Clustering roadside sensing message comprehensive filtering method applying granularity calculation

Abstract

The invention belongs to the field of automatic driving environment perception, and relates to a clustering roadside perception message comprehensive filtering method applying granularity calculation, which comprises the following steps that: vehicle end equipment acquires data information of a roadside unit, and preprocesses the data information; respectively extracting n messages from the processed data information, and forming a sample point set by the n messages; dividing samples in the sample point set; screening out a candidate set according to the divided samples; constructing a common matrix according to the candidate set; fusing the data in the candidate set by adopting a hierarchical clustering algorithm based on condensation on the basis of the common matrix to obtain Pf; calculating a contour coefficient of the Pf, and screening the data according to the contour coefficient; combining the screened data to obtain a minimum non-repeated set; according to the invention, the problem of information redundancy caused by imperfect deployment strategies of RSU roadside equipment and vehicle end acquisition equipment, complicated equipment performance and traffic environment, and shielding of tall buildings and trucks is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of autonomous driving environment perception, and particularly relates to a clustering roadside perception message comprehensive filtering method applying granular computing. Background Art

[0002] With the continuous development of science and technology, car travel has gradually become the first choice for people's commuting and outings; thus, the ownership of private cars has been increasing year by year. As people's dependence on car travel becomes higher and higher, it has gradually led to high requirements for various connected vehicle communication technologies. The gradual development of connected vehicle communication technologies has facilitated people's daily lives and promoted the development of science and technology in the automotive field. Supported by the rapid development of various advanced science and technologies, the cooperative intelligent transportation system solution has gradually become the optimal solution to current traffic problems. Among them, the V2X technology, as a new generation of information and communication technology, focuses on realizing all-round communication between vehicles and the surrounding environment and network, providing environmental perception, information interaction, and collaborative control capabilities for automotive driving and traffic management applications. The RSU is a roadside device with V2X communication capabilities in the vehicle-road cooperation system, mainly responsible for collecting and forwarding roadside traffic and environmental information. Among them, perception messages are an important part of vehicle networking information.

[0003] Considering that connected vehicles do not need all the perception targets in the current traffic environment during driving, to reduce the V2X message processing volume and improve processing efficiency, only the target object information in some regions of interest needs to be concerned. On the other hand, due to the imperfect deployment of RSU devices, there is also a problem of cross-redundancy in the information collected by adjacent RSUs. The excessive amount of perception information data will bring a burden to the transmission process, and a large amount of redundant information not only causes excessive memory occupancy, resulting in memory waste and memory shortage, but also affects the efficiency of the RSU in transmitting effective information to connected vehicles.

[0004] Currently, most of the roadside device message filtering algorithms apply the Kalman filtering algorithm to obtain the optimal estimated state. Then, the non-maximum suppression algorithm (NMS) is used for attribute classification, and the intersection over union is used to calculate the information confidence level to achieve road grouping and data type grouping, and complete the information screening. However, in this method, the grouping categories are too large, and it is difficult to implement individual categories, such as the distance between people and vehicle obstacles, the driving speed and acceleration prediction of people and vehicles, the appearance and disappearance changes of road signs and markings, traffic signal data, extreme weather, road congestion situation prediction, etc. The information classification and redundancy screening. This makes the message transmission redundancy pressure of roadside devices still relatively large. Summary of the Invention

[0005] To solve the problems existing in the above prior art, the present invention proposes a comprehensive filtering method for clustering roadside perception messages applying granular computing, which includes: vehicle-end devices collect data information of roadside units and preprocess the data information; extract n messages from the processed data information respectively, and form a sample point set with the n messages; divide the samples in the sample point set; screen out a candidate set according to the divided samples; construct a co-link matrix based on the candidate set; fuse the data in the candidate set by using an agglomerative hierarchical clustering algorithm based on the co-link matrix to obtain the fused data P f ; calculate the silhouette coefficient of P f , and screen the data in the candidate set according to the silhouette coefficient; merge the screened data to obtain the smallest non-repeating set.

[0006] Advantages of the present invention:

[0007] The overall model of the present invention adopts a clustering method of granular computing to achieve information filtering. After completing time alignment and space alignment, the information transmitted by three types of autonomous driving information sources, namely MEC, SSM, and BSM, is effectively clustered, and the median data of the smallest cluster is taken as the representative data of this cluster for transmission. The present invention reduces the information redundancy problem caused by imperfect deployment strategies of RSU roadside devices and vehicle-end acquisition devices, as well as the complex performance of the devices themselves, traffic environment, and the occlusion of tall buildings and large trucks. On this basis, an information transmission judgment table for extreme weather and congestion conditions is introduced. When the impact of both on driverless driving reaches the critical value, relevant information is transmitted each time to ensure driving safety. Description of the drawings

[0008] Figure 1 is the overall flowchart of the present invention. Specific embodiments

[0009] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0010] A comprehensive filtering method for clustering roadside perception messages applying granular computing, as Figure 1As shown in the figure, the method includes: the vehicle-end device collects the data information of the roadside unit and preprocesses the data information; extracts n messages from the processed data information respectively, and forms a sample point set with the n messages; divides the samples in the sample point set; screens out the candidate set according to the divided samples; constructs a co-association matrix according to the candidate set; based on the co-association matrix, uses the agglomerative hierarchical clustering algorithm to fuse the data in the candidate set to obtain the fused data P f ; calculate the silhouette coefficient of P f , and screen the data in the candidate set according to the silhouette coefficient; merge the screened data to obtain the minimum non-repeating set.

[0011] In this embodiment, before the information obtained by the RSU is exchanged with the neighboring RSU, the received information of SSM, BSM, and MEC time is preprocessed, including operations such as decoding and decryption, and the operation of removing empty perception packets is completed.

[0012] The information after preprocessing is time-aligned by using the method of collecting with a set timer, and the sampling frequency is set to 10HZ. The RSU puts the received valid information into the Ssm_Queue, Bsm_Queue, and Mec_Queue message buffer queues respectively. By setting a timer, n messages are taken out from the three queues at the same time every t f time for processing. Subsequently, using UTM projection, the longitude and latitude of the perceived target are projected onto the plane coordinate system, so as to unify the BSM, SSM, and MEC information into the coordinate system X′O′Y′, and complete the spatial alignment. The coordinate system X′O′Y′ is the UTM coordinate system, and the coordinate system XOY is rotated by an angle θ around the point O to obtain X′O′Y′. The conversion formula is as follows:

[0013]

[0014] If the coordinates of the origin O in the coordinate system X′O′Y′ are recorded as (x0, y0), and the position coordinates of the original information are (x, y), then the updated coordinates (x′, y′) have the following expression:

[0015]

[0016] In this embodiment, a base clustering member selection method based on granularity distance is adopted, and the information to be distributed is selected from the three types of perceived information after time-space alignment as elements to form a clustering integration domain. U represents a set containing 3 n elements, which is called the clustering integration domain; its expression is:

[0017] U = {X1, X2, X3,... X 3n-1 , X 3n}

[0018] Measure the difference of the basic clustering results based on the granularity distance, select the partitioning result with the smallest difference and add it to the selection set to complete the member selection and partitioning.

[0019]

[0020] Among them, K(.) is the set of elements to be clustered, P and Q are elements in the set, U is the total sum of clustering elements, [.] P The direct sum operation of taking elements from a single set is X i For the traversal of elements in set P, X j For the traversal of elements in set Q, is the direct sum operation symbol.

[0021] The partitioning distance in the formula satisfies:

[0022] 0 ≤ dis(K(P), K(Q)) ≤ 1 - 1 / |U|

[0023] Assume that P = {p1, p2, …, p H} are H basic clustering results on the data set U. From the difference between the partitioning result p i and the aggregation of the entire clustering members, select the clustering member p c with the highest similarity in P and add it to the clustering member set S j . The clustering member set is an empty set in the initial state, and the number of members S N = 0.

[0024]

[0025] Among them, p i is the partitioning result.

[0026] Then iteratively select the clustering member p λ with the highest quality and the largest difference from the selected members from the remaining candidate sets and add it to the selection set until the required number of partitions H′ is satisfied to complete the class partitioning.

[0027]

[0028] In the formula, p i , p q ∈ P\S j , i ≠ q, λ = 2, 3, …, H′, S j is the currently selected clustering member set, j is the number of selected member sets, and P\S j is the number of candidate clustering member sets.

[0029] In this embodiment, a co-connected matrix clustering fusion algorithm based on granularity partitioning is used to obtain H clustering results. That is, the H partition results of the domain U are represented as a set P, Divide U into |V pi | subset. |V pi | is the partition result p i The number of clusters included, V pi Represents partition p i The label information value range generated for all elements in U; ​​the relationship between the two is:

[0030]

[0031] Among them, any element X i After dividing p i The generated category label can be represented by the information function f.

[0032] f:U×P→V

[0033] satisfy In summary, a related clustering integrated information system S can be expressed as a four-tuple S = (U, P, V, f).

[0034] Select the clustering results in set S and construct a new matrix including:

[0035]

[0036] Among them, A ij is the common matrix element, δ is the value operation for the traversed elements, λ is the initial count, H is the total number of set elements, X i and X j is the element to be traversed, |V pλ | is the number of clusters contained in the partition result pλ, C pλ (x i ) represents sample X i The label corresponding to the cluster in the partition result pλ.

[0037] In this embodiment, an agglomerative hierarchical clustering algorithm based on average connection is used as a consistent fusion function to fuse the selected integrated members, and the mean of all distances is used as the distance between two combined data points. The distance dis(p i ,p j ),j∈[1,H], and get the final result P f , P f The clustering results are:

[0038] P f={{X1,X2,X3,…X i}{X i+1 ,X i+2 ,X i+3 ,…X j}…{X p ,X p+1 ,X p+2 ,…X n}}

[0039] where n is the number of elements contained in the set U, and the variables i, j, and n are cluster indices. Denote the total number of variables as C, satisfying C = |V pλ |.

[0040] In this embodiment, combining the compactness within the cluster and the separation between clusters, the silhouette coefficient is used to evaluate the quality of clustering. Elements with a silhouette coefficient S (i) close to -1 are removed, and elements with a silhouette coefficient S (i) close to 0 are reclustered. The silhouette coefficient is:

[0041]

[0042] where b (i) is the average distance from data point i to all points in the nearest different cluster, and a (i) is the average distance from data point i to other points within its cluster.

[0043] In this embodiment, the classification obtained after processing according to the silhouette coefficient is further processed. The center point of the same cluster is selected as the representative point, and the smallest cluster is merged. Isolated clusters containing only one element are retained to achieve blind filling of the information of a single information source, so as to obtain the smallest non-redundant set C f =

[0044] {X1,X2,X3,…X i}, where i ≤ |V pλ |, achieving the purpose of reducing the redundancy between the three types of information BSM, SSM, and MEC.

[0045] The merged smallest set is the minimum non-redundant unit information to be transmitted by BSM, SSM, and MEC in the same time-frequency and spatial region. According to the information transmission judgment table for extreme weather and congestion conditions, when the impacts on autonomous driving caused by both reach the critical value, relevant information is transmitted each time, and extreme weather and congestion information is added to the minimum non-redundant unit to form the single-time minimum transmission unit C f ' = C f ∪{X j ,X w}, where C f is the smallest non-redundant set, and X jThe minimum unit obtained by clustering calculation, X w Is the minimum unit obtained according to the extreme weather and congestion situation information transmission table.

[0046] The above-described embodiments further illustrate in detail the object, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made to the present invention within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A comprehensive filtering method for clustered roadside perception messages applying granular computing, characterized in that, Including: The vehicle-end device collects the data information of the roadside unit and preprocesses the data information. Extract n messages from the processed data information respectively, and form a sample point set with the n messages. Divide the samples in the sample point set; screen out the candidate set according to the divided samples; construct a co-linkage matrix according to the candidate set. Based on the co - association matrix, the data in the candidate set is fused using an agglomerative hierarchical clustering algorithm to obtain the fused data P f ; Calculate the silhouette coefficient of P f , and screen the data in the candidate set according to the silhouette coefficient; Merge the screened data to obtain the smallest non - repeating set after screening and filtering.

2. The clustering roadside perception message comprehensive filtering method applying granularity calculation according to claim 1, wherein The data information of the roadside unit includes BSM data, SSM data, and MEC data.

3. The clustering roadside perception message comprehensive filtering method using granularity calculation according to claim 1, characterized in that Preprocessing the data information includes normalizing and dimension-reducing the data; discretizing the associated data; aligning the discretized data in time and space.

4. A comprehensive filtering method for clustering roadside perception messages applying granularity calculation according to claim 1, characterized in that Dividing the samples in the sample point set includes: dividing the sample points, calculating the distance between each division result and other division results; calculating the average value of the distances, and screening the samples in the sample set according to the average value and adding them to the selection set to obtain the divided samples and the undivided samples.

5. A comprehensive filtering method for clustering roadside perception messages applying granularity calculation according to claim 1, characterized in that Screening out the candidate set according to the divided samples includes: forming a clustering member set with the data of the undivided samples; calculating the distance between the clustering member set and the divided samples; calculating the quality and difference of each clustering member in the candidate set according to the distance; setting quality thresholds and difference thresholds, and adding the members greater than the quality thresholds and difference thresholds to the divided samples.

6. A comprehensive filtering method for clustering roadside perception messages applying granularity calculation according to claim 1, characterized in that Constructing a co-linkage matrix according to the candidate set includes: Among them, A ij is the co - association matrix element, δ is the value - taking operation on the traversed element, λ is the initial count, H is the total number of set elements, X i and X j are both elements to be traversed, |V pλ | is the number of clusters included in the partitioning result pλ, C pλ (x i ) represents the label corresponding to the cluster where the sample X i is located in the partitioning result pλ.

7. A comprehensive filtering method for clustering roadside perception messages applying granularity calculation according to claim 1, characterized in that Fusing the data in the candidate set using an agglomerative hierarchical clustering algorithm includes: obtaining the distance values of the data in each partitioning result, taking the mean of all distances as the distance between the combined data of the two partitions, and calculating the distance dis(p i , p j ), j ∈ [1, H] between each data point in the two combined data points and all other data points according to the granularity distance calculation formula to obtain the final clustering classification result; where dis(p i , p j ) is the distance between data point p i and data point p j , and H is the total number of data.

8. A comprehensive filtering method for clustered roadside perception messages applying granularity calculation according to claim 1, characterized in that, Calculate P for the fused data f Silhouette coefficient S i including removing elements of the silhouette coefficient S i close to -1, and reclustering elements of the silhouette coefficient S i close to 0.

9. The clustering roadside perception message comprehensive filtering method applying granularity calculation according to claim 8, characterized in that, The silhouette coefficient calculation formula is Among them, b (i) is the average distance between data point i and all points in the nearest different cluster, and a (i) is the average distance between data point i and other points within its affiliated cluster.

Images