Method for selecting an object using an evaluation unit of a node

Abstract

The invention relates to a method (100) for selecting an object by means of an evaluation unit (12) of a node (10), wherein the method (100) comprises: - Receiving (101), by the evaluation unit (12), object data of objects, wherein the object data for each object includes a state vector and covariance information, - Determining (102), by the evaluation unit (12), an object perception quality score for each object based on the covariance information, - Comparing (103) the determined object perception quality value for each object with an object perception quality threshold according to a threshold requirement for selecting relevant objects, - Checking (104) a fulfillment of the limit requirement based on the comparison (103), wherein the limit requirement defines the condition that must be met for the object perception quality value to be satisfied, - Providing (105) a selection list of objects depending on a positive result of the check (104).

Description

[0001] The invention relates to a method for selecting an object using an evaluation unit of a node. Furthermore, the invention relates to a computer program, a data processing device, a platform, a computer-readable storage medium, and a message for this purpose. State of the art

[0002] Currently, traffic automation and vehicle communication are considered promising technologies for mitigating the negative effects of increasing traffic density. Various communication links are being developed, such as vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), vehicle-to-vulnerable road user (V2VRU), and vehicle-to-network (V2N).

[0003] Collective perception aims to exchange information about the current driving environment between ITS stations within ITS subsystems. This includes defining the syntax and semantics of the collective perception message (CPM) as well as specifying the data and message processing in detail to enhance environmental awareness collaboratively. Vehicles and infrastructure units exchange sensor-detected objects with neighboring vehicles.

[0004] It is known from the prior art that collective perception can play a crucial role in the exchange of information between nodes or ITS stations (Intelligent Transportation System Sensors, ITS-Ss). Shared messages within this collective perception include, for example, sets of perceived objects and regions, as well as their observed status and attributes. The content of these messages can vary depending on the type of information or object data and the detection capabilities of the sending nodes (ITS-S). The status information for perceived objects in such a message can include at least the measurement time and position, and optionally further elements of the kinematic and attitude state. To facilitate the interpretation of the received message by any receiving node or ITS station, the sender can also include information about its sensors, such as sensor types and fields of view.Upon receiving a message, receiving nodes can be informed about the presence, type, and status of detected information, objects, or regions identified by the sending nodes. This awareness enables receiving nodes to support so-called ITS applications, which enhance safety and improve traffic efficiency through communication, thereby reducing travel time. The distributed information from these messages can be beneficial for numerous applications, making the concept of collective awareness or information sharing a crucial component of any implementation.

[0005] Detected objects can be included in the CPM according to inclusion criteria such as the object's dynamic state (speed, acceleration), the confidence level of the object measurement, the object type, or the V2X network's current knowledge of the object. Based on specific standardized object inclusion rules, only a selection of the detected objects is included in the collective perception message. The generation rules of the current ETSI standard filter the detected objects based on their object perception quality to avoid transmitting excessively noisy measurements.

[0006] Existing systems, however, rely on complex rules for determining the quality of perceived objects. This can be inefficient and difficult to maintain. Therefore, the object of the invention is to provide a simplified process for selecting information for efficient transmission in a network. In particular, one objective is to provide an improved filter for object perception quality based on covariance information from tracking algorithms. Disclosure of the invention

[0007] According to aspects of the invention, a method with the features of claim 1, an evaluation unit with the features of claim 10, a data processing device with the features of claim 11, a platform with the features of claim 12, a computer program with the features of claim 13, a computer-readable storage medium with the features of claim 14, and a message with the features of claim 15 are provided. Further features and embodiments of the invention are disclosed in the respective dependent claims, the description, and the drawings. Features and embodiments described in connection with the method according to the invention also apply accordingly to the computer program, the evaluation unit, the data processing device, the platform, the storage medium, and the message according to the invention, and vice versa.

[0008] According to one aspect of the invention, a method for selecting an object uses an evaluation unit of a node, wherein the method comprises: - Receiving, by the evaluation unit, object data of objects, wherein the object data for each object includes a state vector and covariance information, - Determine, by the evaluation unit, an object perception quality score for each object based on the covariance information, - Comparing the specified object perception quality value for each object with an object perception quality threshold according to a threshold requirement for selecting relevant objects, - Verifying compliance with the limit value requirement based on comparison, where the limit value requirement defines the condition that the object perception quality value must meet, - Providing a selection list of objects depending on a positive result of the check.

[0009] It is possible to select relevant objects based on their perceived quality. The method according to the invention preferably uses weighted covariance information, in particular a (weighted or unweighted) covariance matrix derived from the object data, to calculate an object perception quality (OPQ) score for each object. This OPQ score reflects the accuracy of the measurements associated with the object's state vector. Objects with higher OPQ scores indicate greater measurement accuracy or precision and are therefore more reliable. By comparing the calculated OPQ scores with predefined threshold values, those objects that meet or exceed the threshold requirement are included in a selection list. The selection list preferably contains the object data, i.e., the state vectors and the (weighted or unweighted) covariance information of the objects to be selected.

[0010] This filtering process ensures that only objects with sufficient perceived quality are selected for further processing or decision-making. Furthermore, this approach increases the reliability of object perception and improves the accuracy of downstream tasks such as path planning or collision avoidance. By implementing a covariance-based approach to filter out less precise object data based on its covariance matrix, the selection process is also simplified while maintaining accuracy and reliability. Consequently, the method according to the invention advantageously reduces the load on the communication channel, significantly improves efficiency, and ensures that only essential data for decision-making is transmitted to other nodes, vehicles, or road users.

[0011] The minimum information required for an object within a Corporate Perception Service (CPS), wireless sensor network, or similar system consists of its position and velocity relative to the sending node or ITS station. Accordingly, the minimum state space can be defined as a state vector x comprising components or elements of the state vector, or scalars for position and velocity, specifically d x d y v x v y Therefore, an object perception quality can be specified at least for such a state space, with possible extensions for, for example, acceleration, angle, or the like.

[0012] The covariance information can comprise a covariance matrix or at least one component (matrix element) of a covariance matrix with respect to the state vector for each of the objects. In other words, for each object, either the entire covariance matrix or at least one matrix component of the covariance matrix can be included in the covariance information. More precisely, the covariance matrix comprises matrix elements that correspond to the correlation of identical or different elements of the state vector. The covariance information can include at least one column and / or at least one row and / or the diagonal of the covariance matrix.

[0013] Alternatively, the covariance matrix or components of the covariance matrix can be determined or calculated based on the received covariance information. For example, the covariance information can include a variance represented by a confidence value for a defined confidence interval (e.g., 95%) with respect to the elements of the state vector. Additionally, the covariance information can also include Pearson correlation coefficients representing covariances between different elements of the state vector.

[0014] The covariance matrix describes the accuracy or precision of the current measurement. In other words, the covariance matrix determines the shape of the ellipsoid, while the position or center of gravity of the ellipsoid is described by the underlying mean of the distribution. The principal axes of the ellipsoid correspond to the normalized eigenvectors of the covariance matrix, and the lengths of the axes are determined by the associated eigenvalues.

[0015] Furthermore, the covariance matrix can provide a detailed understanding of how the object data and / or its variables or components change together. The diagonal elements of the covariance matrix represent the variances of the individual variables. Variance is a measure of the dispersion of a variable around its mean. The elements in the positions off-diagonal of the matrix represent the covariances between any two different variables. Covariance indicates how strongly two variables change together. If the covariance is positive, the variables tend to increase or decrease simultaneously; if it is negative, one variable increases while the other decreases.

[0016] The procedure may also include at least one of the following steps: - Providing the specific object perception quality value for further processing at the node, - Transferred, by the evaluation unit, the selection list of objects to another unit of the node for further processing of the selected object data.

[0017] This has the advantage that the specific object perception quality value can be used by other modules within the system for tasks such as planning, decision-making, or control. Transferring the selection list or the selected objects to another unit enables further processing of the selected objects, which can potentially lead to measures such as optimizing the transfer of the selected objects.

[0018] According to one embodiment, further processing includes prioritizing the objects contained in the selection list by the evaluation unit and / or the further unit of the node. Preferably, the method further includes: - Initiating a transmission of a message, preferably via a wireless communication channel, to another node of a distributed system that includes the node, based on the provided selection list of objects, in particular based on a prioritization result, wherein the message includes at least some of the object data.

[0019] The additional radio node can be formed by a platform, such as a vehicle or an infrastructure unit. Based on the initiated transmission, a driver information or warning system and / or a driver assistance system and / or an automated driving system of the vehicle can be controlled.

[0020] It is also possible that the condition for verifying the limit requirement specifies that the determined object perception quality value must be equal to, or greater than, or equal to or greater than, or less than, or equal to or less than the (predefined) object perception quality limit.

[0021] Defining specific criteria enables the selection of objects based on their OPQ values. The threshold requirement can define whether the OPQ value should be equal to, greater than, or less than a predefined threshold. This allows for finer control in object selection and enables the system to prioritize objects with specific OPQ properties. For example, objects requiring high precision can be selected by setting a higher threshold, while objects requiring lower accuracy can be selected by setting a lower threshold.

[0022] It is possible that the procedure may also include: - Determining, by the evaluation unit, weighted covariance information based on the received covariance information, - wherein the weighted covariance information preferably comprises a weighted covariance matrix or at least one weighted component of a covariance matrix, and - where the object perception quality score is determined based on the weighted covariance information.

[0023] The weighted covariance matrix, or at least one weighted component of the covariance matrix, can be encompassed by the received covariance information or determined based on the received covariance information.

[0024] It is possible that a weight of each component of the covariance matrix is ​​designed to differ from at least one other weight of a component, or that the weights of all components are equal or identical, i.e., defined in the same way.

[0025] It is possible to adapt the weighting scheme for the components of the covariance matrix to specific applications in order to emphasize certain aspects of object perception accuracy. Assigning different weights allows for a more nuanced assessment of the object's state and uncertainty. For example, in an application where positional accuracy is particularly important, the positional variance components can be assigned a higher weight than the variances of velocity or orientation. Conversely, in a prioritized dynamics assessment, velocity variances can be weighted more heavily. This adaptable weighting scheme improves the adaptability of the OPQ calculation to specific application requirements and increases the overall accuracy and relevance of object perception.

[0026] It is also possible that the covariance information preferably comprises a covariance matrix; that the comparison of the specific object perception quality score is based on comparing each component of the covariance matrix and / or each weighted component of the covariance matrix with corresponding components of an object perception quality threshold matrix; or that the comparison of the specific object perception quality score for each of the objects with an object perception quality threshold is based on an entropy-based model. This allows for a flexible and adaptable OPQ assessment based on different properties and requirements of the perceived objects.

[0027] It is possible to base the comparison of the object perception quality score on the components of the covariance matrix and / or the weighted components of the weighted covariance matrix for each of the objects, with each of the one or more components and / or each of the one or more weighted components being compared to the object perception quality threshold. Optionally, it is also possible to compare a subset of the components to the object perception quality threshold.

[0028] This allows for a more nuanced assessment of object perception quality by independently considering the accuracy of specific parameters such as position, velocity, or acceleration. By focusing on relevant components, the system can better prioritize objects based on the accuracy of the information available for each parameter.

[0029] Furthermore, it is possible that the comparison of the object perception quality value is based on an inverted covariance matrix for each of the objects, with each of the one or more components being compared to the object perception quality limit.

[0030] It is also possible that the comparison of the specific object perception quality value is based on comparing a scalar value derived from the covariance information and / or the weighted covariance information with the object perception quality limit.

[0031] The scalar value can encapsulate the overall accuracy of the representation of an object's state vector. Comparing this scalar value with the object perception quality threshold advantageously simplifies the OPQ evaluation process. This approach offers computational efficiency advantages compared to methods that require comparisons of multiple variances or determinants. Using a single object perception quality threshold for scalar-based OPQ evaluation advantageously streamlines the decision-making process within the OPQ filter.

[0032] Furthermore, it is possible that the node includes a sensor system for perceiving the environment and / or a communication system for receiving a message with object data from at least one other node and / or a V2X system as a sensor, wherein the received object data includes object data of objects that are perceived by the sensor system in the environment and / or by the communication system and / or by the V2X system as a sensor.

[0033] It is possible for the node to be equipped with multiple sensing modalities. The node can integrate a sensor system to directly perceive its environment and collect data about objects within it. Additionally, the node can include a communication system to receive object data from other nodes in the network. This enables collaborative or collective perception, where information from different nodes is combined to create a more comprehensive understanding of the environment. Furthermore, the node can utilize a V2X system as a sensor. V2X technology allows vehicles to communicate with each other, with infrastructure, and with pedestrians, providing additional data about objects and their movements. By integrating these different sensing modalities, the node can gather extensive and diverse information about its surroundings.

[0034] It is possible for the method according to the invention to be used in a vehicle and / or in an infrastructure unit. The vehicle can, for example, be a motor vehicle and / or passenger car and / or at least partially automated / autonomous vehicle. The vehicle can be equipped with vehicle equipment, e.g., for providing an automated driving function and / or a driver assistance system. The vehicle equipment can be designed to control the vehicle at least partially automatically and / or accelerate and / or brake and / or steer. The infrastructure unit can, for example, be a trackside station and / or a traffic light and / or a traffic sign.

[0035] At least one control action for the vehicle can be initiated and / or carried out based on the method according to the invention. The control action can include at least one of the following: braking, steering, accelerating, overtaking maneuver, emergency braking, activation of an alarm system, activation of hazard warning lights, activation of a turn signal, light control, or the like.

[0036] In another aspect of the invention, a computer program, in particular a computer program product, can be provided which includes instructions which, when the computer program is executed by at least one computer and / or evaluation unit and / or data processing device and / or platform, cause the computer and / or evaluation unit and / or data processing device and / or platform to carry out the method according to the invention. The computer program according to the invention can thus have the same advantages that have already been described with regard to the method according to the invention.

[0037] Another aspect of the invention is an evaluation unit for filtering object data, which includes means for carrying out the method according to the invention. The evaluation unit according to the invention thus has the same advantages that have already been described in detail with regard to the method according to the invention.

[0038] In another aspect of the invention, a data processing device can be provided that is configured to execute the method according to the invention. For example, the device can include at least one computer that executes the computer program according to the invention. The computer can comprise at least one processor that can be used to execute the computer program. Furthermore, a non-volatile data storage device can be provided in which the computer program is stored and from which the computer program can be read by the processor for execution.

[0039] In another aspect of the invention, a platform can be provided that is configured as an infrastructure unit, e.g., as a roadside unit (RSU), or as a vehicle. The platform can therefore be stationary or mobile. The vehicle can be a ground vehicle or an aircraft and can be unmanned or manned.

[0040] According to another aspect of the invention, a computer-readable storage medium can be provided which comprises the computer program according to the invention and / or instructions which, when executed by at least one computer and / or evaluation unit and / or data processing device and / or platform, cause the computer and / or evaluation unit and / or data processing device and / or platform to perform the steps of the method according to the invention. The storage medium can be configured as a data storage device such as a hard disk and / or non-volatile data storage and / or memory card and / or solid-state drive. The storage medium can, for example, be integrated into the computer.

[0041] Another aspect of the invention is a message comprising a V2X message, wherein the V2X message contains object data of objects. The object data is processed according to the steps of the method according to the invention. The message according to the invention thus has the same advantages that have already been described in detail with regard to the method according to the invention. Furthermore, this enables improved communication between nodes, vehicles, and / or infrastructure units by providing more accurate and reliable information or object data.

[0042] Furthermore, the method according to the invention can be implemented as a computer-implemented method. Alternatively or additionally, at least one of the disclosed method steps can be computer-implemented and / or automated.

[0043] It is possible for the method according to the invention to be used in a vehicle and / or in an infrastructure unit. The vehicle can, for example, be a motor vehicle and / or passenger car and / or at least partially automated or autonomous vehicle. The vehicle can be equipped with vehicle equipment, for example, for providing an autonomous driving function and / or a driver assistance system. The vehicle equipment can be designed to control, accelerate, brake, and / or steer the vehicle at least partially automatically. The infrastructure unit can, for example, be a trackside station, a traffic light, or a traffic sign.

[0044] Further advantages, features, and details of the invention will become apparent from the following description, in which embodiments of the invention are illustrated in detail with reference to the drawings. In this context, the features mentioned in the claims and in the description can be essential to the invention, either individually or in any combination. The following are shown: Fig. 1: a method, a computer program, a storage medium and a device according to embodiments of the invention, Fig. 2: a schematic diagram of a functional process flow according to embodiments of the invention, Fig. 3: a further schematic diagram for generating messages according to embodiments of the invention, Fig. 4: a schematic overview according to embodiments of the invention, and Fig. 5: A schematic view of a scenario according to embodiments of the invention.

[0045] In the following figures, the same reference numerals are used for the same technical features, even in different embodiments.

[0046] The invention relates to a method for improving object selection in a Collaborative Perception Service (CPS) based on covariance-based filtering of objects with inaccurate measurements, as well as an overall simplified method for message generation. The invention addresses the complexity of existing rules for selecting detected objects, which leads to inefficient resource utilization and redundant information transmission. By implementing a covariance-based approach for filtering out less accurate object data based on its covariance matrix, the selection process is simplified while maintaining accuracy and reliability. The method according to the invention enables improved object perception quality, which advantageously enhances the situational awareness of other nodes, vehicles, or infrastructure systems. In the case of vehicles, this facilitates safer and more efficient traffic flow.

[0047] Fig. Figure 1 shows a method 100 for selecting an object using an evaluation unit 12 of a node 10. The method 100 comprises a step 101. In step 101, the evaluation unit 12 receives object data from objects. The object data includes a state vector and covariance information, preferably a covariance matrix or components thereof, for example, specific variances or a subset thereof, for each of the objects. In step 102, the evaluation unit 12 determines an object perception quality value for each object based on the covariance information. In step 103, the determined object perception quality value for each object is compared with an object perception quality threshold according to a threshold requirement for selecting relevant objects. Then, in step 104, it is checked, based on the comparison in step 103, whether the threshold requirement is met.The threshold requirement defines the condition that the object perception quality value must meet. In step 105, depending on a positive result of the check in step 104, a selection list of objects to be selected is provided.

[0048] Furthermore, it shows Fig. 1 a node 10 or a device 10 for data processing, comprising a computer-readable storage medium 15. The medium 15 contains a computer program 50. Furthermore, it shows Fig. 1. A message 600 comprising a V2X message 610. The V2X message 610 contains object data of objects 40. The message 600 can further comprise at least one additional message type, which is processed by the method 100 according to the invention.

[0049] Fig. Figure 2 shows a schematic diagram of a functional process flow according to embodiments of the invention. In particular, it represents Fig. Figure 2 presents a schematic diagram illustrating the processing of available information. Available information can include data on detected or perceived objects, or information such as temperature, humidity, position, speed, acceleration, or other measured or inferred data.

[0050] Furthermore, it shows Fig. 2 two blocks 12, 20, where the first block 12 can represent an evaluation unit 12 and the second block can represent an evaluation function 20 or another evaluation unit 20. The evaluation unit 12 can be designed as a pre-filter to pre-select or choose information, for example, object data for an object. The evaluation function 20 can further process available information based on a relative information gain, also referred to as information value, and decide, based on the available resources 203, to transmit or distribute the relevant information 20, 21 205.

[0051] The evaluation unit 12 can, for example, be designed to select an object that has been detected or perceived by a sensor system of node 10. Therefore, the evaluation unit 12 can receive such information or data 201 in order to process it further based on the available resources 203 or to provide it in relation to the generated message(s) 204.

[0052] In one embodiment, the pre-selected information or data can be further processed, for example, by another traffic relevance function, an evaluation function 20, or the like. In another embodiment, the information can be forwarded to another node or another function of a distributed system.

[0053] In another embodiment, the evaluation unit 12 can receive information or data about a message from one or more nodes 201. The one or more nodes can be elements or components of a distributed system.

[0054] Furthermore, it shows Fig. 2 an evaluation function 20 that prioritizes the preselected 202 information or object data. The (available) information or data or object data 202 can be selected for transmission 205 based on an estimated information value or information gain that they offer to the receiving sink nodes. An information gain of available information, in particular an information value, can be estimated in relation to the evaluated information using an information entropy model.

[0055] Furthermore, the evaluation unit 12, as in Fig. 2 shows how to select information based on the object perception quality.

[0056] For each perceived object, as in Fig. Figure 3 illustrates that a state vector, along with its associated covariance matrix or specific matrix elements, is present. The covariance is a measure of measurement quality. It also encompasses the information age, since tracking filters such as a Kalman filter or a particle filter can lead to decreasing precision over time if only state predictions are made without subsequent updates. In other words, if new measurements are not taken, the information age increases and the state estimates become less accurate because they rely more heavily on model-based transition predictions than on actual measurement updates.

[0057] Therefore, the evaluation unit 12, as in Fig. Figure 2 illustrates how covariance can be used to estimate object perception quality (OPQ) by defining precision or accuracy thresholds for the various parameters or components of the state vector. For example, a variance (or, correspondingly, the estimated absolute accuracy for a 95% confidence interval) of the x and y positions should be greater than or equal to a position accuracy threshold p. a_thr be, where p a_thr may be tied to a minimum accuracy that receivers need to use the data (e.g., track accuracy × constant).

[0058] Alternatively, a scalar value can be defined, e.g., based on the information entropy of the data. This scalar is then compared by the evaluation unit 12 with a single threshold value, instead of comparing each component of the state vector with a threshold value.

[0059] The pre-filtering and selection process by the evaluation unit 12 or the OPQ filter 12 can be followed by further filters such as a traffic relevance filter or an information value function 20, in functional units, as in Fig. 2 shown, using the rules according to the invention.

[0060] The function for object perception quality (OPQ) should have the form object perception quality = f(χn,Σnxn), where ∂f(χn,∑nxn)∂|∑nxn|≤0 This means that its value increases with increasing noise in χ. n decreases or remains constant, or any combination of the components of χ n Σnxn represents the corresponding covariance matrix of χ. nor their available components. Depending on the implementation, the function can return either a scalar or another vector (usually with n or fewer components) or a matrix (also typically with reduced dimensionality).

[0061] Furthermore, the OPQ is compared with threshold values ​​within the OPQ filter. These threshold values ​​can be different for each component, for example, 2 m (2 meters) for x and y, and 0.3 m / s for v. x , v y or 4 m (4 meters) for e.g.

[0062] In one embodiment, a weighted covariance matrix can be used to calculate f: ∑weightedn×n=Wn×n∑n×nWn×nT

[0063] With W = E (where E is the identity matrix) all components would be equally weighted, resulting in the original, unweighted covariance matrix.

[0064] In other embodiments, it is optionally possible to use a precision-based, a determinant-based, or an entropy-based approach: 1. Precision-based: Λweighted=∑weighted−1 (positive definite) 2. Determinant-based: Σ gewichtet (Scalar ∈ [0,∞], the lower the number, the higher the OPQ) 3. Entropy-based: Hweighted=12log2((2πe)n∑weighted) (Scalar ∈[0,∞], lower for higher OPQ)

[0065] In further embodiments, the above-mentioned threshold comparison can be carried out, for example, for the precision-based, the determinant-based or the entropy-based approach as follows: 1. Precision-based: diag (Λ gewichtet) as a component-wise comparison with an n-dimensional threshold vector, where rules may, for example, stipulate that each component must be greater than the respective threshold. Since the components are already weighted, this can also be expressed without loss as diag(Λ gewichtet ) ≥ threshold value for all n components. Another embodiment may be that at least m of n components must be greater than or equal to the threshold value; 2. Determinant-based: |Σ gewichtet | ≤ Threshold 3. Entropy-based: H gewichtet ≤ Threshold

[0066] Only objects that meet the above conditions can pass the filter. The filter can also be defined in reverse, so that objects that do not pass the filter are filtered out.

[0067] Fig. Figure 3 shows another schematic diagram for message generation according to embodiments of the invention. Different generation rules can exist to create a message 300, as shown in Fig. 3 shown, to build.

[0068] For example, message 300 can be a Collective Perception Message, a Shared Information Message, or something similar. Based on these generation rules, such a message 300 can contain various elements or container types inherent in every message 300. Message 300 can include a header element 304, which comprises a PDU header, a management container, a station data container, and a generation delta time. The PDU header can contain information about the protocol version or message type. The management container can include components for the reference time and reference position. Furthermore, the message can include a sensor information container 305, a free space supplement 306, and a container for perceived objects 301.The sensor information container 305 can contain information about individual sensors or sensor systems that generated the data of detected objects. The kinematic and positional state of an object can be represented in a corresponding state vector, which includes various components relating to the kinematic and positional state. Furthermore, the transmission frequency of message 300 can be set between 1 and 20 Hz, depending on the utilization of the communication channel. The detected objects 301 can be included in message 300 based on various inclusion criteria, such as object quality, object type (e.g., VRU), or the like. Based on these criteria, objects 302 can be excluded due to their object perception quality. The capacity of the communication channel can then limit how many of these objects are ultimately selected for transmission.Objects 303 that are already known to potential recipients are considered "redundant" and excluded. However, the excluded objects 303 can be transmitted via other channels 320, if available.

[0069] If multi-channel operation (MCO) with different communication channels is available, parts of the information or object data can be outsourced to different channels (310, 320). This is described in Fig. This is illustrated by the two messages 300 and 350, where message 350 can contain elements 304 and 303. Element or container 303 contains the outsourced part of the information (320). This message 350 also includes a header element 304. Furthermore, it is possible that information, object data, or objects 303 may be omitted due to redundancy.

[0070] Each communication channel can support a specific message size, which depends, among other things, on communication technologies such as 4G-V2X, ITS-G5, or 5G-V2X. Therefore, message segmentation may be necessary if too much information or object data is selected for transmission.

[0071] In another embodiment (not shown), such as in distributed systems like wireless sensor networks, the exchange of information between different nodes in a network is often dynamic. Source nodes possess certain available information or data that are required by one or more sink nodes. This information or data can consist of numerous subsets, such as temperature, pressure, timestamp, position, velocity, etc., also referred to as the Available Information Vector (AIV), which can be understood as a generalization of a system state vector. Furthermore, the sink nodes have specific information requirements, also referred to as the Required Information Vector (RIV).As a dynamic system, the size of both vectors, AIV and RIV, can vary over time and adapt to the conditions of the environment and the system.

[0072] To exchange available information, the nodes of the distributed system are interconnected via wired or wireless communication links. The bandwidth of these connections can be limited depending on the physical implementation, which often presents a complex optimization problem.

[0073] Conventional systems employ different approaches, such as: periodic transmissions of all available data to the source nodes (full AIVs) but at low frequencies to limit the load on the communication nodes; partial transmissions of AIVs at higher frequencies for more critical components; and request-based transmissions where sink nodes request updates to specific components of their RIVs. While the first two approaches are relatively static and do not consider the information actually needed by the sink nodes, the latter approach can utilize some of the valuable communication resources for transmitting the requests.

[0074] According to embodiments of the invention, information or data or object data can be selected for transmission based on an estimated information value or information gain that it offers to the receiving sink node.

[0075] The process can include specific steps that can be performed at one or more source nodes.

[0076] Available information can be evaluated at one or more sink nodes of the distributed system based on a defined system architecture. The value of receiving specific information or data, which might be part of the AIV (Automated Information System), can depend on the estimated prior knowledge of the respective receiving or sinking node. This knowledge can be estimated in various ways.

[0077] It is possible that the prior knowledge of the receiver or source node is estimated without any external information from other source nodes (independently), i.e., based exclusively on local information.

[0078] Furthermore, it is possible that the sending source node assumes that information it previously transmitted is already known. Depending on how quickly the network topology changes in a given situation, only very recently transmitted data should be assumed to be known.

[0079] It is also possible that the sending source node analyzes the intrinsic entropy of each component of the AIV. For example, when transmitting weather data, prevailing weather conditions can be expected to be already known to the receiving sink nodes.

[0080] If the sending source node also receives information from other source nodes, it can assume that this information has also been received by the sinking nodes and is therefore known to them.

[0081] In another embodiment, a central instance can exist that receives and aggregates information or data from all source nodes. Optionally, the central instance can explicitly request missing information from the network. Furthermore, it can disseminate a display of available information, for example, by using a Bloom filter or by means of an object list in the message, whereby the message can optionally be enriched with additional information, such as object positions.

[0082] Furthermore, it can distribute all aggregated information to the source nodes. As another option, the central instance can forward information based on the estimated utility or the relative information gain for the respective sink nodes. This can be done, for example, by means of an evaluation using relative information entropy.

[0083] In another embodiment, a cluster head of a decentralized, cluster-specific network topology can aggregate information that it receives from its so-called leaf nodes.

[0084] Furthermore, the decision to forward information within the network can also be based on the additional steps described below.

[0085] According to embodiments of the invention, the relative information gain of available information, for example, the information value, can be estimated at one or more source nodes of the distributed system in relation to the evaluated information using an information entropy model. The model defines a relationship between two (probability) distributions of the respective information. After the information available to the sink node has been estimated using one or a combination of the methods described above (e.g., centralized or decentralized), the sink node can estimate the potential information gain of each of its AIV components for the respective sink nodes.

[0086] This can be estimated or calculated, for example, as follows:

[0087] One possibility is to use an approach based on the Kullback-Leibler divergence (D KL ) 2 . DKL is a measure of the relative entropy between two probability distributions and can be used to estimate the information gain that a component of the AIV provides relative to the prior knowledge of the receiving or sinking node. Let N0 be the relevant AIV component of the sending source node and N1 the estimated prior knowledge of the sinking node about this quantity; Σ0 and L1 the covariance matrices of N0 and N1, respectively; µ0 and µ1 the means of N0 and N1, respectively. Then D KL Calculate as follows: DKL(N0‖N1)=12(tr(∑1−1∑0)−k+(μ1−μ0)T∑1−1(μ1−μ0)+ln(det∑1det∑0)) A higher divergence then means a higher information gain and consequently a higher information gain or information value (Vol).

[0088] Another approach can be based on angular similarity (S θThe cosine similarity is a measure of the similarity between two vectors and is calculated as the cosine of the angle between them. For two components of the AlV, A and B, the cosine similarity is calculated as follows: Cosine similarity=SC(A,B):=cos(θ)=A⋅B‖A‖‖B‖=∑i=1nAiBi∑i=1nAi2⋅∑i=1nBi2

[0089] In normalized form, this is called angular distance: Angular distance=Dθ:=arccos(cosine similarity)π=θπ

[0090] The complement of the angular distance is the angular similarity S. θ It lies in the interval [0,1] and is calculated as: Angular similarity=Sθ:=1−angular distance=1−θπ

[0091] Other options may be based on other KLD derivatives, such as the Population Stability Index (PSI). The PSI is a symmetrical version of the KLD and is calculated as: PSI(P,Q)=DKL(P‖Q)+DKL(Q‖P), or based on the Jensen-Shannon divergence (JSD). The JSD is a symmetrical version of the KLD and lies in the region [0,1]. It can be calculated as follows: JSD(P‖Q)=12D(P‖M)+12D(Q‖M), with M=12(P+Q).

[0092] According to embodiments of the invention, resources in a network can be allocated based on the estimated (relative) information gain in order to prioritize information transmission. Once the information gain or information value of each component of the available information vector (AIV) has been estimated, it is necessary to evaluate which components should be transmitted and how frequently. This decision requires a reasonable compromise between providing high-quality information on the one hand and avoiding overloading the transmission channel on the other.To achieve this, one possibility is that, if the communication channel itself and its current utilization are not monitored, one or more (predefined) fixed thresholds regarding information gain or information value are used to decide whether a component of the AIV at one or more source nodes should be included in a newly generated message. The message generation rate can be fixed. The other possibility is based on the experience that a priori fixed transmission patterns rarely offer optimal solutions for the aforementioned trade-off. If, however, the channel is constantly monitored, message generation can be dynamically optimized.The data rate available for the information dissemination or transmission service can be estimated, and this service can then be optimized based on the value of the information transmitted by regulating both the overall frequency of message generation and the quantity and selection of available information or data from which each message may consist.

[0093] Fig. Figure 4 shows a schematic overview according to embodiments of the invention. In particular, it illustrates Fig. 4 the influence of an information value threshold 430 on available information, such as multiple message objects 410, to be selected or prioritized for transmission. As in Fig. As shown in Figure 4, the threshold of 430 is set such that the information or objects 41, 42, 43, with their respective information values ​​31, 32, 33, which exceed the threshold of 430, can be transmitted. Consequently, objects 44, 45, 46 are omitted or not considered for transmission, since their respective information values ​​34, 35, 36 are below the set threshold of 430.

[0094] In another embodiment (not shown), an application example might be a wireless sensor network (WSN) comprising distributed sensor nodes for forest fire detection. A sensor node might include temperature and / or humidity sensors. Such a sensor node might assume that the temperature and humidity values ​​of an information sink are known if they are close to the usual values ​​for a given date and time. Each time the sensor node takes a new measurement, it might assess the value of its new information, for example, by using the angular similarity between the measured values ​​and the expected values. For example:

[0095] An expected measurement result could be, for example, a temperature T = 28 degrees Celsius and humidity H = 46%, which can be further processed as: M=(2846) For the date and time, a temperature T = 30 degrees Celsius and a humidity H = 50% are expected, which can also be formulated as follows: E=(3050) To determine a (relative) information gain, the node can calculate the following: cos(θ)=M⋅E‖M‖‖E‖=0.9999797158186168

[0096] This value close to 1 means that the information is quite similar to the expected value mentioned above and should therefore only be transmitted if sufficient resources are available.

[0097] In the event of an unexpected measurement result, the temperature could be, for example, T = 71 degrees Celsius and the humidity H = 5%, which is used as follows: M=(715)

[0098] For date and time, a temperature T = 30 degrees Celsius and a humidity H = 50% are expected, which is used as follows: E=(3050)

[0099] To determine the (relative) information gain, the node can calculate: cos(θ)=M⋅E‖M‖‖E‖=0.5734623443633283 and Dθ=arccos(0.5734623443633283)π=θπ=0.3055998877857852Sθ=1−Dθ=0.6944001122142148

[0100] The similarity score used in this example is much lower, which means a higher information gain for the sink.

[0101] Therefore, this measurement should be preferred if a choice must be made between the two measurements as to which one should be transmitted.

[0102] Fig. Figure 5 shows a schematic representation of a scenario according to embodiments of the invention. In particular, a traffic scenario is depicted in which a first vehicle 61 and a second vehicle 62 detect a cyclist 70 or a bicycle 70. The cyclist can also be referred to as an endangered road user 70. The third vehicle 60, which wants to turn onto the main road on which the first vehicle 61 is traveling, can be seen in this Fig. In the scenario depicted, the cyclist 70 cannot be detected. This is because his view is obstructed by a building 80. The third vehicle 60 can receive shared messages from the other two vehicles 61 and 62. These messages can be so-called Vehicle-to-Everything (V2X) messages. The third vehicle 60 can benefit from these messages transmitted by the other two vehicles 61 and 62. Due to their different perspectives, the second vehicle 62 can measure different information or data, such as the characteristics of the cyclist 70, with varying degrees of accuracy. For example, the first vehicle 61 can accurately measure the width and speed of the cyclist 70, but not his length, and vice versa for the second vehicle 62. The first vehicle 61 can transmit its message first, and the second vehicle 62 can receive it.Furthermore, the second vehicle 62 can assume that the third vehicle 60 can also receive the same message from the first vehicle 61. In this embodiment of . Fig. Vehicles 60, 61, and 62 are members of a distributed system for exchanging information about the vehicles in the area. Fig. The intersection shown in Figure 2. Such a system can also be called a Vehicle Ad Hoc Network (VANET). This network is set up for a specific period of time at the "intersection" location, i.e., as long as vehicles 60, 61, 62 and the cyclist are near this intersection.

[0103] If the second vehicle 62 can then generate a new message itself, it can estimate an information gain, for example, the value of the information it can provide, based on a Kullback-Leibler divergence. The information value of a transmitted object can be determined by the relative entropy between the object's state known to VANET and the measured state. Since fusion algorithms like the Kalman filter typically assume a normal distribution of the data, the relative entropy can be calculated from the respective example distributions (alternatively, Poisson, Rice, or other distributions can be used): NLEM(r→LEM,∑LEM)=(2π)−k2|∑LEM|−12e−12(r→−r→LEM)T∑LEM−1(r→−r→LEM) NVANET(r→VANET,∑VANET)=(2π)−k2|∑VANET|−12e−12(r→−r→VANET)T∑VANET−1(r→−r→VANET) where r→LEM,r→VANET∈ℝk and Σ VANETthe corresponding covariance matrix of the information (or a subset thereof) as it is known to VANET.

[0104] LEM is to be understood as a local environmental model that represents the vehicle's surroundings. GEM is to be understood as a global environmental model that represents the overall situation, for example, a traffic scenario or a forest fire scenario.

[0105] Since it is generally unknown, it can be calculated, for example, as follows: ∑VANET≈∑V2X=∑LEM∑GEM(∑LEM−∑GEM)−1

[0106] The relative entropy can now be derived from the Kullback-Leibler divergence: DKL(NLEM‖NVANET)=∫ℝkNLEM logNLEMNVANETdr→≥0 By combining these equations and performing some algebraic calculations, one obtains the computationally less complex representation: DKL=12{tr(∑VANET−1∑LEM)+(r→VANET−r→LEM)T∑VANET−1(r→VANET−r→LEM)−k+ln|∑VANET||∑LEM|}≥0

[0107] Relative entropy is measured in Napier digits. Division by In 2 thus yields the divergence in bits. Therefore, D KL / ln 2 is interpreted as the expected number of additional bits that must be transmitted by the node or vehicle to provide the network with knowledge about N. LEM given N VANET to complete.

[0108] According to Gibbs' inequality, relative entropy is always non-negative D. KL ≥ 0, where equality only holds if N LEM = N VANET almost everywhere. If, instead, unfused data is to be transmitted, the relative entropy can be preserved by N LEM by N meas replaced and D KL (N meas ||N VANET ) is calculated.

[0109] In some cases, a next step may involve mapping to a different space. This can preferably be done by normalizing the information gain to the range [0, 1].

[0110] This could be achieved, for example, by applying f(DKL)=1−e−DKL to be achieved, whereby D KL : [0, ∞] → [0,1] is mapped. However, using DKL / ln (2) is an equally valid option (though then it is not normalized to [0, 1]).

[0111] For each possibility (taking into account the fused or unfused data available at the transmitting stations), an analysis of the potential information gain expected for VANET can now be calculated. Various heuristics can be used for this purpose. A brute-force solution that yields an optimal result could, for example, examine all permissible combinations of data elements that can be contained in an object.

[0112] Finally, given a specific amount of channel resources, the selection of objects and their level of detail can be modeled as a variant of the knapsack problem. Corresponding solution approaches could then be applied.

[0113] In another embodiment, the information transmitted or distributed jointly, for example across different V2X message types, can include trajectory data. These trajectories can be described at varying levels of detail. The trajectories can be future-oriented (for maneuver planning and coordination) or past-oriented (for better correlation of measurements or traces, e.g., to ensure that object A is still the same object A that was detected a few seconds earlier). Currently, trajectories are typically sampled uniformly or generated according to simple rules, such as intervals.

[0114] In another embodiment, for example, a vehicle X is driving straight ahead at a high, constant speed, while another vehicle Y is about to initiate a critical left turn.

[0115] In this embodiment, the position of vehicle X is subject to significantly greater changes. With a scanning threshold of, for example, 4 m, as is common in CAM, this would lead to a comparatively large number of trajectory points that would have to be transferred.

[0116] For vehicle Y, the positions change significantly less, so considerably fewer trajectory points need to be included (even with further sampling thresholds, such as a rotation change of 0.5°). However, since the receivers can easily predict the constant speed of vehicle X, much of this information is not really valuable.

[0117] On the other hand, the left turn maneuver by vehicle Y might come as a surprise. However, considering the information gained leads to a significantly more efficient allocation of resources. The greater the deviations from expected states, the more resources are allocated (in this case, more sampled trajectory points). For example, if vehicle X suddenly begins to brake hard, more resources would immediately be made available for its monitoring.

[0118] For another vehicle Z, which is moving within a roundabout, fewer resources are required, since its position, orientation, speed, etc., can be easily estimated based on the last transmitted trajectory points. Once Z leaves the roundabout, the value of the available information increases, so this information is transmitted as soon as possible.

[0119] In another embodiment, it can be assumed that, for example, two observations are present at a vehicle. Firstly, an object-related observation of the environment, which can already be communicated via V2X by other nodes or vehicles, or infrastructure elements, such as a trackside substation (RSU), and secondly, a self-observation or so-called ego-perception.

[0120] These two observations can capture some of the same objects, but with different accuracy, precision, resolution, etc. (diverse observation), and some different objects that contribute to expanding or completing the ego vehicle's field of view.

[0121] Both V2X data and ego perception can differ from each other. This means that available information, such as the position of a detected object (across multiple observations), can differ from the ground truth. Covariances can also vary. Based on the V2X data received over the air, the ego vehicle (e.g., as a source node) can decide whether its observation improves the overall perception when communicating with other vehicles, route substations (RSUs), or any nodes or ITS stations.

[0122] In one embodiment, V2X-based data and ego-vehicle observations can statistically represent object features through first and second moments. The respective object features are summarized in a state vector f, which is statistically characterized by the mean vector µ (first moment) and the covariance matrix Σ (second moment). Therefore, the ego-vehicle data, also referred to as the local environment model (LEM), and the V2X data are represented by the pairs (µ ∈ Σ ... LEM , Σ LEM ) and (µ V2X , Σ V2X ) represents. Both datasets are available to the ego vehicle. The ego vehicle, as the source node, can prioritize its information based on its quality and the added value (information gain; information value) it offers other connected nodes compared to data already available there. An example of such priorities 30 is in Fig.Figure 4 illustrates how objects 40 are selected for transmission based on their information gain, for example, their respective information values ​​31, 32, 33, 34, 35, 36, and a dynamic priority threshold 430. Furthermore, the Ego vehicle can integrate the received V2X data into its Local Environmental Model (LEM), thus obtaining data with the moments (µ). GEM , E GEM The V2X data and the local data of the Ego vehicle are also collectively referred to as the local environmental model (“Global Environmental Model”, GEM).

[0123] Findings from experimental cases show, for example: a) Given μGEM=μLEM=μV2X and det(∑LEM)>>det(∑V2X)≈det(∑GEM)→ observation of Ego vehicles is not helpful when combined with V2X data b) Given μGEM=μLEM=μV2X and det(∑GEM)>>det(∑LEM)≈det(∑V2X)→Observation of Ego vehicles helps considerably when combined with V2X data c) Given μGEM=μLEM=μV2X and det(∑LEM)>>det(∑V2X)≈det(∑GEM)→Observation of Ego vehicles becomes helpful when combined with V2X data d) Given μLEM=μGEM−α, where α(real number) is a predefined threshold (which can be derived based on association rules) → observation of Ego vehicles may possibly help when combined with V2X data Cases a, b, and c are intuitively understandable. Case d is based on the fact that the vehicle itself has different observations (assuming the criteria for correctness and trustworthiness are met) than the collective perception based on received messages. In this case, sharing this information with other vehicles is more valuable. The information value (Vol) of the vehicle itself, as an example of a source node, can be defined as a metric to quantify the value of the vehicle's observations in improving overall perception performance (that is, the higher the Vol, the better if it is shared). Therefore, for case d, Vol1 > Vol2 if |α1| > |α2|, where |α| < T / 2 (a predefined threshold from the association function and its requirements) can optionally be specified.

[0124] The following function characterizes the information gain or information value (Vol) for the source node information, such as the data of the ego vehicle with the feature vector f, which is represented by (µ LEM ,Σ LEM ) is described. VoILEM=max‖μLEM−μGEM‖→∞min‖∑LEM−∑GEM‖→0G(μLEM,μGEM,∑LEM,∑GEM)ordet(∑LEM)−det(∑GEM)→0

[0125] Any function G that satisfies the given conditions can be used as a Vol function. This means that any function G should be maximized if, for example, the norm ||µ LEM - µ GEM || increases, and should be minimized when ||Σ LEM - Σ GEM || or det(Σ LEM ) - det(Σ GEM ) decreases. This means that it decreases in ||Σ LEM - µ GEM || monotonically increasing and in ||Σ LEM - Σ GEMThe function G must be monotonically decreasing. The function G must be differentiable. To be sufficiently general, the norm and the determinant can be replaced by any other function with similar behavior.

[0126] In other embodiments, some functions that satisfy the above rule and conditions include, for example, an information value based on the revised Kullback-Leibler divergence. The Kullback-Leibler divergence of two Gaussian distributions can be specified as follows (All: distribution of the fused data, Ev: distribution of the Ego vehicle data, K is the dimension of the feature vector). DKL(GEM‖LEM)=12[tr(∑LEM−1∑GEM)+(μLEM−μGEM)T∑LEM−1(μLEM−μGEM)−K+log(det(∑LEM)det(∑GEM))] The formula above takes into account divergence in the mean rather than convergence. Therefore, the revised KL divergence reflects divergence in covariance separation and convergence in mean separation. Therefore: RDKL(GEM‖LEM)=12[tr(∑LEM−1∑GEM)−(μLEM−μGEM)T∑LEM−1(μLEM−μGEM)−K+log(det(∑LEM)det(∑GEM))] Therefore, the value of information is inversely proportional to the divergence in the covariance and the convergence in the mean (revised KL divergence) and can be written as follows: VoI=A*exp(−B−RDKL+C) Note that in the revised KL divergence, the -K term is crucial for normalization, dimensional correction, and adjustment of the divergence measure. A, B, and C are constants.

[0127] Furthermore, the only difference between D KL and RD KLin the sign of the Mahalanobis distance term. In some cases, a reversal from -K to +K may also be suitable for a better representation of the information value (Vol).

[0128] Weighted sum of the determinant of the Kalman gain (for fusion) and the Bhattacharya distance. VoI=det(∑V2X(∑V2X+∑LEM)−1+β(μLEM−μGEM)T∑LEM−1(μLEM−μGEM)) where E V2X and Σ LEM Covariance matrices are, µ LEM and µ GEM Mean vectors are, and β is a suitable scaling factor.

[0129] Any custom function design can be based on the following formulas d∑=‖∑GEM−∑LEM‖F or ∑=det(∑GEM−−∑LEM)or∑=det(∑GEM)−−det(∑LEM) dμ=‖μGEM−−μLEM‖or VoI=β1dμ−−β2dΣ where β are suitable scaling factors (weights).

[0130] The information value function 20 can advantageously take into account both deterministic (e.g. the determinant of covariance matrices) and probabilistic (e.g. the Bhattacharya distance) measures.

[0131] Users can define their own custom functions that integrate specific comparison metrics. Examples include the difference in Frobenius norms between covariance matrices or the Euclidean distance between means.

[0132] The Vol function 20 can combine these different metrics using weighted sums, allowing users to prioritize specific factors based on their application requirements.

[0133] The above explanation of the embodiments describes the present invention by means of examples. Naturally, individual features of the embodiments can be combined with one another as desired, provided this is technically feasible without departing from the scope of the present invention.

Claims

[1] Method (100) for selecting an object using an evaluation unit (12) of a node (10), wherein the method (100) comprises: - Received (101) by the evaluation unit (12) object data from objects, wherein the object data for each of the objects includes a state vector and covariance information, - Determining (102), by the evaluation unit (12), an object perception quality score for each of the objects based on the covariance information, - Comparing (103) the determined object perception quality value for each of the objects with an object perception quality threshold according to a threshold requirement for selecting relevant objects, - Checking (104) a fulfillment of the limit requirement based on the comparison (103), wherein the limit requirement defines the condition that must be met for the object perception quality value to be satisfied, - Providing (105) a selection list of objects to be selected depending on a positive result of the check (104). [2] Method (100) according to claim 1, characterized by , that the procedure (100) further comprises at least one of the following: - Providing the specified object perception quality value for further processing at the node (10), - Transferred, by the evaluation unit (12), the selection list of objects to another unit (20) of the node (10) for further processing of the selected object data. [3] Method (100) according to any one of the preceding claims, characterized by , that the condition for verification (104) of the limit requirement specifies that the specified object perception quality value must be equal to, or greater than, or equal to or greater than, or less than, or equal to or less than the object perception quality limit. [4] Method (100) according to any one of the preceding claims, further comprising: - Determine, by the evaluation unit (12), weighted covariance information based on the received covariance information, - wherein the weighted covariance information preferably comprises a weighted covariance matrix or at least one weighted component of a covariance matrix, and - where the object perception quality score is determined based on the weighted covariance information. [5] Method (100) according to claim 4, characterized by , that a weight of each of the components of the covariance matrix is ​​designed such that it differs from at least one other weight of a component, or that the weights of all components are the same. [6] Method (100) according to any one of the preceding claims, characterized by , that - the covariance information preferably includes a covariance matrix, - the comparison (103) of the determined object perception quality score is based on comparing each component of the covariance matrix and / or each weighted component of the covariance matrix with corresponding components of an object perception quality threshold matrix, or - the comparison (103) of the determined object perception quality value for each of the objects with an object perception quality limit based on an entropy-based model. [7] Method (100) according to any one of claims 4 to 6, characterized by , that the comparison (103) of the object perception quality score is based on the components of the covariance matrix and / or the weighted components of the weighted covariance matrix for each of the objects, wherein each of the one or more components and / or one or more weighted components is compared with the object perception quality threshold. [8] Method (100) according to any one of the preceding claims, characterized by , that the comparison (103) of the determined object perception quality value is based on comparing a scalar value derived from the covariance information and / or the weighted covariance information with the object perception quality limit. [9] Method (100) according to any one of the preceding claims, characterized by , that the node (10) comprises a sensor system (11) for perceiving the environment, and / or a communication system (13) for receiving a message with object data from at least one other node, and / or a V2X system as a sensor, wherein the received object data comprises object data of objects that are perceived by the sensor system (11) in the environment and / or by the communication system (13) and / or by the V2X system as a sensor (14). [10] Evaluation unit (12) for filtering object data, comprising means for carrying out the method (100) according to any one of claims 1 to 9. [11] Data processing device (10) comprising means for carrying out the method (100) according to any one of claims 1 to 9. [12] Platform configured as an infrastructure unit or as a vehicle, comprising the evaluation unit (12) according to claim 10 and / or the data processing device (10) according to claim 11. [13] Computer program (20) comprising instructions which, when the computer program (20) is executed by the evaluation unit (12) according to claim 10 and / or the data processing device (10) according to claim 11 and / or the platform according to claim 12, cause the evaluation unit (12) and / or the data processing device (10) and / or the platform to execute the method (100) according to any one of claims 1 to 9. [14] Computer-readable storage medium (15) on which the computer program (20) according to claim 12 is stored. [15] Message (600) comprising a V2X message (610), wherein the V2X message (610) comprises object data of objects (40), characterized by , that the object data are processed according to the steps of the method (100) according to one of claims 1 to 9.

Images