A method for collecting and aligning intelligent driving test data in mixed scenarios with multiple vehicle models

Abstract

The present invention relates to the technical field of driving test data collection, and discloses a method for intelligent driving test data collection and alignment in a mixed scene of multiple vehicle models, the method comprising: constructing a target recognition model with an adaptive border to perform target recognition on driving scene video frames; extracting features from target recognition areas in different driving scene video frames; calculating similarity of feature vectors corresponding to target recognition areas in different driving scene video frames, and treating target recognition areas with similarity exceeding a preset threshold as same-target heterogeneous areas; integrating same-target heterogeneous areas to form driving test alignment data. Based on the similarity of target recognition areas in different driving scene video frames, the present invention treats target recognition areas with similarity exceeding a preset threshold as same-target heterogeneous areas, realizes clustering and integration processing of images of the same vehicle model under different viewing angles, and forms driving test alignment data of the same vehicle model under a mixed scene of multiple vehicle models and multiple viewing angles.

Description

Technical Field

[0001] The present invention relates to the technical field of driving test data collection, and in particular to an intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario. Background Art

[0002] With the rapid development of artificial intelligence and autonomous driving technologies, intelligent driving systems are gradually being applied to actual traffic scenarios. However, due to the complexity and diversity of the road environment and the differences between different models, the testing and verification of intelligent driving systems face huge challenges. Especially in multi-model mixed scenarios, how to effectively collect and align test data has become a key issue that needs to be solved urgently. Multi-model mixed scenarios refer to the presence of multiple types and brands of vehicles in the same road environment, including traditional fuel vehicles, new energy vehicles, and autonomous driving vehicles. This scenario is closer to the real traffic environment and can more comprehensively evaluate the performance of the intelligent driving system. However, different models have significant differences in sensor configuration, control strategy, dynamic characteristics, etc., which brings great difficulties to data collection and alignment. Traditional intelligent driving test methods are mainly conducted on a single model or similar models, which is difficult to adapt to the complexity of multi-model mixed scenarios. Summary of the invention

[0003] In view of this, the present invention proposes a method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models to cope with the challenges in a mixed scenario of multiple vehicle models and improve the test efficiency and accuracy of the intelligent driving system.

[0004] To achieve the above object, the present invention provides a method for collecting and aligning intelligent driving test data in a multi-vehicle mixed scenario, comprising the following steps:

[0005] S1: collecting driving scene video data through a camera, collecting a vehicle parameter data sequence through a sensor, and preprocessing the driving scene video data to obtain preprocessed driving scene video data, wherein the driving scene video data consists of a driving scene video frame and a video frame timestamp;

[0006] S2: construct an adaptive bounding box target recognition model to perform target recognition on the driving scene video frame, and obtain the target recognition area of ​​the driving scene video frame;

[0007] S3: Extracting features of target recognition areas in different driving scene video frames to obtain feature vectors of the target recognition areas, wherein the cross-vision Transformer is an implementation method of the feature extraction;

[0008] S4: Calculate the similarity of feature vectors corresponding to target recognition areas in different driving scene video frames, and regard target recognition areas with similarity exceeding a preset threshold as same-target different-source areas;

[0009] S5: Integrate the same target heterogeneous regions to obtain the image data sequence and vehicle parameter data sequence of the same vehicle model to form the driving test alignment data.

[0010] As a further improvement method of the present invention:

[0011] Optionally, in step S1, collecting driving scene video data by a camera and collecting vehicle parameter data sequence by a sensor include:

[0012] The driving scene video data is collected by the camera, wherein the driving scene video data consists of driving scene video frames and video frame timestamps. for:

[0013] ;

[0014] ;

[0015] in:

[0016] represents the driving scene video data captured by the mth camera, , M represents the total number of cameras in the driving scene;

[0017] Represents driving scene video data The nth driving scene video frame in , N represents the number of driving scene video frames in the driving scene video data, Indicates the video frame timestamp corresponding to the nth driving scene video frame;

[0018] Preprocessing the driving scene video data to obtain preprocessed driving scene video data;

[0019] The vehicle parameter data sequence is collected by using sensors, wherein the collected vehicle parameter data sequence set is:

[0020] ;

[0021] in:

[0022] C represents the set of vehicle parameter data sequences;

[0023] represents the vehicle parameter data sequence corresponding to the e-th vehicle model, Indicates that the e-th model is at timestamp The vehicle parameter data includes the vehicle's instantaneous speed, acceleration, yaw angular velocity, four-wheel lateral force, four-wheel steering angle and four-wheel drive torque.

[0024] Optionally, preprocessing the driving scene video data to obtain preprocessed driving scene video data includes:

[0025] The driving scene video data is preprocessed to obtain preprocessed driving scene video data, wherein the driving scene video data The preprocessing process is:

[0026] S11: Acquiring driving scene video data All driving scene video frames in;

[0027] S12: grayscale processing is performed on any pixel in the driving scene video frame, wherein the driving scene video frame The pixel at row x and column y The grayscale processing formula is:

[0028] ;

[0029] in:

[0030] Represents pixels Gray value of

[0031] In order, it represents pixels Color values ​​in R, G, B color channels;

[0032] S13: Perform brightness correction on the gray value of the pixel, where the gray value The brightness correction formula is:

[0033] ;

[0034] S14: The pixels after brightness correction are used to form a preprocessing result of a driving scene video frame, wherein the driving scene video frame The preprocessing results are:

[0035] ;

[0036] in:

[0037] Represents a driving scene video frame The preprocessing results are It is in the form of a matrix of X rows and Y columns, where X represents the number of pixel rows in the driving scene video frame, and Y represents the number of pixel columns in the driving scene video frame.

[0038] Optionally, in step S2, constructing an adaptive bounding box object recognition model includes:

[0039] Constructing an adaptive border target recognition model, wherein the target recognition model takes the preprocessed driving scene video frame as input and takes the target recognition area in the driving scene video frame as output, wherein the adaptive border target recognition model includes an input layer, a background score calculation layer, an initial border generation layer, an alignment convolution layer, and an output layer;

[0040] The input layer is used to receive the preprocessed driving scene video frames;

[0041] The background score calculation layer is a convolutional neural network structure, which is used to calculate the background score of any pixel in the driving scene video frame;

[0042] The initial border generation layer performs target-background classification based on the background score of the pixel and generates the initial border of the target area;

[0043] The alignment convolution layer is used to map and sample the pixels in the initial border, capture the geometric information and context information of the border, and correct the initial border;

[0044] The output layer is used to extract the image area in the corrected bounding box as the target recognition area of ​​the driving scene video frame.

[0045] Optionally, the performing target recognition on the driving scene video frame using the target recognition model includes:

[0046] The target recognition model is used to perform target recognition on the driving scene video frame, where the driving scene video frame is preprocessed The target recognition process is:

[0047] S21: The input layer receives the preprocessed driving scene video frames ;

[0048] S22: Background score calculation layer calculates driving scene video frames The background score of any pixel in the driving scene video frame The background score of the pixel at row x and column y in is:

[0049] ;

[0050] in:

[0051] Represents a driving scene video frame The background score of the pixel in the xth row and yth column;

[0052] Indicates the grayscale change threshold of adjacent video frames, Indicates the grayscale change threshold of the initial video frame;

[0053] represents the adjustment coefficient; in the embodiment of the present invention, Set to 0.01;

[0054] represents an exponential function with a natural constant as base;

[0055] S23: The initial border generation layer performs target-background classification based on the background score of the pixel, obtains the probability that the pixel is the target pixel, and generates an initial border containing the target pixel, where the driving scene video frame The probability that the pixel in the xth row and yth column is the target pixel is:

[0056] ;

[0057] in:

[0058] Represents a driving scene video frame The probability that the pixel in the xth row and yth column is the target pixel;

[0059] Represents a driving scene video frame The pixel at row x and column y is the center The mean background score of the neighborhood pixel area;

[0060] Composing driving scene video frames The target pixel probability matrix: ;

[0061] Using the object detection convolution matrix Perform convolution calculation on the target pixel probability matrix and get The coordinates of the border vertices of the group target area, where Represents a driving scene video frame The number of targets in , where the coordinates of the border vertices of the k-th target area are:

[0062] ;

[0063] in:

[0064] Represents the coordinates of the border vertices of the kth group of target areas, Represents the coordinates of the four vertices of the border;

[0065] In an embodiment of the present invention, the pixel coordinates in the convolution result of the target pixel probability matrix that are higher than a preset threshold are used as a vertex position coordinate, the two nearest vertex position coordinates are used as diagonal coordinates in the vertex position coordinates of the border, and the four vertex position coordinates of the border are generated. If the target pixel ratio in the four vertex position coordinates is higher than the preset threshold, the border is retained, otherwise it is not retained.

[0066] S24: The aligned convolution layer is used to map and sample the pixels in the initial border, where the border vertex position coordinates The pixel position in the frame The corresponding sampling pixel position is , ,

[0067] , , , , , ;in The pixel positions are The distance to the left, top, right, and bottom of the border;

[0068] Based on sampling pixel positions in driving scene video frames The 9 pixels obtained by sampling are used as the neighborhood pixel area of ​​the pixels in the initial border, and the border weight matrix is ​​used to calculate the neighborhood pixel area of ​​the pixels in the initial border. Perform convolution processing on the neighborhood pixel area to obtain the mapping value of the neighborhood pixel area;

[0069] S25: Correcting the initial border so that the average mapping value in the border is higher than a preset mapping threshold;

[0070] S26: Extract the image area in the corrected border as the driving scene video frame target identification area.

[0071] Optionally, in step S3, feature extraction of target recognition areas in different driving scene video frames includes:

[0072] The target recognition area in different driving scene video frames is feature extracted to obtain the feature vector of the target recognition area, wherein the cross-vision transformer is the implementation method of the feature extraction, and the target recognition area The feature extraction process is:

[0073] S31: Target recognition area Perform residual convolution calculation to obtain the target recognition area Multi-scale residual convolution features:

[0074] ;

[0075] in:

[0076] Indicates the target recognition area The residual convolution feature at scale z, , Z represents the maximum scale;

[0077] represents the convolution operator;

[0078] express The pixel-sized residual convolution matrix, express The pixel-sized residual convolution matrix;

[0079] S32: Calculate the cross attention of residual convolution features at different scales, where the residual convolution features The cross attention is:

[0080] ;

[0081] in:

[0082] Represents the residual convolution feature Cross-attention;

[0083] represents an activation function; in the embodiment of the present invention, the selected activation function is a ReLU function;

[0084] represents the attention convolution weight matrix;

[0085] S33: Constructing the target recognition area The local eigenvector of :

[0086] ;

[0087] S34: Constructing the target recognition area The global eigenvector of :

[0088] ;

[0089] S35: Constructing the target recognition area The eigenvector of .

[0090] Optionally, in step S4, similarity calculation of feature vectors corresponding to target recognition areas in different driving scene video frames may include:

[0091] The similarity of the feature vectors corresponding to the target recognition area in different driving scene video frames is calculated, where the feature vector and The similarity calculation process is as follows:

[0092] S41: Calculate eigenvectors and The distance of the residual convolution feature in , where the residual convolution feature and The distance between them is:

[0093] ;

[0094] in:

[0095] represents the L2 norm;

[0096] Represents the residual convolution feature and The distance between

[0097] S42: Calculate the feature vector and The local distance between ,in:

[0098] ;

[0099] in:

[0100] Represents local distance Iteration intermediate variables, ;

[0101] S43: Calculate the characteristic vector and The global distance between :

[0102] ;

[0103] S44: Generate feature vector and Similarity:

[0104] ;

[0105] in:

[0106] Represents the feature vector and The similarity of the feature vectors is the similarity of the target recognition area;

[0107] The target recognition region with a similarity exceeding a preset threshold is regarded as a same-target different-source region, and the vehicle type recognized by the target recognition region in the same-target different-source region is consistent.

[0108] Optionally, in step S5, the same target heterogeneous regions are integrated to form driving test alignment data, including:

[0109] The same target heterogeneous regions are integrated to obtain the image data sequence and vehicle parameter data sequence of the same vehicle model, which constitute the driving test alignment data. The image data sequence set of the e-th vehicle model is:

[0110] ;

[0111] ;

[0112] in:

[0113] Represents a set of image data sequences of the e-th type of vehicle;

[0114] Indicates the timestamp of the e-th vehicle model in the video frame All target identification areas;

[0115] Indicates the timestamp of the mth camera in the video frame The target recognition area of ​​the e-th type of vehicle captured;

[0116] The driving test alignment data is:

[0117] ;

[0118] in:

[0119] represents the driving test alignment data; Represents the vehicle parameter data sequence corresponding to the e-th vehicle model.

[0120] In order to solve the above problem, the present invention provides an electronic device, the electronic device comprising:

[0121] A memory storing at least one instruction;

[0122] Communication interface, enabling electronic equipment to communicate; and

[0123] The processor executes the instructions stored in the memory to implement the above-mentioned intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario.

[0124] In order to solve the above problems, the present invention also provides a computer-readable storage medium, which stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario.

[0125] Compared with the prior art, the present invention proposes a method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models. The technology has the following advantages:

[0126] Firstly, this scheme adaptively corrects the grayscale value of the pixel in combination with the brightness of the pixel in the video frame to obtain the preprocessed driving scene video data. The background score of the pixel is calculated according to the difference between the pixels at the same position in different video frames. The grayscale frame is converted into a background score frame, and convolution calculation is performed to obtain the initial border of different target areas in the driving scene video frame. The pixels in the initial border are vertically mapped to form 9 sampling pixel positions of the pixel, and the mapping value of the pixels in the border is obtained by convolution calculation. The initial border is corrected to obtain several target recognition areas in the driving scene video frame, thereby realizing target recognition of different models in a mixed scene of multiple models.

[0127] At the same time, this scheme combines the multi-scale residual convolution method and the attention mechanism to extract features from the target recognition area, obtain the local feature vector and the global feature vector of the target recognition area, and calculate the distance between the feature vectors to obtain the local distance and the global distance of the target recognition area in the video frames of different driving scenes. The local distance and the global distance are converted into the similarity between the feature vectors, and the target recognition area with a similarity exceeding a preset threshold is regarded as the same-target heterogeneous area, so as to realize clustering and integration processing of images of the same model under different perspectives to form the same-target heterogeneous area, and integrate the same-target heterogeneous area to obtain the image data sequence and the vehicle parameter data sequence of the same model, so as to constitute the driving test alignment data of the same model in the mixed scene of multiple models and multiple perspectives. BRIEF DESCRIPTION OF THE DRAWINGS

[0128] Figure 1 A flowchart of a method for collecting and aligning intelligent driving test data in a multi-vehicle mixed scenario provided by an embodiment of the present invention;

[0129] Figure 2 A schematic diagram of the structure of an electronic device for implementing a method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models provided by an embodiment of the present invention;

[0130] In the figure: 1 electronic device, 10 processor, 11 memory, 12 program, 13 communication interface.

[0131] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0132] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0133] The embodiment of the present application provides a method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models. The execution subject of the method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiment of the present application. In other words, the method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0134] S1: collecting driving scene video data through a camera, collecting a vehicle parameter data sequence through a sensor, and preprocessing the driving scene video data to obtain preprocessed driving scene video data, wherein the driving scene video data consists of a driving scene video frame and a video frame timestamp.

[0135] In the step S1, the driving scene video data is collected by a camera, and the vehicle parameter data sequence is collected by a sensor, including:

[0136] The driving scene video data is collected by the camera, wherein the driving scene video data consists of driving scene video frames and video frame timestamps. for:

[0137] ;

[0138] ;

[0139] in:

[0140] represents the driving scene video data captured by the mth camera, , M represents the total number of cameras in the driving scene;

[0141] Represents driving scene video data The nth driving scene video frame in , N represents the number of driving scene video frames in the driving scene video data, Indicates the video frame timestamp corresponding to the nth driving scene video frame;

[0142] Preprocessing the driving scene video data to obtain preprocessed driving scene video data;

[0143] The vehicle parameter data sequence is collected by using sensors, wherein the collected vehicle parameter data sequence set is:

[0144] ;

[0145] in:

[0146] C represents the set of vehicle parameter data sequences;

[0147] represents the vehicle parameter data sequence corresponding to the e-th vehicle model, Indicates that the e-th model is at timestamp The vehicle parameter data includes the vehicle's instantaneous speed, acceleration, yaw angular velocity, four-wheel lateral force, four-wheel steering angle and four-wheel drive torque.

[0148] The preprocessing of the driving scene video data to obtain the preprocessed driving scene video data includes:

[0149] The driving scene video data is preprocessed to obtain preprocessed driving scene video data, wherein the driving scene video data The preprocessing process is:

[0150] S11: Acquiring driving scene video data All driving scene video frames in;

[0151] S12: grayscale processing is performed on any pixel in the driving scene video frame, wherein the driving scene video frame The pixel at row x and column y The grayscale processing formula is:

[0152] ;

[0153] in:

[0154] Represents pixels Gray value of

[0155] In order, it represents pixels Color values ​​in R, G, B color channels;

[0156] S13: Perform brightness correction on the gray value of the pixel, where the gray value The brightness correction formula is:

[0157] ;

[0158] S14: The pixels after brightness correction are used to form a preprocessing result of a driving scene video frame, wherein the driving scene video frame The preprocessing results are:

[0159] ;

[0160] in:

[0161] Represents a driving scene video frame The preprocessing results are It is in the form of a matrix of X rows and Y columns, where X represents the number of pixel rows in the driving scene video frame, and Y represents the number of pixel columns in the driving scene video frame.

[0162] S2: Construct an adaptive bounding box target recognition model to perform target recognition on the driving scene video frame to obtain the target recognition area of ​​the driving scene video frame.

[0163] The target recognition model with adaptive border is constructed in step S2, including:

[0164] Constructing an adaptive border target recognition model, wherein the target recognition model takes the preprocessed driving scene video frame as input and takes the target recognition area in the driving scene video frame as output, wherein the adaptive border target recognition model includes an input layer, a background score calculation layer, an initial border generation layer, an alignment convolution layer, and an output layer;

[0165] The input layer is used to receive the preprocessed driving scene video frames;

[0166] The background score calculation layer is a convolutional neural network structure, which is used to calculate the background score of any pixel in the driving scene video frame;

[0167] The initial border generation layer performs target-background classification based on the background score of the pixel and generates the initial border of the target area;

[0168] The alignment convolution layer is used to map and sample the pixels in the initial border, capture the geometric information and context information of the border, and correct the initial border;

[0169] The output layer is used to extract the image area in the corrected bounding box as the target recognition area of ​​the driving scene video frame.

[0170] The method of using the target recognition model to perform target recognition on the driving scene video frame includes:

[0171] The target recognition model is used to perform target recognition on the driving scene video frame, where the driving scene video frame is preprocessed The target recognition process is:

[0172] S21: The input layer receives the preprocessed driving scene video frames ;

[0173] S22: Background score calculation layer calculates driving scene video frames The background score of any pixel in the driving scene video frame The background score of the pixel at row x and column y in is:

[0174] ;

[0175] in:

[0176] Represents a driving scene video frame The background score of the pixel in the xth row and yth column;

[0177] Indicates the grayscale change threshold of adjacent video frames, Indicates the grayscale change threshold of the initial video frame;

[0178] represents the adjustment coefficient; in the embodiment of the present invention, Set to 0.01;

[0179] represents an exponential function with a natural constant as base;

[0180] S23: The initial border generation layer performs target-background classification based on the background score of the pixel, obtains the probability that the pixel is the target pixel, and generates an initial border containing the target pixel, where the driving scene video frame The probability that the pixel in the xth row and yth column is the target pixel is:

[0181] ;

[0182] in:

[0183] Represents a driving scene video frame The probability that the pixel in the xth row and yth column is the target pixel;

[0184] Represents a driving scene video frame The pixel at row x and column y is the center The mean background score of the neighborhood pixel area;

[0185] Composing driving scene video frames The target pixel probability matrix: ;

[0186] Using the object detection convolution matrix Perform convolution calculation on the target pixel probability matrix and get The coordinates of the border vertices of the group target area, where Represents a driving scene video frame The number of targets in , where the coordinates of the border vertices of the k-th target area are:

[0187] ;

[0188] in:

[0189] Represents the coordinates of the border vertices of the kth group of target areas, Represents the coordinates of the four vertices of the border;

[0190] In an embodiment of the present invention, the pixel coordinates in the convolution result of the target pixel probability matrix that are higher than a preset threshold are used as a vertex position coordinate, the two nearest vertex position coordinates are used as diagonal coordinates in the vertex position coordinates of the border, and the four vertex position coordinates of the border are generated. If the target pixel ratio in the four vertex position coordinates is higher than the preset threshold, the border is retained, otherwise it is not retained.

[0191] S24: The aligned convolution layer is used to map and sample the pixels in the initial border, where the border vertex position coordinates The pixel position in the frame The corresponding sampling pixel position is , ,

[0192] , , , , , ;in The pixel positions are The distance to the left, top, right, and bottom of the border;

[0193] Based on sampling pixel positions in driving scene video frames The 9 pixels obtained by sampling are used as the neighborhood pixel area of ​​the pixels in the initial border, and the border weight matrix is ​​used to calculate the neighborhood pixel area of ​​the pixels in the initial border. Perform convolution processing on the neighborhood pixel area to obtain the mapping value of the neighborhood pixel area;

[0194] S25: Correcting the initial border so that the average mapping value in the border is higher than a preset mapping threshold;

[0195] S26: Extract the image area in the corrected border as the driving scene video frame target identification area.

[0196] S3: Extract features of the target recognition area in different driving scene video frames to obtain feature vectors of the target recognition area.

[0197] In the step S3, feature extraction is performed on the target recognition area in the video frames of different driving scenes, including:

[0198] The target recognition area in different driving scene video frames is feature extracted to obtain the feature vector of the target recognition area, wherein the cross-vision transformer is the implementation method of the feature extraction, and the target recognition area The feature extraction process is:

[0199] S31: Target recognition area Perform residual convolution calculation to obtain the target recognition area Multi-scale residual convolution features:

[0200] ;

[0201] in:

[0202] Indicates the target recognition area The residual convolution feature at scale z, , Z represents the maximum scale;

[0203] represents the convolution operator;

[0204] express The pixel-sized residual convolution matrix, express The pixel-sized residual convolution matrix;

[0205] S32: Calculate the cross attention of residual convolution features at different scales, where the residual convolution features The cross attention is:

[0206] ;

[0207] in:

[0208] Represents the residual convolution feature Cross-attention;

[0209] represents an activation function; in the embodiment of the present invention, the selected activation function is a ReLU function;

[0210] represents the attention convolution weight matrix;

[0211] S33: Constructing the target recognition area The local eigenvector of :

[0212] ;

[0213] S34: Constructing the target recognition area The global eigenvector of :

[0214] ;

[0215] S35: Constructing the target recognition area The eigenvector of .

[0216] S4: Calculate the similarity of feature vectors corresponding to target recognition areas in different driving scene video frames, and regard target recognition areas with similarity exceeding a preset threshold as same-target different-source areas.

[0217] In the step S4, similarity calculation is performed on the feature vectors corresponding to the target recognition areas in the different driving scene video frames, including:

[0218] The similarity of the feature vectors corresponding to the target recognition area in different driving scene video frames is calculated, where the feature vector and The similarity calculation process is as follows:

[0219] S41: Calculate eigenvectors and The distance of the residual convolution feature in , where the residual convolution feature and The distance between them is:

[0220] ;

[0221] in:

[0222] represents the L2 norm;

[0223] Represents the residual convolution feature and The distance between

[0224] S42: Calculate the feature vector and The local distance between ,in:

[0225] ;

[0226] in:

[0227] Represents local distance Iteration intermediate variables, ;

[0228] S43: Calculate the characteristic vector and The global distance between :

[0229] ;

[0230] S44: Generate feature vector and Similarity:

[0231] ;

[0232] in:

[0233] Represents the feature vector and The similarity of the feature vectors is the similarity of the target recognition area;

[0234] The target recognition region with a similarity exceeding a preset threshold is regarded as a same-target different-source region, and the vehicle type recognized by the target recognition region in the same-target different-source region is consistent.

[0235] S5: Integrate the same target heterogeneous regions to obtain the image data sequence and vehicle parameter data sequence of the same vehicle model to form the driving test alignment data.

[0236] In step S5, the same target heterogeneous regions are integrated to form driving test alignment data, including:

[0237] The same target heterogeneous regions are integrated to obtain the image data sequence and vehicle parameter data sequence of the same vehicle model, which constitute the driving test alignment data. The image data sequence set of the e-th vehicle model is:

[0238] ;

[0239] ;

[0240] in:

[0241] Represents a set of image data sequences of the e-th type of vehicle;

[0242] Indicates the timestamp of the e-th vehicle model in the video frame All target identification areas;

[0243] Indicates the timestamp of the mth camera in the video frame The target recognition area of ​​the e-th type of vehicle captured;

[0244] The driving test alignment data is:

[0245] ;

[0246] in:

[0247] represents the driving test alignment data; Represents the vehicle parameter data sequence corresponding to the e-th vehicle model.

[0248] like Figure 2 , is a schematic diagram of the structure of an electronic device for implementing a method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models provided by an embodiment of the present invention.

[0249] The electronic device 1 may include a processor 10 , a memory 11 , a communication interface 13 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10 , such as a program 12 .

[0250] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 may not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the program 12, etc., but also be used to temporarily store data that has been output or is to be output.

[0251] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as the program 12 for realizing intelligent driving test data collection and alignment in a multi-vehicle mixed scenario), and calls the data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0252] The communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices, and to achieve connection and communication between internal components of the electronic device.

[0253] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0254] Figure 2 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 2 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0255] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0256] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0257] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0258] The program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0259] The driving scene video data is collected by a camera, and a vehicle parameter data sequence is collected by a sensor, and the driving scene video data is preprocessed to obtain the preprocessed driving scene video data;

[0260] An adaptive bounding box target recognition model is constructed to perform target recognition on driving scene video frames, and a target recognition area of ​​the driving scene video frames is obtained;

[0261] Extract features of target recognition areas in video frames of different driving scenes to obtain feature vectors of target recognition areas;

[0262] The similarity of the feature vectors corresponding to the target recognition areas in the video frames of different driving scenes is calculated, and the target recognition areas with similarity exceeding a preset threshold are regarded as same-target different-source areas;

[0263] The same target heterogeneous regions are integrated to obtain the image data sequence and vehicle parameter data sequence of the same vehicle model, forming the driving test alignment data.

[0264] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figure 1 to Figure 2 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0265] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0266] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0267] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for collecting and aligning intelligent driving test data in a mixed scenario of multiple vehicle models, characterized in that: The method comprises: S1: collecting driving scene video data through a camera, collecting a vehicle parameter data sequence through a sensor, and preprocessing the driving scene video data to obtain preprocessed driving scene video data, wherein the driving scene video data consists of a driving scene video frame and a video frame timestamp; The driving scene video data consists of driving scene video frames and video frame timestamps. for: ； in: represents the driving scene video data captured by the mth camera, , M represents the total number of cameras in the driving scene; Represents driving scene video data The nth driving scene video frame in , N represents the number of driving scene video frames in the driving scene video data, Indicates the video frame timestamp corresponding to the nth driving scene video frame; Preprocessing the driving scene video data to obtain preprocessed driving scene video data; The vehicle parameter data sequence is collected by using sensors, wherein the collected vehicle parameter data sequence set is: ； in: C represents the set of vehicle parameter data sequences; represents the vehicle parameter data sequence corresponding to the e-th vehicle model, Indicates that the e-th type of vehicle model is at timestamp The vehicle parameter data includes the instantaneous speed, acceleration, yaw rate, four-wheel lateral force, four-wheel steering angle and four-wheel drive torque of the vehicle; S2: construct an adaptive bounding box target recognition model to perform target recognition on the driving scene video frame, and obtain the target recognition area of ​​the driving scene video frame; S3: extracting features of target recognition areas in different driving scene video frames to obtain feature vectors of the target recognition areas; S4: Calculate the similarity of feature vectors corresponding to target recognition areas in different driving scene video frames, and regard target recognition areas with similarity exceeding a preset threshold as same-target different-source areas; S5: Integrate the same target heterogeneous regions to obtain the image data sequence and vehicle parameter data sequence of the same vehicle model to form the driving test alignment data.

2. The intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario as claimed in claim 1 is characterized in that: The preprocessing of the driving scene video data to obtain the preprocessed driving scene video data includes: The driving scene video data is preprocessed to obtain preprocessed driving scene video data, wherein the driving scene video data The preprocessing process is: S11: Acquiring driving scene video data All driving scene video frames in; S12: grayscale processing is performed on any pixel in the driving scene video frame, wherein the driving scene video frame The pixel at row x and column y The grayscale processing formula is: ； in: Represents pixels Gray value of In order, it represents pixels Color values ​​in R, G, B color channels; S13: Perform brightness correction on the gray value of the pixel, where the gray value The brightness correction formula is: ； S14: The pixels after brightness correction are used to form a preprocessing result of a driving scene video frame, wherein the driving scene video frame The preprocessing results are: in: Represents a driving scene video frame The preprocessing results are It is in the form of a matrix of X rows and Y columns, where X represents the number of pixel rows in the driving scene video frame, and Y represents the number of pixel columns in the driving scene video frame.

3. The intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario as claimed in claim 1 is characterized in that: The target recognition model with adaptive border is constructed in step S2, including: Constructing an adaptive border target recognition model, wherein the target recognition model takes the preprocessed driving scene video frame as input and takes the target recognition area in the driving scene video frame as output, wherein the adaptive border target recognition model includes an input layer, a background score calculation layer, an initial border generation layer, an alignment convolution layer, and an output layer; The input layer is used to receive the preprocessed driving scene video frames; The background score calculation layer is a convolutional neural network structure, which is used to calculate the background score of any pixel in the driving scene video frame; The initial border generation layer performs target-background classification based on the background score of the pixel and generates the initial border of the target area; The alignment convolution layer is used to map and sample the pixels in the initial border, capture the geometric information and context information of the border, and correct the initial border; The output layer is used to extract the image area in the corrected bounding box as the target recognition area of ​​the driving scene video frame.

4. The intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario as claimed in claim 3 is characterized in that: Use the target recognition model to perform target recognition on driving scene video frames, including: The target recognition model is used to perform target recognition on the driving scene video frame, where the driving scene video frame is preprocessed The target recognition process is: S21: The input layer receives the preprocessed driving scene video frames ; S22: Background score calculation layer calculates driving scene video frames The background score of any pixel in the driving scene video frame The background score of the pixel in row x and column y is: in: Represents a driving scene video frame The background score of the pixel in row x and column y; Indicates the grayscale change threshold of adjacent video frames, Indicates the grayscale change threshold of the initial video frame; represents the adjustment coefficient; represents an exponential function with a natural constant as base; S23: The initial border generation layer performs target-background classification based on the background score of the pixel, obtains the probability that the pixel is the target pixel, and generates an initial border containing the target pixel, where the driving scene video frame The probability that the pixel in the xth row and yth column is the target pixel is: in: Represents a driving scene video frame The probability that the pixel in the xth row and yth column is the target pixel; Represents a driving scene video frame The pixel at row x and column y is the center The mean background score of the neighborhood pixel area; Composing driving scene video frames The target pixel probability matrix: ; Using the object detection convolution matrix Perform convolution calculation on the target pixel probability matrix and get The coordinates of the border vertices of the group target area, where Represents a driving scene video frame The number of targets in , where the coordinates of the border vertices of the k-th target area are: in: Represents the coordinates of the border vertices of the kth group of target areas, Represents the coordinates of the four vertices of the border; S24: The aligned convolution layer is used to map and sample the pixels in the initial border, where the border vertex position coordinates The pixel position in the frame The corresponding sampling pixel position is , , , , , , , ;in The pixel positions are The distance to the left, top, right, and bottom of the border; Based on the sampling pixel position in the driving scene video frame The 9 pixels obtained by sampling are used as the neighborhood pixel area of ​​the pixels in the initial border, and the border weight matrix is ​​used to calculate the neighborhood pixel area of ​​the pixels in the initial border. Perform convolution processing on the neighborhood pixel area to obtain the mapping value of the neighborhood pixel area; S25: Correcting the initial border so that the average mapping value in the border is higher than a preset mapping threshold; S26: Extract the image area in the corrected border as the driving scene video frame target identification area.

5. The intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario as claimed in claim 4 is characterized in that: In the step S3, feature extraction is performed on the target recognition area in the video frames of different driving scenes, including: The feature vector of the target recognition area is obtained by extracting the features of the target recognition area in the video frames of different driving scenes. The feature extraction process is: S31: Target recognition area Perform residual convolution calculation to obtain the target recognition area Multi-scale residual convolution features: in: Indicates the target recognition area The residual convolution feature at scale z, , Z represents the maximum scale; represents the convolution operator; express The pixel-sized residual convolution matrix, express The pixel-sized residual convolution matrix; S32: Calculate the cross attention of residual convolution features at different scales, where the residual convolution features The cross attention is: ； in: Represents the residual convolution feature Cross-attention; represents the activation function; represents the attention convolution weight matrix; S33: Constructing the target recognition area The local eigenvector of : ； S34: Constructing the target recognition area The global eigenvector of : ； S35: Constructing the target recognition area The eigenvector of .

6. The intelligent driving test data collection and alignment method in a multi-vehicle mixed scenario as claimed in claim 5 is characterized in that: In the step S4, similarity calculation is performed on the feature vectors corresponding to the target recognition areas in the different driving scene video frames, including: The similarity of the feature vectors corresponding to the target recognition area in different driving scene video frames is calculated, where the feature vector and The similarity calculation process is as follows: S41: Calculate eigenvectors and The distance of the residual convolution feature in , where the residual convolution feature and The distance between them is: ； in: represents the L2 norm; Represents the residual convolution feature and The distance between S42: Calculate the feature vector and The local distance between ,in: ； in: Represents local distance Iteration intermediate variables, ; S43: Calculate the characteristic vector and The global distance between : ； S44: Generate feature vector and Similarity: ； in: Represents the feature vector and The similarity of the feature vectors is the similarity of the target recognition area; The target recognition region with a similarity exceeding a preset threshold is regarded as a same-target different-source region, and the vehicle type recognized by the target recognition region in the same-target different-source region is consistent.

7. The method for collecting and aligning intelligent driving test data in a multi-vehicle mixed scenario as claimed in claim 6, characterized in that: In step S5, the same target heterogeneous regions are integrated to form driving test alignment data, including: The same target heterogeneous regions are integrated to obtain the image data sequence and vehicle parameter data sequence of the same vehicle model, which constitute the driving test alignment data. The image data sequence set of the e-th vehicle model is: ； in: Represents a set of image data sequences of the e-th type of vehicle; Indicates the timestamp of the e-th vehicle model in the video frame All target identification areas; Indicates the timestamp of the mth camera in the video frame The target recognition area of ​​the e-th type of vehicle captured; The driving test alignment data is: ； in: represents the driving test alignment data; Represents the vehicle parameter data sequence corresponding to the e-th vehicle model.

Images