes can be turned off, which can reduce the reception delay of emergency roadside sensing data and ensure a rapid response to emergency information.

[0032] In some embodiments, the third interface 113 is an Ethernet interface, which is used to receive traffic event alarms and cache the received traffic event alarms to obtain traffic event data. A FIFO (First-In-First-Out) mechanism can be used to write the traffic events received via Ethernet into a cache queue for temporary storage. The cache depth (the maximum number of events that can be stored) is 1000 frames, and an alarm is triggered when it overflows.

[0033] In some embodiments, the Gigabit Ethernet interface conforms to the IEEE 802.3ab standard, supports full-duplex transmission, and is used to receive traffic event alarms sent by traffic event detectors (such as cameras, millimeter-wave radar, traffic flow detectors, etc.). The interface controller of the Gigabit Ethernet interface integrates a DMA (Direct Memory Access) engine, which detects the data frame length of the traffic event alarm through hardware logic. When the frame length is greater than 1500 bytes (such as accident alarm data containing images), DMA transfer is automatically triggered, bypassing the central processing unit 130 to directly write the traffic event into memory. This improves transmission efficiency, reduces CPU utilization, and avoids transmission blockage caused by CPU overload.

[0034] In some embodiments, the central processing unit 130 (CPU, which may adopt an ARM Cortex-A76 architecture) is used to: schedule the CAN bus to receive data in the first time slot of the scheduling period, the mobile communication module to receive data in the second time slot of the scheduling period, and the third time slot of the scheduling period in each scheduling cycle, wherein the first time slot, the second time slot, and the third time slot are the division results of the scheduling cycle.

[0035] For example, the preset scheduling period is 10ms. Each period is divided into three time slots: a first time slot of 3ms, a second time slot of 4ms, and a third time slot of 3ms. A hardware timer ensures that the time synchronization accuracy of the first interface 111, the second interface 112, and the third interface 113 receiving raw data is ≤5ms. The hardware timer enables precise time slot switching, ensuring that there are no conflicts in the reception of raw data by the three types of interfaces.

[0036] In some embodiments, the interface adaptation module 110 further includes a hardware timer, which is used to control the switching of the first time slot, the second time slot and the third time slot.

[0037] In some embodiments, the multi-source heterogeneous data processing device further includes a timing module based on a global navigation satellite system. This timing module adds timestamps to the data frames of each dimension of the received vehicle status data, roadside perception data, and traffic event data. Adding high-precision timestamps to the data frames of all accessed data of each dimension results in higher time synchronization accuracy and further addresses the problem of synchronization latency exceeding 100 milliseconds caused by multi-gateway conversion.

[0038] In some embodiments, the vehicle status data, roadside perception data, and traffic event data received through the first interface 111, the second interface 112, and the third interface 113 are converted into serialized data of the same format based on a predefined standardized data structure. The predefined standardized data structure includes a unique data identifier, timestamp, data source type, device identifier, core data fields, and data precision.

[0039] If the data source is on the vehicle side, the device identifier is the ECU number of the vehicle status signal source. If the data source is on the road side, the device identifier is the RSU unit number. The data source type corresponds one-to-one with the first interface 111, the second interface 112, and the third interface 113.

[0040] To address the issue of heterogeneous formats in multi-source data, some embodiments predefine standardized data structures to uniformly convert data received from all interfaces into serialized data in the same format, such as Protobuf. This serialized data structure includes a unique data identifier, a GNSS timing timestamp (in milliseconds), data source type, device ID (e.g., ECU number, RSU number), core data fields (key-value pairs), and key information such as data accuracy (e.g., GPS positioning accuracy). During format conversion, the protocol type of the data source is automatically identified using a protocol feature library, such as the frame header 0x01 of a CAN bus interface or the UDP port 5683 of a 5G communication module. After parsing the received data frames according to the protocol type, the core data fields of the data frames are extracted and encapsulated, thus reducing conversion latency and ensuring real-time data transmission.

[0041] After receiving vehicle status data, roadside perception data, and traffic event data, the central processing unit 130 further performs the following: constructs a three-dimensional feature space based on the vehicle status data, roadside perception data, and traffic event data to obtain onboard data evidence, roadside data evidence, and event data evidence; generates basic probability assignments (BPAs) for the target driving problem based on the onboard data evidence, roadside data evidence, and event data evidence, respectively, with each basic probability assignment including the support for each decision objective of the target driving problem; fuses the various basic probability assignments to obtain a fused basic probability assignment, which includes new support for each decision objective of the target driving problem; and controls the target vehicle based on the fused basic probability assignment.

[0042] In some embodiments, when constructing a three-dimensional feature space based on vehicle state data, roadside perception data, and traffic event data, the central processing unit 130 is configured to: preprocess vehicle state data from the vehicle side to obtain a first standard dataset; preprocess roadside perception data from the roadside to obtain a second standard dataset; preprocess traffic event data from the traffic event detector to obtain a third standard dataset; construct a three-dimensional feature space on the first standard dataset to obtain vehicle data evidence; construct a three-dimensional feature space on the second standard dataset to obtain roadside data evidence; and construct a three-dimensional feature space on the third standard dataset to obtain event data evidence.

[0043] It should be noted that the vehicle status data, roadside perception data, and traffic event data that are preprocessed are structured data frames formed after the above format conversion and encapsulation. By preprocessing the structured data frames of vehicle status data, roadside perception data, and traffic event data, the first standard dataset, the second standard dataset, and the third standard dataset are obtained, providing reliable input for subsequent fusion decision-making.

[0044] In existing technologies, the detection of abnormal data frames in collected data generally relies on fixed thresholds, resulting in a high false detection rate; missing value filling generally uses mean or linear interpolation, which has a large error in scenarios with sudden changes in traffic flow (such as the switching between morning and evening peak hours).

[0045] To address the issue that existing technologies for anomaly detection and missing value imputation have significant errors when applied to intelligent driving scenarios, in some embodiments of this invention, when the central processing unit 130 preprocesses the vehicle state data from the vehicle side, it uses an anomaly detection model based on the Isolation Forest (iForest) algorithm to remove anomaly frames from the vehicle state data, and employs a time-series prediction model to impute missing values ​​in the vehicle state data, thereby obtaining a first standard dataset.

[0046] To address the issue that existing technologies for anomaly detection and missing value imputation have significant errors when applied to intelligent driving scenarios, in some embodiments, when the central processing unit 130 preprocesses the roadside perception data from the roadside, it uses an anomaly detection model based on the isolated forest algorithm to remove anomaly frames from the roadside perception data, and employs a temporal prediction model to impute missing values ​​in the roadside perception data, thereby obtaining a second standard dataset.

[0047] To address the issue that existing technologies for anomaly detection and missing value imputation have significant errors when applied to intelligent driving scenarios, in some embodiments, when the central processing unit 130 preprocesses traffic event data from the traffic event detector, it uses an anomaly detection model based on the isolated forest algorithm to remove anomaly frames from the traffic event data, and employs a time-series prediction model to impute missing values ​​in the traffic event data, thus obtaining a third standard dataset.

[0048] Understandably, the anomaly detection model described above can consist of multiple isolated trees, such as 100 trees. The construction process for each tree is as follows: a small subset of samples is randomly sampled from the original training set to construct a subset. Anomaly samples are scarce, and a small number of samples can cover the anomaly features, while also reducing computational load and improving the speed of tree construction. Next, core features across multiple dimensions are extracted from the subsample set. A feature dimension is randomly selected from these core features, and a segmentation threshold is randomly chosen between the minimum and maximum eigenvalues ​​of that feature dimension. The samples in the subsample set are then divided into left and right subtrees (samples with eigenvalues ​​less than the segmentation threshold go to the left subtree, and samples with eigenvalues ​​greater than or equal to the segmentation threshold go to the right subtree). This process of randomly selecting feature dimensions and corresponding segmentation thresholds is repeated for both subtrees to recursively divide the subtrees until one of the following stopping conditions is met, at which point the tree construction is complete: Condition 1: Each subtree contains only one sample or the tree depth reaches a preset value (10 levels); Condition 2: One of the subtrees contains only one sample. The shorter the path length of a sample in the tree, the more likely it is to be isolated, resulting in a higher probability of anomaly and a higher anomaly score. The path length of each sample in the tree refers to the number of partitions from the root to the leaf node where the sample is located; the shorter the path, the more likely the sample is to be an anomaly.

[0049] When constructing the anomaly detection model, vehicle speed, acceleration, and GPS positioning variance of roadside equipment are extracted as core features from the original training set to form a three-dimensional feature vector, comprehensively covering the key feature dimensions of vehicle driving status and positioning reliability. The GPS positioning variance is obtained by calculating the coordinate dispersion of five consecutive positioning points, reflecting the stability of the positioning data; the larger the variance, the less reliable the positioning. All dimensional features in the three-dimensional feature vector are normalized and mapped to the interval between 0 and 1 to eliminate the influence of dimensional differences on the model. The construction of each tree in the anomaly detection model can randomly select feature dimensions from the three core features of vehicle speed, acceleration, and GPS positioning variance in the subsample set.

[0050] To further improve detection accuracy, when constructing the anomaly detection model, in addition to extracting vehicle speed, acceleration, and GPS positioning variance of roadside equipment as core features from the original training set, a third-order difference time series (DTA) feature of vehicle speed is also extracted. This DTA represents the rate of change of acceleration and can accurately capture drastic fluctuations in vehicle speed within a short period of time. This third-order difference time series feature of vehicle speed is obtained by calculating the third difference between the current vehicle speed and the vehicle speed at several previous moments, thereby improving the accuracy of detecting abnormal data frames and reducing the false detection rate.

[0051] When using an anomaly detection model to remove abnormal data frames from vehicle status data, roadside perception data, and traffic event data, each received data frame is input into the anomaly detection model, which calculates an anomaly score for that data frame. If the anomaly score is greater than a score threshold (e.g., 0.8), the data frame is determined to be an abnormal data frame. For example, a vehicle speed greater than 300 km / h, an acceleration greater than 100 m / s², and a GPS positioning variance greater than 0.5 are all considered abnormal data frames.

[0052] In some embodiments, to address the problem of low accuracy in filling missing values, this invention employs LSTM (Long Short-Term Memory) to construct a time-series prediction model, thereby achieving accurate repair of missing data.

[0053] In some embodiments, the input layer of the time-series prediction model is used to extract multi-dimensional raw features from vehicle state data, roadside perception data, and traffic event data after removing anomalous data frames. For example, for vehicle state data after removing anomalous data frames, the extracted multi-dimensional raw features include multiple parameters such as vehicle speed, acceleration, fuel consumption, and engine speed. For roadside perception data after removing anomalous data frames, the extracted multi-dimensional raw features include multiple parameters such as GPS positioning, road grade, and weather conditions. For traffic event data after removing anomalous data frames, the extracted multi-dimensional raw features include accident warnings and construction area coordinates. The hidden layer of the time-series prediction model has a multi-layer LSTM network (e.g., a two-layer LSTM network), with each LSTM network containing 128 neurons. Each LSTM network in the hidden layer processes the raw features of each dimension through gating mechanisms and nonlinear transformations of neurons to generate high-level time-series features. Dropout=0.2 is added to the hidden layer to randomly discard 20% of the neuron connections to prevent overfitting in the prediction process. The output layer of the time series prediction model makes predictions based on high-level time series features to obtain predicted values ​​for filling missing data (such as missing rotational speed values) and fills in the missing data to obtain the first standard dataset, the second standard dataset, and the third standard dataset.

[0054] In some embodiments, the time-series prediction model is trained using the Adam optimizer (learning rate 0.001). The training data consists of historical vehicle state data, roadside perception data, and traffic event data, such as vehicle state data, roadside perception data, and traffic event data from the past three months under various driving scenarios, including highway, urban, and suburban driving scenarios. Five-fold cross-validation is used to reduce prediction error, for example, to keep the prediction error below 5%. For instance, if the actual vehicle speed is 60 km / h, the predicted value range is 57-63 km / h. During training, mean squared error can be used as the loss function, and the model is iterated for a preset number of rounds to bring the loss of the time-series prediction model to below 0.002.

[0055] In some embodiments, in order to construct a three-dimensional feature space for the first standard dataset, the second standard dataset, and the third standard dataset, the central processing unit 130 is configured to: construct a three-dimensional feature space for the first standard dataset in terms of time dimension, spatial dimension, and attribute dimension to obtain vehicle-mounted data evidence, wherein the vehicle-mounted data evidence includes time dimension features, spatial dimension features, and attribute dimension features; construct a feature space for the second standard dataset in terms of time dimension, spatial dimension, and attribute dimension to obtain roadside data evidence, wherein the roadside data evidence includes time dimension features, spatial dimension features, and attribute dimension features; and construct a feature space for the third standard dataset in terms of time dimension, spatial dimension, and attribute dimension to obtain event data evidence, wherein the event data evidence includes time dimension features, spatial dimension features, and attribute dimension features.

[0056] Understandably, the construction of the feature space in the time dimension employs a preset sliding time window, aligning data from all data sources by timestamp. For example, a sliding time window of 100ms includes standard data from all data sources within 100ms before and after the current moment: vehicle status data (vehicle speed, acceleration, fuel consumption, engine speed, etc.), roadside perception data (radar GPS positioning, road grade, weather conditions, etc.), and traffic event data (accident alarms, construction area coordinates, etc.). This ensures the continuity of time-series data while controlling computational complexity.

[0057] Understandably, the construction of the feature space in the spatial dimension is based on UTM (Universal Transverse Mercator Projection) projection, which divides the road into a grid of preset sizes. Each grid is assigned a unique ID to link different data sources within the same grid. For example, the feature space construction in the spatial dimension could be based on UTM projection, dividing the road into a 500m × 500m grid. The unique ID assigned to each grid is composed of the UTM plane coordinates of the lower left corner of the grid, such as "Grid-352000-483000". This grid allows for the association of vehicle status data (such as vehicle speed and location within the grid), roadside perception data (such as GPS positioning of radar within the grid, road grade, weather conditions, etc.), and traffic event data (such as accident alarms within the grid) within the same spatial area, achieving the aggregation of spatial features and facilitating subsequent fusion calculations.

[0058] Understandably, the feature space construction at the attribute level involves converting attributes such as vehicle type, road grade, and weather conditions into numerical features using One-Hot encoding. This attribute-level feature space construction yields 18 core features covering three main categories: vehicles, roads, and environment. Vehicle features include vehicle type (passenger car / truck / bus), speed, acceleration, tire pressure, fuel consumption, and engine speed; road features include road grade (highway / urban / suburban), number of lanes, road surface type (asphalt / cement), and speed limit; and environmental features include weather conditions (sunny / rainy / snowy / foggy), visibility, ambient temperature, and road surface humidity. All attribute features are converted into numerical features using One-Hot encoding for subsequent fusion calculations.

[0059] In some embodiments, to fuse vehicle-mounted data evidence, roadside data evidence, and event data evidence, the central processing unit 130 is configured to: generate basic probability assignments corresponding to vehicle-mounted data evidence, roadside data evidence, and event data evidence, each basic probability assignment also referred to as a basic trust assignment, including the support for each decision objective. Next, the three basic probability assignments are fused to obtain a fused basic probability assignment; a conflict coefficient is determined based on the three basic probability assignments; if the conflict coefficient is less than a preset coefficient threshold, the fused basic probability assignment is output; if the conflict coefficient is greater than or equal to the preset coefficient threshold, the three basic probability assignments are corrected and then re-fused to obtain the fused basic probability assignment.

[0060] For example, the target driving problem is a decision-making problem used for autonomous driving control or assisted driving control. Taking the target driving problem "Is there an obstacle ahead?" as an example, the corresponding decision objectives are: obstacle ahead, obstacle ahead, and uncertainty about the presence of an obstacle. These decision objectives constitute the various incompatible answers to the driving problem. For the decision objective "obstacle ahead", the support obtained based on onboard data evidence is 0.6 (based on image recognition confidence), the support obtained based on road test data evidence is 0.7 (based on radar detection confidence), and the support obtained based on traffic event data evidence is 0.9.

[0061] It should be noted that in scenarios involving autonomous driving control or assisted driving control of the target vehicle, various driving questions will be raised. The driving question "Is there an obstacle ahead?" mentioned above is only an example.

[0062] In some embodiments, the Dempster combination rule can be used to select two basic probability assignments from the three basic probability assignments for the target driving problem and fuse them to obtain a first basic probability assignment. The first basic probability assignment is then fused with the remaining basic probability assignment to obtain a fused basic probability assignment, which includes the new support for each decision objective.

[0063] In other embodiments, the Dempster combination rule can be used to fuse the three basic probability assignments for the target driving problem in one go, resulting in a fused basic probability assignment.

[0064] Understandably, the conflict coefficient characterizes the degree of contradiction between two or three types of data evidence. The conflict coefficient is obtained by calculating the sum of the support of all mutually exclusive evidence. The smaller the conflict coefficient, the higher the consistency of the various types of data evidence.

[0065] In some embodiments, when modifying the three basic probability assignments, a belief entropy correction can be used to reduce the weight of conflicting evidence. Belief entropy reflects the uncertainty of the evidence; the higher the uncertainty, the lower the weight, thereby avoiding distortion of the fusion result.

[0066] When the conflict coefficient is no more than 0.7 (low conflict), the fused basic probability assignment is directly output; when the conflict coefficient is greater than 0.7 (high conflict, such as vehicle data supporting "no obstacles" while roadside data supports "obstacles"), the basic probability assignment is corrected using belief entropy to reduce the weight of conflicting evidence. Belief entropy reflects the uncertainty of the evidence; the higher the uncertainty, the lower the weight, thus avoiding distortion of the fusion result.

[0067] In some embodiments, if the fused basic probability allocation satisfies the conflict coefficient being less than a preset coefficient threshold, then the target vehicle is controlled based on the fused basic probability allocation.

[0068] The embodiments of the present invention reduce the time consumption of a single fusion by means of optimization techniques such as feature space dimensionality reduction (retaining only 18 core features) and parallel computing (parallel reasoning of 100 isolated trees), so as to meet the response requirements of L4 autonomous driving in emergency scenarios.

[0069] In other embodiments, to further ensure the reliability of the fusion result, after obtaining the fused basic probability allocation that satisfies the conflict coefficient being less than a preset coefficient threshold, the central processing unit is further configured to: obtain a historical basic probability allocation from a historical database as a reference basic probability allocation based on the fused basic probability allocation and traffic state data representing the current traffic state; determine the deviation rate between the fused basic probability allocation and the reference basic probability allocation; if the deviation rate is less than a preset deviation threshold, control the target vehicle based on the fused basic probability allocation; if the deviation rate is greater than or equal to the preset deviation threshold, correct the three basic probability allocations and re-fuse them to reduce the deviation rate until the fused basic probability allocation satisfies the deviation rate between itself and the reference basic probability allocation being less than the preset deviation threshold.

[0070] In some embodiments, the constructed historical database stores historical basic probability assignments and traffic condition data (such as accident records and congestion conditions) for various driving issues over the past year, indexed by grid ID and time period, for example, "Grid-352000-483000_Weekday 7:00-8:00". When a new fusion result (the fused basic probability assignment) is generated, a historical basic probability assignment from the same grid, time period, and traffic condition is queried from the historical database as a reference basic probability assignment.

[0071] In some embodiments, the calculated deviation rate is the absolute value of the difference between the fused basic probability allocation and the reference basic probability allocation, divided by 100%. When the deviation rate does not exceed 5% (accuracy not less than 95%), the fused basic probability allocation is determined to be reliable and output; when the deviation rate is greater than 5%, re-fusion is triggered until the deviation rate does not exceed 5%. This mechanism makes the fusion accuracy fluctuation smaller in complex scenarios (such as rainy and snowy weather, construction sections), which is superior to the existing technology.

[0072] Existing data processing devices generally suffer from a narrow environmental adaptability range, operating only in conventional environments from -20℃ to 85℃, and are unable to adapt to extreme scenarios such as high temperatures (e.g., 60℃ road surface temperature in desert areas during summer) and low temperatures (e.g., -30℃ in frigid regions during winter). This invention also achieves stable operation over a wide temperature range through material selection and algorithm compensation. The PCB substrate of the interface adapter module 110 is made of polyetheretherketone (PEEK) material. The temperature resistance range of PEEK material is extended from -20℃ to 130℃ for FR-4 to -40℃ to 250℃, and the signal transmission attenuation rate is reduced to 1.2% in a high-temperature environment of 60℃, ensuring stable interface communication.

[0073] The antenna radome of the 5G communication module is made of quartz glass fiber composite material. The dielectric constant of this material is around 3.8, and the fluctuation does not exceed 0.2 in the range of -40℃ to 200℃, so as to avoid communication signal distortion caused by high temperature.

[0074] Regarding temperature compensation in the algorithm, a temperature compensation factor is added to the time-series prediction model. When the ambient temperature is detected to be greater than the first temperature threshold (e.g., 40℃ or a similar value), the predicted value used to fill the missing data of vehicle speed is automatically corrected: based on the physical characteristics of the decrease in tire friction coefficient at high temperatures, the predicted value is lowered by a preset value, such as 5%, further controlling the prediction error within 3.5% (the error does not exceed 5% under normal temperature).

[0075] In terms of hardware material replacement, the sealing ring of the CAN bus interface uses fluorosilicone rubber, with a temperature resistance range extended from -40℃ to -60℃. Its sealing performance retention rate exceeds 95% in a low-temperature environment of -30℃ (compared to only 70% for traditional nitrile rubber), preventing water and dust ingress into the interface due to low temperatures. The PHY (Physical Layer Transceiver) chip of the Ethernet interface uses a temperature-compensated crystal oscillator (TCXO), improving frequency stability to ±0.5ppm in the temperature range of -40℃ to 85℃, reducing clock drift and data transmission error rate at low temperatures. This achieves adaptation to low-temperature environments (-60℃ to 85℃).

[0076] Low temperatures can reduce radar detection range and weaken signal strength, but cameras are less affected by low temperatures. In some embodiments, a temperature attenuation factor is introduced. Before fusing the various basic probability assignments, if the ambient temperature is below a second temperature threshold (e.g., -20°C), the weight of roadside data evidence for the basic probability assignment of the target driving problem is reduced: for each decision objective, the support of roadside data evidence for that decision objective is multiplied by the temperature attenuation factor (e.g., 0.8). While reducing the weight of roadside data evidence for the basic probability assignment of the target driving problem, the weight of vehicle-mounted data evidence for the basic probability assignment of the target driving problem can be increased.

[0077] Based on the same inventive concept, embodiments of the present invention also provide a vehicle infotainment system, including the multi-source heterogeneous data processing device of any of the above embodiments.

[0078] The multi-source heterogeneous data processing device and vehicle-mounted system for vehicles provided in this invention improves the data access rate to meet the large data volume transmission requirements of roadside lidar point clouds, etc.; the anomaly detection accuracy is improved, and the false detection rate is reduced, far below the industry average. The missing value filling error is reduced, and the fusion decision delay is shortened, meeting the emergency response requirements of L4 autonomous driving. System reliability is improved; the multi-source data synchronization accuracy is improved, and the synchronization performance is significantly optimized; the environmental adaptability temperature range is expanded from -20℃ to 85℃ to -60℃ to 250℃, achieving stable operation over a wide temperature range.

[0079] In a crossroads scenario at the Dongfeng Motor Wuhan Demonstration Zone (a scenario including typical conditions such as complex traffic flow, pedestrian crossing, and sudden obstacles), the device of this invention was tested in the field: the test data volume was 1000 multi-source heterogeneous data (including vehicle speed and acceleration data from the CAN bus interface, roadside radar data from the 5G communication module, and traffic event alarm data from the Ethernet interface); the average processing time was 15 milliseconds / 1000 data, which is 75% faster than the traditional solution (60 milliseconds / 1000 data); the deviation rate between the fusion result and the actual traffic state was 3.2% (less than 5%), successfully supporting the reliable operation of the "crossroads collision warning" function, with a warning response time of no more than 30 milliseconds, which is 62.5% faster than the traditional solution (warning response time 80 milliseconds).

[0080] In high-temperature scenarios (summer in desert regions, ambient temperature 60℃), the device operated continuously for 72 hours without failure, with a data transmission attenuation rate of 1.1% and a fusion accuracy of 95.3%. In low-temperature scenarios (winter in frigid regions, ambient temperature -30℃), the device maintained a 96% sealing performance, a data error rate of 10 to the power of -8, and a fusion accuracy of 95.1%, fully verifying its wide temperature range adaptability.

[0081] Based on the same inventive concept, this invention provides a method for processing multi-source heterogeneous data of vehicles. As shown in Figure 2, the method includes the following steps S201 to S205: Step S201: Control the time-division multiplexing bus to schedule the first, second, and third interfaces of the interface adaptation module to receive data in turn according to a preset scheduling cycle, so that the first interface receives vehicle status data of the target vehicle, the second interface receives roadside perception data, and the third interface receives traffic event data; Step S202: Construct a three-dimensional feature space based on the vehicle status data, roadside perception data, and traffic event data to obtain corresponding vehicle-mounted data evidence, roadside data evidence, and event data evidence; Step S203: Generate basic probability distributions for the target driving problem based on the vehicle-mounted data evidence, roadside data evidence, and event data evidence, respectively. Each basic probability distribution includes the support degree of each decision objective for the target driving problem; Step S204: Fuse the basic probability distributions to obtain a fused basic probability distribution, which includes the new support degree of each decision objective for the target driving problem; Step S205: Control the target vehicle according to the fused basic probability distribution.

[0082] The multi-source heterogeneous data processing method provided in this embodiment of the invention is applied to a central processing unit. More implementation details can be found in the aforementioned multi-source heterogeneous data processing device, which will not be repeated here for the sake of brevity.

[0083] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable code.

[0084] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer instructions. These computer instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0085] These computer instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0086] These computer instructions may also be loaded onto a computer or other programmable execution data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0087] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0088] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-source heterogeneous data processing device for vehicles, characterized in that, Applied to a target vehicle, the multi-source heterogeneous data processing device includes: an interface adaptation module for receiving heterogeneous raw data from multiple data sources; the interface adaptation module integrating a first interface, a second interface, and a third interface; the heterogeneous raw data from the multiple data sources including vehicle status data of the target vehicle received through the first interface, roadside perception data received through the second interface, and traffic event data received through the third interface; a central processing unit (CPU) and a time-division multiplexing bus; the CPU being connected to the interface adaptation module via the time-division multiplexing bus; the CPU controlling the time-division multiplexing bus to receive data according to a preset scheduling cycle, so that the first interface, the second interface, and the third interface take turns receiving the vehicle status data, roadside perception data, and traffic event data. The central processing unit (CPU) is configured to: construct a three-dimensional feature space based on the vehicle state data, roadside perception data, and traffic event data to obtain corresponding vehicle-mounted data evidence, roadside data evidence, and event data evidence; generate basic probability distributions for the vehicle-mounted data evidence, roadside data evidence, and event data evidence for a target driving problem, each basic probability distribution including the support degree for each decision objective of the target driving problem; fuse the various basic probability distributions to obtain a fused basic probability distribution, the fused basic probability distribution including the new support degree for each decision objective of the target driving problem; and control the target vehicle based on the fused basic probability distribution.

2. The multi-source heterogeneous data processing device for vehicles as described in claim 1, characterized in that, The central processing unit is used to: preprocess the vehicle status data from the vehicle side to obtain a first standard dataset; The roadside sensing data from the roadside is preprocessed to obtain a second standard dataset; The traffic event data from the traffic event detector is preprocessed to obtain a third standard dataset; a three-dimensional feature space is constructed on the first standard dataset to obtain the vehicle-mounted data evidence; a three-dimensional feature space is constructed on the second standard dataset to obtain the roadside data evidence; and a three-dimensional feature space is constructed on the third standard dataset to obtain the event data evidence.

3. The multi-source heterogeneous data processing device for vehicles as described in claim 2, characterized in that, The first interface is a CAN bus interface, used to receive various vehicle status signals, parse each vehicle status signal according to a structured format and store it to obtain the vehicle status data; the second interface is a mobile communication module, used to parse the received mobile communication signals to extract various roadside sensing signals, store each roadside sensing signal by device identifier to obtain the roadside sensing data. The third interface is an Ethernet interface, used to receive traffic event alarms and cache the received traffic event alarms to obtain the traffic event data.

4. The multi-source heterogeneous data processing device for vehicles as described in claim 2, characterized in that, The central processing unit is used to: remove abnormal data frames from the vehicle status data using an anomaly detection model built based on the isolated forest algorithm, and fill in the missing values ​​in the vehicle status data using a time-series prediction model to obtain the first standard dataset. The abnormal data frames in the roadside sensing data are removed based on the anomaly detection model, and the missing values ​​in the roadside sensing data are filled using the time-series prediction model to obtain the second standard dataset. The third standard dataset is obtained by removing abnormal data frames from the traffic event data based on the anomaly detection model and filling the missing values ​​in the traffic event data using the time-series prediction model.

5. The multi-source heterogeneous data processing device for vehicles as described in claim 2, characterized in that, The central processing unit is configured to: construct a three-dimensional feature space of time dimension, spatial dimension, and attribute dimension for the first standard dataset to obtain the vehicle-mounted data evidence, wherein the vehicle-mounted data evidence includes time dimension features, spatial dimension features, and attribute dimension features; construct a three-dimensional feature space of time dimension, spatial dimension, and attribute dimension for the second standard dataset to obtain the roadside data evidence, wherein the roadside data evidence includes time dimension features, spatial dimension features, and attribute dimension features; and construct a three-dimensional feature space of time dimension, spatial dimension, and attribute dimension for the third standard dataset to obtain the event data evidence, wherein the event data evidence includes time dimension features, spatial dimension features, and attribute dimension features.

6. The multi-source heterogeneous data processing device for vehicles as described in any one of claims 1-5, characterized in that, The central processing unit is used to: fuse the three basic probability assignments corresponding to the vehicle data evidence, the roadside data evidence, and the event data evidence to obtain the fused basic probability assignments; Based on the three basic probability assignments, a conflict coefficient is determined; if the conflict coefficient is less than a preset coefficient threshold, the fused basic probability assignment is obtained; if the conflict coefficient is greater than or equal to the preset coefficient threshold, the three basic probability assignments are corrected and then re-fused to obtain the fused basic probability assignment.

7. The multi-source heterogeneous data processing device for vehicles as described in claim 6, characterized in that, The central processing unit is configured to: after obtaining the fused basic probability allocation, obtain a historical basic probability allocation from a historical database as a reference basic probability allocation based on the fused basic probability allocation and traffic state data representing the current traffic state, wherein the historical database stores historical basic probability allocations corresponding to different historical time periods, different spatial grids, and different traffic states; determine the deviation rate between the fused basic probability allocation and the reference basic probability allocation; and if the deviation rate is less than a preset deviation threshold, control the target vehicle based on the fused basic probability allocation. If the deviation rate is greater than or equal to the preset deviation threshold, the three basic probability assignments are corrected and then re-fused.

8. The multi-source heterogeneous data processing device for vehicles as described in claim 3, characterized in that, The central processing unit is configured to: schedule the CAN bus to receive data in a first time slot within the scheduling period, schedule the mobile communication module to receive data in a second time slot within the scheduling period, and schedule the Ethernet interface to receive data in a third time slot within the scheduling period, wherein the first time slot, the second time slot, and the third time slot are the division results of the scheduling period; the interface adaptation module further includes a hardware timer, which is used to control the switching of the first time slot, the second time slot, and the third time slot; the multi-source heterogeneous data processing device further includes a timing module based on the Global Navigation Satellite System, which is used to add a timestamp to each data frame of the received vehicle status data, roadside perception data, and traffic event data.

9. The multi-source heterogeneous data processing device for vehicles as described in claim 8, characterized in that, The central processing unit is also used to: convert the received vehicle status data, roadside perception data and traffic event data into serialized data of the same format based on a predefined standardized data structure and store them. Each piece of serialized data includes a unique data identifier, timestamp, data source type, device identifier, core data fields and data precision.

10. A method for processing multi-source heterogeneous data, characterized in that, include: The control time-division multiplexing bus schedules the first, second, and third interfaces of the interface adaptation module to receive data in turn according to a preset scheduling cycle. This allows the first interface to receive vehicle status data of the target vehicle, the second interface to receive roadside perception data, and the third interface to receive traffic event data. A three-dimensional feature space is constructed based on the vehicle status data, roadside perception data, and traffic event data to obtain corresponding vehicle-mounted data evidence, roadside data evidence, and event data evidence. Basic probability distributions for the target driving problem are generated for each of the vehicle-mounted data evidence, roadside data evidence, and event data evidence. Each basic probability distribution includes the support degree for each decision objective of the target driving problem. The basic probability distributions are then fused to obtain a fused basic probability distribution, which includes new support degrees for each decision objective of the target driving problem. The target vehicle is then controlled based on the fused basic probability distribution.

Images