Identity recognition method, device and system

Abstract

This invention provides an identity recognition method, apparatus, and system. The method includes: preprocessing collected vehicle-side raw data and roadside raw data to obtain first vehicle-side data and first roadside data; obtaining normalized feature vectors of the vehicle-side data and roadside data based on the first vehicle-side data and the first roadside data; matching the vehicle-side data and roadside data based on the similarity between the normalized feature vectors of the vehicle-side data and the roadside data to obtain an identity matching result; the identity matching result includes the association relationship between vehicle identifiers in the vehicle-side data and target object identifiers in the roadside data, thereby achieving target identity recognition in vehicle-road cooperative systems.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation and big data technology, and in particular to an identity recognition method, device and system. Background Technology

[0002] In recent years, the automotive industry has experienced rapid development, with the number of cars in China showing a year-on-year growth trend. Along with the increasing number of motor vehicles, a series of traffic-related problems, such as traffic congestion and traffic violations, have become more frequent. Faced with increasingly complex road traffic conditions, and for the needs of road traffic supervision and operation, traffic management departments or operating departments typically install roadside cameras in scenarios such as urban traffic and highways to monitor road conditions in real time. The data collected by roadside equipment can not only be used to monitor traffic flow, vehicle speed, and road congestion on target road sections, but also play a role in identifying traffic accidents and violations, and tracking special vehicles.

[0003] For certain regulatory or security needs, traffic management departments need to identify and verify the identities of specific target vehicles. However, in practice, there are many difficulties:

[0004] 1) While current target detection technologies based on image processing and deep learning are relatively mature, in practical applications, due to external factors such as camera shooting angle and lighting environment, the detection accuracy is low when the camera captures license plate information for vehicle identification in certain scenarios, which fails to achieve the relevant regulatory objectives.

[0005] 2) With the continuous development of vehicle-to-everything (V2X) technology, more and more vehicles will be equipped with OBU (On-Board Unit) terminals with V2X communication capabilities, which can send the vehicle's dynamic information to the relevant V2X platform in real time. If the target vehicle is equipped with an OBU on-board terminal, its location, speed, vehicle type, and other information will be sent to the relevant platform in real time. However, due to errors such as GPS (Global Positioning System) data drift and network latency, it is difficult to accurately locate the real-time dynamics of the target vehicle based solely on the information reported by the vehicle. Furthermore, it is impossible to query the target vehicle's road driving image data relying solely on OBU data. Summary of the Invention

[0006] The purpose of this invention is to provide an identity recognition method, apparatus, and device to solve the problem of low detection accuracy caused by using only a single-source sensor for target detection and identity recognition in the prior art.

[0007] To address the above problems, embodiments of the present invention provide an identity recognition method, including:

[0008] The collected raw vehicle-side data and raw roadside data are preprocessed to obtain the first vehicle-side data and the first roadside data;

[0009] Based on the first vehicle-side data and the first roadside data, obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data;

[0010] By comparing the similarity between the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data, the vehicle-side data and the roadside data are matched to obtain the identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data.

[0011] The raw data from the vehicle side includes at least one of the following: vehicle identification, vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position.

[0012] The preprocessing of the collected raw vehicle-side data to obtain the first vehicle-side data includes:

[0013] Data cleaning is performed on at least one of the vehicle speed, the vehicle heading angle, and the vehicle position to obtain the first vehicle-side data.

[0014] The roadside raw data includes: road traffic monitoring videos or images collected by roadside sensing devices;

[0015] The preprocessing of the collected raw roadside data to obtain the first roadside data includes:

[0016] Target object detection is performed on the raw roadside data to determine at least one of the following: target object type, target object size, target object speed, target object heading angle, and target object position.

[0017] Data cleaning is performed on at least one of the target object's speed, the target object's heading angle, and the target object's position to obtain the first roadside data.

[0018] Specifically, based on the first vehicle-side data and the first roadside data, the normalized feature vectors of the vehicle-side data and the roadside data are obtained, including:

[0019] Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data;

[0020] The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data.

[0021] The feature vector of the vehicle-side data is composed of the feature attributes of the vehicle-side data, including: vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; the feature vector of the roadside data is composed of the feature attributes of the roadside data, including: target object type, target object size, target object speed, target object heading angle, and target object position.

[0022] Specifically, based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one candidate set for similarity calculation, including:

[0023] Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain multiple candidate sets to be matched. Each candidate set to be matched includes: a set of vehicle-side data and multiple sets of roadside data with trajectories similar to the set of vehicle-side data.

[0024] Based on vehicle-side data, multiple sets of roadside data in the candidate set to be matched are filtered for temporal relationships to obtain a similarity calculation candidate set; the similarity calculation candidate set is a subset of the candidate set to be matched.

[0025] Specifically, the vehicle-side data and roadside data are matched based on the similarity between the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data to obtain identity matching results, including:

[0026] Calculate the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data;

[0027] If, at multiple consecutive time points, the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data is higher than a threshold value, it is determined that the vehicle in the vehicle-side data matches the target object in the roadside data.

[0028] The method further includes:

[0029] Obtain vehicle identification;

[0030] Based on the vehicle identifier, query the identity matching result to determine the target object identifier associated with the vehicle identifier;

[0031] Based on the target object identifier, determine the image-level roadside raw data associated with the target object identifier.

[0032] This invention also provides an identity recognition device, comprising:

[0033] The first processing module is used to preprocess the collected raw vehicle-side data and raw roadside data to obtain the first vehicle-side data and the first roadside data.

[0034] The second processing module is used to obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data based on the first vehicle-side data and the first roadside data.

[0035] The matching module is used to match vehicle-side data and roadside data based on the similarity between the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data to obtain an identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data.

[0036] This invention also provides an identity recognition system, including a processor and a transceiver. The transceiver receives and transmits data under the control of the processor, and the processor is used to perform the following operations:

[0037] The collected raw vehicle-side data and raw roadside data are preprocessed to obtain the first vehicle-side data and the first roadside data;

[0038] Based on the first vehicle-side data and the first roadside data, obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data;

[0039] By comparing the similarity between the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data, the vehicle-side data and the roadside data are matched to obtain the identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data.

[0040] The raw data from the vehicle side includes at least one of the following: vehicle identification, vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position.

[0041] The processor is also used to perform the following operations:

[0042] Data cleaning is performed on at least one of the vehicle speed, the vehicle heading angle, and the vehicle position to obtain the first vehicle-side data.

[0043] The roadside raw data includes: road traffic monitoring videos or images collected by roadside sensing devices;

[0044] The processor is also used to perform the following operations:

[0045] Target object detection is performed on the raw roadside data to determine at least one of the following: target object type, target object size, target object speed, target object heading angle, and target object position.

[0046] Data cleaning is performed on at least one of the target object's speed, the target object's heading angle, and the target object's position to obtain the first roadside data.

[0047] The processor is also used to perform the following operations:

[0048] Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data;

[0049] The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data.

[0050] The feature vector of the vehicle-side data is composed of the feature attributes of the vehicle-side data, including: vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; the feature vector of the roadside data is composed of the feature attributes of the roadside data, including: target object type, target object size, target object speed, target object heading angle, and target object position.

[0051] The processor is also used to perform the following operations:

[0052] Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain multiple candidate sets to be matched. Each candidate set to be matched includes: a set of vehicle-side data and multiple sets of roadside data with trajectories similar to the set of vehicle-side data.

[0053] Based on vehicle-side data, multiple sets of roadside data in the candidate set to be matched are filtered for temporal relationships to obtain a similarity calculation candidate set; the similarity calculation candidate set is a subset of the candidate set to be matched.

[0054] The processor is also used to perform the following operations:

[0055] Calculate the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data;

[0056] If, at multiple consecutive time points, the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data is higher than a threshold value, it is determined that the vehicle in the vehicle-side data matches the target object in the roadside data.

[0057] The processor is also used to perform the following operations:

[0058] Obtain vehicle identification;

[0059] Based on the vehicle identifier, query the identity matching result to determine the target object identifier associated with the vehicle identifier;

[0060] Based on the target object identifier, determine the image-level roadside raw data associated with the target object identifier.

[0061] This invention also provides an identity recognition device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the identity recognition method described above.

[0062] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the identity recognition method described above.

[0063] The above-described technical solution of the present invention has at least the following beneficial effects:

[0064] In the identity recognition method, apparatus, and device of this invention, multiple sources of information are aggregated by collecting roadside raw data and vehicle-side raw data. The collected roadside raw data and vehicle-side raw data are analyzed and calculated. By comparing the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data, the vehicle-side data and roadside data are matched to obtain the identity matching result, thereby realizing the target identity recognition of vehicle-road cooperative system. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating the steps of the identity recognition method provided in an embodiment of the present invention.

[0066] Figure 2 This is a schematic diagram of the structure of the identity recognition device provided in an embodiment of the present invention;

[0067] Figure 3 This is a schematic diagram showing the structure of the identity recognition system provided in an embodiment of the present invention. Detailed Implementation

[0068] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0069] like Figure 1 As shown, an embodiment of the present invention provides an identity recognition method, including:

[0070] Step 101: Preprocess the collected raw vehicle-side data and raw roadside data to obtain the first vehicle-side data and the first roadside data;

[0071] Step 102: Based on the first vehicle-side data and the first roadside data, obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data;

[0072] Step 103: Match vehicle-side data and roadside data by the similarity between the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data to obtain identity matching results; the identity matching results include the association between vehicle identifiers in vehicle-side data and target object identifiers in roadside data.

[0073] Optionally, the normalized feature vector of the vehicle-side data is composed of the feature attributes of the vehicle-side data, including: vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; the normalized feature vector of the roadside data is composed of the characteristic attributes of the roadside data, including: target object type, target object size, target object speed, target object heading angle, and target object position.

[0074] The purpose of this invention is to achieve target identity recognition and authentication by collecting and recording data from vehicle-side and roadside sensors, and then fusing and analyzing data from different sources. Embodiments of this invention fully utilize multi-source information from vehicle-side and roadside perception data, increasing information redundancy and effectively improving recognition efficiency.

[0075] In an optional embodiment of the present invention, the raw vehicle-side data includes at least one of the following: vehicle identification, vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; for example, basic vehicle-side information (such as vehicle ID, vehicle type, and vehicle size) and vehicle operating status information (such as vehicle speed, heading angle, and position information) can be collected through the vehicle-mounted terminal equipment.

[0076] The corresponding step 101, which involves preprocessing the collected raw vehicle-side data to obtain the first vehicle-side data, includes:

[0077] Data cleaning is performed on at least one of the vehicle speed, the vehicle heading angle, and the vehicle position to obtain the first vehicle-side data.

[0078] To address potential data gaps or anomalies in the collected raw vehicle-side data, data cleaning is performed on the raw vehicle-side data. Specifically, this includes speed cleaning, heading angle cleaning, and position coordinate value cleaning.

[0079] Among them, velocity cleaning can be understood as: comparing the velocity data recorded at adjacent time points, calculating the velocity change rate and comparing it with a threshold. If the change in the calculated result is greater than the threshold, then fitting is performed based on the velocity at adjacent time points.

[0080] Heading angle cleaning can be understood as: comparing the heading angle data recorded at adjacent times, calculating the rate of change of heading angle and comparing it with a threshold. If the change in the calculated result is greater than the threshold, then fitting is performed based on the heading angle at adjacent times.

[0081] Position coordinate cleaning can be understood as: calculating the rate of change of velocity between adjacent time points, and if the rate of change of velocity is greater than a threshold, then fitting is performed based on the position coordinates between adjacent time points.

[0082] In another optional embodiment of the present invention, the roadside raw data includes: road traffic monitoring videos or images collected by roadside sensing devices; for example, road traffic monitoring videos or images can be collected based on roadside sensing devices (such as roadside surveillance cameras).

[0083] Accordingly, the preprocessing of the collected roadside raw data in step 101 to obtain the first roadside data includes:

[0084] Target object detection is performed on the raw roadside data to determine at least one of the following: target object type, target object size, target object speed, target object heading angle, and target object position.

[0085] Data cleaning is performed on at least one of the target object's speed, the target object's heading angle, and the target object's position to obtain the first roadside data.

[0086] Target object detection and analysis are performed on raw roadside data collected by roadside sensing devices. This mainly includes the detection and identification of motor vehicles and non-motor vehicles, as well as the identification of target-related attributes, such as the target object's speed, heading angle, position information, and target object type and size.

[0087] For example, target detection on the collected raw roadside data mainly involves detecting vehicle targets and obtaining information such as vehicle type and size. This step can be achieved using deep neural network learning methods, specifically using target detection algorithms such as Faster R-CNN and YOLO.

[0088] Furthermore, based on the detected target, information such as the target's velocity, position, and heading angle is calculated. Specifically:

[0089] Speed ​​and heading angle information can be obtained by the dynamic difference between adjacent data frames.

[0090] Location: The target location information obtained from video detection is a relative coordinate in the image coordinate system. To facilitate the fusion calculation of roadside data and vehicle-side reported absolute position coordinates, this embodiment of the invention performs coordinate system transformation on the relative position coordinates of the roadside-sensed target, converting them to an absolute position coordinate system. Specifically:

[0091]

[0092] Pos target Pos represents the absolute position coordinates of the roadside sensing target. station Let ρ be the absolute position coordinates of the roadside sensing device, and ρ be the rotation matrix of the sensing target relative to the roadside sensing device.

[0093] Data cleaning is performed on at least one of the following: the velocity of the target object, the heading angle of the target object, and the position of the target object after processing.

[0094] Among them, velocity cleaning can be understood as: comparing the velocity data recorded at adjacent time points, calculating the velocity change rate and comparing it with a threshold. If the change in the calculated result is greater than the threshold, then fitting is performed based on the velocity at adjacent time points.

[0095] Heading angle cleaning can be understood as: comparing the heading angle data recorded at adjacent times, calculating the rate of change of heading angle and comparing it with a threshold. If the change in the calculated result is greater than the threshold, then fitting is performed based on the heading angle at adjacent times.

[0096] Position coordinate cleaning can be understood as: calculating the rate of change of velocity between adjacent time points, and if the rate of change of velocity is greater than a threshold, then fitting is performed based on the position coordinates between adjacent time points.

[0097] If the preprocessed vehicle-side data and road-side data are directly fused and calculated, the following problems will arise:

[0098] The calculation process generates a large amount of computation, increasing computational complexity; some data attributes in vehicle-side and roadside data are not closely related in time and space, and such data, as noise, will affect the target identification effect.

[0099] This invention provides a two-stage cascaded spatiotemporal data filtering strategy, the purpose of which is to: (1) narrow the range of effective computational data; and (2) remove redundant noise interference. This strategy specifically consists of two stages:

[0100] a) First stage: Spatiotemporal coarse-grained data screening and feature extraction based on spatial trajectory relationships. The purpose is to achieve coarse-grained screening of targets with similar trajectories in vehicle-side data and roadside data based on the trajectory relationships between data, and to extract the feature attributes of the corresponding data, thereby narrowing the effective calculation range and eliminating redundant noise.

[0101] b) Second stage: Fine-grained spatiotemporal data screening and feature processing based on time sequence relationship, with the aim of: reducing the data time misalignment problem caused by the asynchronous clocks of the two types of data and transmission delay; and narrowing the data range for similar calculations between the two types of data.

[0102] The corresponding step 102 includes:

[0103] Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data;

[0104] The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data.

[0105] The feature vector of the vehicle-side data is composed of the feature attributes of the vehicle-side data, including: vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; the feature vector of the roadside data is composed of the feature attributes of the roadside data, including: target object type, target object size, target object speed, target object heading angle, and target object position.

[0106] The two-stage cascaded data filtering rule proposed in this invention effectively reduces data computation complexity and mitigates the impact of data noise.

[0107] As an optional embodiment, data screening is respectively performed based on the spatial trajectory relationship and the temporal relationship between the first vehicle-side data and the first roadside data to obtain at least one candidate set for similarity calculation, including:

[0108] (First stage) Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data screening is performed to obtain multiple candidate sets for matching. Each candidate set for matching includes: a set of vehicle-side data, and multiple sets of roadside data whose trajectories are similar to that of this set of vehicle-side data;

[0109] (Second stage) Taking the vehicle-side data as a reference, temporal relationship screening is performed on the multiple sets of roadside data in the candidate sets for matching to obtain a candidate set for similarity calculation; the candidate set for similarity calculation is a subset of the candidate sets for matching.

[0110] Among them, in the first stage: spatio-temporal coarse-grained data screening and feature extraction based on the spatial trajectory relationship are used to achieve the screening of targets with similar trajectories in the two types of data. Specifically,

[0111] If the first vehicle-side data includes M sets of data (i.e., M vehicles report their own data), and the first roadside data includes N sets of data (i.e., N target objects are captured by the roadside), taking the i-th set (0 < i ≤ M) of vehicle-side data as a reference, spatial trajectory similarity calculation is respectively performed with the N sets of roadside data, the roadside data with a relatively high similarity degree to the i-th set of vehicle-side data is screened, and the screened roadside data is used as the coarse-grained candidate combination for matching of the i-th set of vehicle-side data.<于

[0112] Suppose X and Y respectively represent vehicle-side data sequences and roadside data sequences with lengths of n and m, where, X = [x t1 , x t2 , x t3 , x t4 , ……, x tn ; Y = [y t1 , y t2 , y t3 , y t4 , ……, y tm . Calculate the common length res(X, Y) of the two sets of trajectory data sequences as:

[0113]

[0114] Furthermore, calculate the similarity Similarity between the two trajectories as:

[0115] Similarity = 1 - (res(X, Y)) / min(len X , len Y ); It should be noted that there may be some inaccuracies in the translation due to the complexity and potential ambiguity of the original patent text. It is recommended to review and verify the translation in combination with the specific patent context and relevant professional knowledge.

[0116] where len X is the length of X, and len Y is the length of Y.

[0117] According to the preset threshold Thresh_traj, determine the similarity between N groups of roadside data and the trajectory of the i-th group of vehicle-side data, and obtain K groups (0 ≤ K ≤ N) of roadside data similar to the trajectory of the i-th group of vehicle-side data, which serve as the candidate set {Candidate_match_set} to be matched for the i-th group of vehicle-side data i .

[0118] Further extract vehicle-side data attributes such as vehicle position, vehicle speed, vehicle heading angle, vehicle type, and vehicle size from the vehicle-side data in the candidate set to be matched, and obtain the feature vector Feature_vector of the vehicle-side data bsm .

[0119] Feature_vector bsm = [position, speed, angle, vehicleType, size]

[0120] Further extract roadside data attributes such as target object type, target object position, target object speed, target object heading angle, and target object size from the roadside data in the candidate set to be matched, and obtain the feature vector Feature_vector of the target object target .

[0121] Feature_vector target

[0122] = [position target , speed target , angle target , type target , size target

[0123] Second stage: Fine-grained spatio-temporal data screening and feature processing based on temporal relationships, to achieve screening of roadside data at times close to its time sequence with the vehicle-side data as the benchmark, and perform feature normalization processing on the vehicle-side data used as the benchmark and the roadside data at the similar times obtained by screening. Specifically

[0124] Taking the vehicle-side data as the reference value, if there are M groups of vehicle-side data, set the timestamp of the j-th (0 < j ≤ the total amount of data in the i-th group) (i.e., the data at the j-th moment in the i-th group of vehicle-side data) of the i-th group (0 < i ≤ M) of vehicle-side data as the reference timestamp T for this round of screening and matching calculation subset

[0125] ​Based on the baseline timestamp T, a candidate set {Candidate_match_set} is calculated for the i-th group of vehicle-side data obtained in the first stage. i Perform fine-grained filtering of time-series relationships, searching for roadside data with times close to the baseline timestamp, with a search range of: T search_range ~(T-△t, T+△t); Based on the above strategy, a candidate set {Candidate_computing_set} is selected for calculating the similarity between vehicle-side and roadside data under the condition of time j. i,j Repeat the above steps to calculate the similarity candidate set for all time conditions of the i-th group of vehicle-side data.

[0126] Given the inconsistency in the dimensions of vehicle-side and roadside data features, to eliminate the impact of differences in dimensions and scale of each feature dimension, the feature vectors of vehicle-side and roadside data in the aforementioned similarity calculation candidate set are normalized to obtain a normalized feature vector, Normalized_feature_vector. Implementation details:

[0127]

[0128] Where μ is the mean of the original features, σ is the standard deviation of the original features, and P is the original feature.

[0129] Finally, the normalized similarity calculation candidate set {Normalized_Candidate_computing_set} is obtained. i,j For example, as shown below,

[0130]

[0131] In at least one embodiment of the present invention, step 103 includes:

[0132] Calculate the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data;

[0133] If, at multiple consecutive time points, the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data is higher than a threshold value, it is determined that the vehicle in the vehicle-side data matches the target object in the roadside data.

[0134] This invention calculates the similarity between vehicle-side data feature vectors and roadside data feature vectors in the normalized similarity calculation candidate set. If, at multiple consecutive time points, the feature vector of a roadside target object and the feature vector of the vehicle-side data both have a high matching degree, then the two targets are considered to point to the same target entity. Specifically:

[0135] Taking the vehicle side data of the i-th group (i.e., the vehicle side data reported by the i-th vehicle) as an example, if the vehicle side data of the i-th group contains a total of n moments, then the similarity calculation candidate set of the vehicle side data at the j-th moment (0 < j ≤ n) of the i-th group can be expressed as {Normalized_Candidate_computing_set} i,j ;

[0136] Furthermore, calculate the normalized feature vectors of the vehicle side data at the j-th moment of the i-th group and the normalized feature vectors of each roadside candidate target in the candidate set respectively. Specifically, the similarity degree is measured based on the cosine distance for similarity calculation Similarity(X,Y):

[0137]

[0138]

[0139]

[0140] Finally, at the j-th moment, the similarity between the i-th vehicle in the vehicle side data and each roadside candidate target object is obtained:

[0141]

[0142] Furthermore, the similarity between the i-th vehicle in the vehicle side data and the roadside data candidate target at each moment is obtained:

[0143]

[0144] Furthermore, the similarity scores of each roadside target object (target) in the candidate set with the i-th vehicle on the vehicle side at different moments are comprehensively calculated to obtain the comprehensive similarity score between each roadside perception target object and the i-th vehicle in the vehicle side data:

[0145] [[ID=3�]]

[0146] Similarity bsmi-target =(Score target1 ,Score target2 ,……,Score targetN )

[0147] Finally, in Similarity bsmi-targetThe roadside target with the highest score is identified as pointing to the same target as the i-th vehicle in the vehicle-side data, thus achieving matching and identity verification between vehicle-side and roadside data. If the vehicle-side and roadside data successfully match and verify the identity, the IDs of the matched vehicle-side and roadside data are associated to form (Vehicle ID - Target ID), which serves as the matching result for target identity authentication.

[0148] In at least one embodiment of the present invention, the method further includes:

[0149] Obtain vehicle identification;

[0150] Based on the vehicle identifier, query the identity matching result to determine the target object identifier associated with the vehicle identifier;

[0151] Based on the target object identifier, determine the image-level roadside raw data associated with the target object identifier.

[0152] For example, a mapping relationship between roadside target object IDs and their corresponding images can be pre-built; using vehicle IDs as search criteria, the corresponding roadside target object IDs can be found by querying (vehicle ID - roadside target object ID); by querying the mapping relationship between roadside target object IDs and their corresponding images, image-level driving data of the target vehicle in a specific time and area can be retrieved.

[0153] The identity recognition method based on vehicle-road cooperative data proposed in this invention realizes the fusion analysis of vehicle-side data and roadside data, and can further achieve: continuous tracking or protection of special targets; self-localization of target vehicles; and further, can provide auxiliary conditions for expanding the perception range of target vehicles and analyzing the surrounding environment.

[0154] In summary, the identity recognition method provided by this invention fully utilizes multi-source information from vehicle-side data and roadside perception data. Compared with traditional methods for target recognition and identity authentication based on a single data source, it increases information redundancy and can effectively improve the accuracy of target identity recognition. Furthermore, based on a two-stage cascaded spatiotemporal data filtering rule of spatial trajectory relationship + temporal relationship, it effectively reduces data noise and the computational complexity of matching vehicle-side data and roadside data.

[0155] like Figure 2 As shown, embodiments of the present invention also provide an identity recognition device, comprising:

[0156] The first processing module 201 is used to preprocess the collected vehicle-side raw data and roadside raw data to obtain the first vehicle-side data and the first roadside data.

[0157] The second processing module 202 is used to obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data based on the first vehicle-side data and the first roadside data.

[0158] The matching module 203 is used to match vehicle-side data and roadside data based on the similarity between the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data to obtain an identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data.

[0159] As an optional embodiment, the raw vehicle-side data includes at least one of the following: vehicle identification, vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position;

[0160] The first processing module is further configured to:

[0161] Data cleaning is performed on at least one of the vehicle speed, the vehicle heading angle, and the vehicle position to obtain the first vehicle-side data.

[0162] As an optional embodiment, the roadside raw data includes: road traffic monitoring videos or images collected by roadside sensing devices;

[0163] The first processing module is further configured to:

[0164] Target object detection is performed on the raw roadside data to determine at least one of the following: target object type, target object size, target object speed, target object heading angle, and target object position.

[0165] Data cleaning is performed on at least one of the target object's speed, the target object's heading angle, and the target object's position to obtain the first roadside data.

[0166] As an optional embodiment, the second processing module is further configured to:

[0167] Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data;

[0168] The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data.

[0169] The feature vector of the vehicle-side data is composed of the feature attributes of the vehicle-side data, including: vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; the feature vector of the roadside data is composed of the feature attributes of the roadside data, including: target object type, target object size, target object speed, target object heading angle, and target object position.

[0170] As an optional embodiment, the second processing module is further configured to:

[0171] Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain multiple candidate sets to be matched. Each candidate set to be matched includes: a set of vehicle-side data and multiple sets of roadside data with trajectories similar to the set of vehicle-side data.

[0172] Based on vehicle-side data, multiple sets of roadside data in the candidate set to be matched are filtered for temporal relationships to obtain a similarity calculation candidate set; the similarity calculation candidate set is a subset of the candidate set to be matched.

[0173] As an optional embodiment, the matching module is further configured to:

[0174] Calculate the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data;

[0175] If, at multiple consecutive time points, the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data is higher than a threshold value, it is determined that the vehicle in the vehicle-side data matches the target object in the roadside data.

[0176] As an optional embodiment, the apparatus further includes:

[0177] The identification acquisition module is used to acquire vehicle identification.

[0178] The tag determination module is used to query the identity matching result based on the vehicle identifier and determine the target object identifier associated with the vehicle identifier;

[0179] The data determination module is used to determine the image-level roadside raw data associated with the target object identifier based on the target object identifier.

[0180] The identity recognition method provided in this invention makes full use of multi-source information from vehicle-side data and roadside perception data. Compared with traditional methods for target recognition and identity authentication based on a single data source, it increases information redundancy and can effectively improve the accuracy of target identity recognition. Furthermore, based on a two-stage cascaded spatiotemporal data filtering rule of spatial trajectory relationship + temporal relationship, it effectively reduces data noise and the computational complexity of matching vehicle-side data and roadside data.

[0181] It should be noted that the identity recognition device provided in the embodiments of the present invention is a device capable of executing the above-described identity recognition method. Therefore, all embodiments of the above-described identity recognition method are applicable to this device and can achieve the same or similar beneficial effects.

[0182] like Figure 3 As shown, this embodiment of the invention also provides an identity recognition system, including a processor 300 and a transceiver 310. The transceiver 310 receives and transmits data under the control of the processor 300, and the processor 300 is used to perform the following operations:

[0183] The collected raw vehicle-side data and raw roadside data are preprocessed to obtain the first vehicle-side data and the first roadside data;

[0184] Based on the first vehicle-side data and the first roadside data, obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data;

[0185] By comparing the similarity between the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data, the vehicle-side data and the roadside data are matched to obtain the identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data.

[0186] As an optional embodiment, the raw vehicle-side data includes at least one of the following: vehicle identification, vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position;

[0187] The processor is also used to perform the following operations:

[0188] Data cleaning is performed on at least one of the vehicle speed, the vehicle heading angle, and the vehicle position to obtain the first vehicle-side data.

[0189] As an optional embodiment, the roadside raw data includes: road traffic monitoring videos or images collected by roadside sensing devices;

[0190] The processor is also used to perform the following operations:

[0191] Target object detection is performed on the raw roadside data to determine at least one of the following: target object type, target object size, target object speed, target object heading angle, and target object position.

[0192] Data cleaning is performed on at least one of the target object's speed, the target object's heading angle, and the target object's position to obtain the first roadside data.

[0193] As an optional embodiment, the processor is also configured to perform the following operations:

[0194] Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data;

[0195] The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data.

[0196] The feature vector of the vehicle-side data is composed of the feature attributes of the vehicle-side data, including: vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; the feature vector of the roadside data is composed of the feature attributes of the roadside data, including: target object type, target object size, target object speed, target object heading angle, and target object position.

[0197] As an optional embodiment, the processor is also configured to perform the following operations:

[0198] Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain multiple candidate sets to be matched. Each candidate set to be matched includes: a set of vehicle-side data and multiple sets of roadside data with trajectories similar to the set of vehicle-side data.

[0199] Based on vehicle-side data, multiple sets of roadside data in the candidate set to be matched are filtered for temporal relationships to obtain a similarity calculation candidate set; the similarity calculation candidate set is a subset of the candidate set to be matched.

[0200] As an optional embodiment, the processor is also configured to perform the following operations:

[0201] Calculate the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data;

[0202] If, at multiple consecutive time points, the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data is higher than a threshold value, it is determined that the vehicle in the vehicle-side data matches the target object in the roadside data.

[0203] As an optional embodiment, the processor is also configured to perform the following operations:

[0204] Obtain vehicle identification;

[0205] Based on the vehicle identifier, query the identity matching result to determine the target object identifier associated with the vehicle identifier;

[0206] Based on the target object identifier, determine the image-level roadside raw data associated with the target object identifier.

[0207] The identity recognition method provided in this invention makes full use of multi-source information from vehicle-side data and roadside perception data. Compared with traditional methods for target recognition and identity authentication based on a single data source, it increases information redundancy and can effectively improve the accuracy of target identity recognition. Furthermore, based on a two-stage cascaded spatiotemporal data filtering rule of spatial trajectory relationship + temporal relationship, it effectively reduces data noise and the computational complexity of matching vehicle-side data and roadside data.

[0208] It should be noted that the identity recognition system provided in the embodiments of the present invention is a system capable of executing the above-described identity recognition method. Therefore, all embodiments of the above-described identity recognition method are applicable to this system and can achieve the same or similar beneficial effects.

[0209] This invention also provides an identity recognition device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the various processes in the identity recognition method embodiments described above and achieves the same technical effect. To avoid repetition, these will not be repeated here.

[0210] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, this program implements the various processes described in the above-described identity recognition method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0211] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0212] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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 program instructions. These computer program 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, generate instructions for implementing the flowchart... Figure 1 A device for one or more processes and / or the functions specified in one or more boxes.

[0213] These computer program 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 a paper article including an instruction means, the instruction means being implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0214] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment, causing the computer or other programmable equipment to perform a series of operational steps to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0215] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An identity recognition method, characterized in that, include: The collected raw vehicle-side data and raw roadside data are preprocessed to obtain the first vehicle-side data and the first roadside data; Based on the first vehicle-side data and the first roadside data, obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data; By comparing the similarity between the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data, the vehicle-side data and the roadside data are matched to obtain the identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data; The step of obtaining the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data based on the first vehicle-side data and the first roadside data includes: Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data; The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data. Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one candidate set for similarity calculation, including: Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain multiple candidate sets to be matched. Each candidate set to be matched includes: a set of vehicle-side data and multiple sets of roadside data with trajectories similar to the set of vehicle-side data. Based on vehicle-side data, multiple sets of roadside data in the candidate set to be matched are filtered for temporal relationships to obtain a similarity calculation candidate set; the similarity calculation candidate set is a subset of the candidate set to be matched. The method further includes: Obtain vehicle identification; Based on the vehicle identifier, query the identity matching result to determine the target object identifier associated with the vehicle identifier; Based on the target object identifier, determine the image-level roadside raw data associated with the target object identifier.

2. The method according to claim 1, characterized in that, The raw data from the vehicle side includes at least one of the following: vehicle identification, vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position. The preprocessing of the collected raw vehicle-side data to obtain the first vehicle-side data includes: Data cleaning is performed on at least one of the vehicle speed, the vehicle heading angle, and the vehicle position to obtain the first vehicle-side data.

3. The method according to claim 1, characterized in that, The raw roadside data includes: road traffic monitoring videos or images collected by roadside sensing devices; The preprocessing of the collected raw roadside data to obtain the first roadside data includes: Target object detection is performed on the raw roadside data to determine at least one of the following: target object type, target object size, target object speed, target object heading angle, and target object position. Data cleaning is performed on at least one of the target object's speed, the target object's heading angle, and the target object's position to obtain the first roadside data.

4. The method according to claim 1, characterized in that, in, The feature vector of the vehicle-side data is composed of the feature attributes of the vehicle-side data, including: vehicle type, vehicle size, vehicle speed, vehicle heading angle, and vehicle position; the feature vector of the roadside data is composed of the feature attributes of the roadside data, including: target object type, target object size, target object speed, target object heading angle, and target object position.

5. The method according to claim 1, characterized in that, By analyzing the similarity between the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data, matching is performed between the vehicle-side data and the roadside data to obtain identity matching results, including: Calculate the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data; If, at multiple consecutive time points, the similarity between the normalized feature vector of the target object in the roadside data and the normalized feature vector of the vehicle in the vehicle-side data is higher than a threshold value, it is determined that the vehicle in the vehicle-side data matches the target object in the roadside data.

6. An identity recognition device, characterized in that, include: The first processing module is used to preprocess the collected raw vehicle-side data and raw roadside data to obtain the first vehicle-side data and the first roadside data. The second processing module is used to obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data based on the first vehicle-side data and the first roadside data. The matching module is used to match vehicle-side data and roadside data based on the similarity between the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data to obtain an identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data; The second processing module is further used for: Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data; The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data. The second processing module is further used for: Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain multiple candidate sets to be matched. Each candidate set to be matched includes: a set of vehicle-side data and multiple sets of roadside data with trajectories similar to the set of vehicle-side data. Based on vehicle-side data, multiple sets of roadside data in the candidate set to be matched are filtered for temporal relationships to obtain a similarity calculation candidate set; the similarity calculation candidate set is a subset of the candidate set to be matched. The identification acquisition module is used to acquire vehicle identification. The tag determination module is used to query the identity matching result based on the vehicle identifier and determine the target object identifier associated with the vehicle identifier; The data determination module is used to determine the image-level roadside raw data associated with the target object identifier based on the target object identifier.

7. An identity recognition system, comprising a processor and a transceiver, wherein the transceiver receives and transmits data under the control of the processor, characterized in that, The processor is used to perform the following operations: The collected raw vehicle-side data and raw roadside data are preprocessed to obtain the first vehicle-side data and the first roadside data; Based on the first vehicle-side data and the first roadside data, obtain the normalized feature vector of the vehicle-side data and the normalized feature vector of the roadside data; By comparing the similarity between the normalized feature vectors of the vehicle-side data and the normalized feature vectors of the roadside data, the vehicle-side data and the roadside data are matched to obtain the identity matching result; the identity matching result includes the association between the vehicle identifier in the vehicle-side data and the target object identifier in the roadside data; The processor is also used to perform the following operations: Based on the spatial trajectory relationship and temporal relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain at least one similarity calculation candidate set; each similarity calculation candidate set includes: a set of vehicle-side data, and multiple sets of roadside data with trajectories similar to the set of vehicle-side data; The feature vectors of vehicle-side data and roadside data in the similarity calculation candidate set are normalized to obtain normalized feature vectors of vehicle-side data and roadside data. The processor is also used to perform the following operations: Based on the spatial trajectory relationship between the first vehicle-side data and the first roadside data, data filtering is performed to obtain multiple candidate sets to be matched. Each candidate set to be matched includes: a set of vehicle-side data and multiple sets of roadside data with trajectories similar to the set of vehicle-side data. Based on vehicle-side data, multiple sets of roadside data in the candidate set to be matched are filtered for temporal relationships to obtain a similarity calculation candidate set; the similarity calculation candidate set is a subset of the candidate set to be matched. The processor is also used to perform the following operations: Obtain vehicle identification; Based on the vehicle identifier, query the identity matching result to determine the target object identifier associated with the vehicle identifier; Based on the target object identifier, determine the image-level roadside raw data associated with the target object identifier.

8. An identity recognition device, comprising a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that, When the processor executes the program, it implements the identity recognition method as described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the identity recognition method as described in any one of claims 1-5.

Images