Vehicle video detection method, vehicle video detection device and readable storage medium

By using the detection model of the three-dimensional grouped convolution layer and the long and short-term memory network layer in vehicle video detection, the characteristics and sequence information of the vehicle video file are extracted, and automated video verification and detection are realized, and the realization of vehicle video in the prior art is solved. The safety hazards and low efficiency of sexual testing have improved the accuracy and efficiency of the registration process.

CN111859018BActive Publication Date: 2024-11-19BEIJING DIDI INFINITY TECH & DEV CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010071919.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-01-21
Publication Date
2024-11-19
Estimated Expiration
2040-01-21

AI Technical Summary

Technical Problem

The lack of effective vehicle video authenticity detection solutions in the prior art leads to potential safety hazards in the online car-hailing registration process, and video fake verification requires human-computer interaction and low efficiency.

Method used

Using detection models including three-dimensional packet convolutional layer, two-dimensional feature extraction layer, long and short-term memory network layer and classification layer, the feature information and sequence information in the vehicle video file are extracted, and false detection is automatically performed to achieve human-computer interaction without human-computer interaction. video verification.

Benefits of technology

It improves the accuracy and efficiency of vehicle video verification results, reduces the calculation amount and hardware requirements, saves manpower and material resources, and improves the user registration experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111859018B_ABST
    Figure CN111859018B_ABST
Patent Text Reader

Abstract

The present invention provides a vehicle video detection method, a vehicle video detection device and a readable storage medium. The vehicle video detection method includes: obtaining a vehicle video file and inputting the vehicle video file into a detection model; extracting feature information and sequence information corresponding to the feature information in the vehicle video file through the detection model, and determining the authentication result of the vehicle video file according to the feature information and the sequence information. The present invention can effectively extract the sequence information in the vehicle video file through the detection model structure, and determine the authentication result of the vehicle video file through the sequence information and the feature information, thereby improving the accuracy of the vehicle authentication result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of video detection technology, and in particular to a vehicle video detection method, a vehicle video detection device and a computer-readable storage medium. Background Art

[0002] In online car-hailing registration, the verification of vehicle information is very critical. In the verification of vehicle information, the vehicle video contains the most information, which is particularly important for determining the vehicle information provided by the owner. If the vehicle information provided by the owner is not true, there may be potential safety hazards in the owner's future service process. In order to ensure the authenticity of the vehicle information uploaded by the owner, the vehicle information is verified through the vehicle video. Video verification in related technologies requires human-computer interaction and is not suitable for vehicle verification scenarios; in video-based applications, current research mainly focuses on tracking targets in videos, behavior and other content analysis, and there is no video verification-related solution. Therefore, how to detect the authenticity of vehicle videos has become an urgent problem to be solved. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art or related art.

[0004] To this end, a first aspect of the present invention provides a vehicle video detection method.

[0005] A second aspect of the present invention provides a vehicle video detection device.

[0006] A third aspect of the present invention provides a computer-readable storage medium.

[0007] In order to achieve the above-mentioned purpose, according to the technical solution of the first aspect of the present invention, a vehicle video detection method is provided, including: obtaining a vehicle video file, and inputting the vehicle video file into a detection model; extracting feature information and sequence information corresponding to the feature information in the vehicle video file through the detection model, and determining the authentication result of the vehicle video file according to the feature information and the sequence information.

[0008] In this technical solution, after obtaining the vehicle video file, the vehicle video file is input into the detection model for detection. The detection model includes a three-dimensional grouped convolution layer, a two-dimensional feature extraction layer, a long short-term memory network layer and a classification layer. The feature information in the vehicle video file is extracted by the three-dimensional grouped convolution layer and the two-dimensional feature extraction layer, and the sequence information corresponding to the feature information in the vehicle video file is extracted by the three-dimensional grouped convolution layer, wherein the feature information extracted by the three-dimensional grouped convolution layer is part of the feature information of the video file, and the extracted sequence information is also the local sequence information corresponding to the part of the feature information. Compared with the three-dimensional convolution in the related art, the three-dimensional grouped convolution in this application has the effect of small calculation amount and high model performance. The long short-term memory network layer and the classification layer can analyze the sequence information and feature information to determine the authentication result of the vehicle video file. The sequence information in the vehicle video file can be effectively extracted through the detection model structure, and the authentication result of the vehicle video file is determined by the sequence information and feature information, thereby improving the accuracy of the vehicle authentication result.

[0009] It can be understood that applying the above-mentioned vehicle video detection method to the online car-hailing registration process enables users to complete the video authentication step without human-computer interaction, thereby improving user registration efficiency and further enhancing user experience.

[0010] It can be understood that the vehicle video detection method is executed on the server side, that is, the vehicle video detection method can be configured on the server side. The user uploads the vehicle video file to the server side so that the server side obtains the vehicle video file sent by the user, thereby realizing online detection of the vehicle video file, avoiding manual authentication of the vehicle video file, and saving manpower and material resources.

[0011] In any of the above technical solutions, the step of extracting feature information and sequence information corresponding to the feature information in the vehicle video file specifically includes: intercepting a frame sequence of a set length in the vehicle video file; extracting feature information and sequence information corresponding to the feature information in the frame sequence.

[0012] In this technical solution, when extracting feature information and sequence information of a vehicle video file, a partial frame sequence is intercepted from the acquired vehicle video file, and only the feature information and sequence information of the intercepted partial frame sequence is extracted. There is no need to extract the entire acquired vehicle video file, thereby greatly reducing the amount of calculation and improving the performance of the model.

[0013] It can be understood that the present application utilizes the three-dimensional grouping convolutional layer in the detection model to group and intercept the vehicle video files, extract feature information, and extract sequence information. For example, if the video frame at frame 8 is intercepted, the video frames at frames 6, 7, 9, and 10 will be intercepted at the same time, thereby realizing the interception of a partial frame sequence.

[0014] In any of the above technical solutions, the step of extracting feature information from the frame sequence specifically includes: extracting video features from the frame sequence; converting the video features into two-dimensional features, and extracting feature information from the two-dimensional features.

[0015] In this technical solution, the video features in the frame sequence are first extracted through a three-dimensional grouped convolution layer, and then the video features are converted into two-dimensional features through a two-dimensional feature extraction layer, and the feature information in the two-dimensional features is extracted. Among them, the three-dimensional grouped convolution layer extracts a large number of video features, and paired video features are extracted again through two-dimensional feature extraction, which achieves the goal of reducing the feature amount while ensuring the accuracy of feature information, thereby further improving the performance of the detection model, reducing hardware requirements, and achieving cost savings.

[0016] It can be understood that the two-dimensional feature extraction layer in the detection model contains 60 convolutional layers, and a large number of 3×3, 1×3 and 3×1 convolution kernels are used in the convolutional layers. While ensuring high accuracy, a large amount of network inference time is reduced, thereby improving the efficiency of the detection model.

[0017] In any of the above technical solutions, the step of determining the authentication result of the vehicle video file according to the feature information and the sequence information specifically includes: determining the authentication score of the vehicle video file according to the feature information and the sequence information; and determining the authentication result according to the authentication score.

[0018] In this technical solution, the long short-term memory network layer and the classification layer in the detection model can determine the authentication score of the vehicle video file based on the feature information and sequence information, and determine the authentication result based on the obtained authentication score. The detection model can automatically authenticate the vehicle video file and automatically output the authentication result. The authentication result is determined by the authentication score, which has high accuracy and configurability.

[0019] In any of the above technical solutions, the step of determining the authentication score of the vehicle video file based on the feature information specifically includes: parsing the feature information according to the sequence information to generate the authentication information of the vehicle video file; and classifying the authentication information to generate the authentication score of the vehicle video file.

[0020] In this technical solution, the feature information is parsed through the sequence information through the long short-term memory network layer to generate the authentication information of the vehicle video file, and the authentication information is input into the classification layer for category judgment, thereby generating the authentication score of the vehicle video file.

[0021] It is understandable that the long short-term memory network layer can learn sequence information and convert the sequence information into a basis for judging whether the vehicle video file is fake. The convolutional neural network in the related art has limited learning ability for video sequence information. The long short-term memory network layer is based on the context information of the sequence and can fully learn the relationship between video sequences. The relationship between feature information can be detected based on the sequence information.

[0022] It can be understood that the detection results generated by the long short-term memory network layer are input into the classification layer for classification to obtain a relatively intuitive authentication score.

[0023] In any of the above technical solutions, the step of determining the authentication result according to the authentication score specifically includes: obtaining a preset score; and determining the authentication result of the vehicle video file according to a numerical relationship between the preset score and the authentication score.

[0024] In this technical solution, the preset score is compared with the authentication score generated by the classification layer, the numerical relationship between the preset score and the authentication score generated by the classification layer is determined, and the authentication result of the vehicle video file is determined according to the numerical relationship. By setting the preset score, the authentication standard can be adjusted.

[0025] It is understandable that staff can set preset scores based on actual needs to adjust the authentication standards for vehicle video files.

[0026] In any of the above technical solutions, the step of determining the authentication result of the vehicle video file according to the numerical relationship between the preset score and the authentication score specifically includes: if the authentication score is greater than or equal to the preset score, determining that the vehicle video file is a false video file; if the authentication score is less than the preset score, determining that the vehicle video file is a real video file.

[0027] In this technical solution, when the authentication score is greater than or equal to the preset score, the vehicle video file is considered to be a false video file; when the authentication score is less than the preset score, the vehicle video file is considered to be a real video file.

[0028] It is understandable that after the authentication result is determined according to the preset score, the authentication result is transmitted back to the user end, prompting the user whether it is necessary to re-upload the vehicle video file for re-authentication.

[0029] In any of the above technical solutions, before the step of obtaining the vehicle video file, it also includes: obtaining an initial video file and annotating the initial video file to obtain a video data set; training the video data set to obtain a detection model; the initial video file includes: a real video file and a false video file.

[0030] In this technical solution, before obtaining the vehicle video file, it is necessary to train the detection model online, obtain the initial video file, and annotate the data set in the initial video file. The detection model can be obtained by training the data set, wherein the initial video file includes an equal amount of real video files and false video files, which can ensure that the trained detection model has a good detection accuracy.

[0031] Specifically, 50,000 real video files and 50,000 fake video files can be obtained.

[0032] It is understandable that when annotating a data set, you can choose to use manual annotation to annotate the data set in the initial video file to improve the accuracy of the data set annotation; you can also choose to annotate the data set in the initial video file through automatic annotation by the device to improve the efficiency of data set annotation.

[0033] According to the technical solution of the second aspect of the present invention, there is provided a vehicle video detection device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0034] The computer program is executed by the processor to implement the steps of the vehicle video detection method as described in any one of the above technical solutions; the vehicle video detection device has all the beneficial effects of the vehicle video detection method as described in any one of the above technical solutions, which will not be repeated here.

[0035] According to the technical solution of the third aspect of the present invention, a computer-readable storage medium is provided, on which a control program is stored. When the control program is executed by a processor, the steps of the vehicle video detection method as described in any of the above technical solutions are implemented.

[0036] Additional aspects and advantages of the present invention will be given in part in the following description, and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0038] Figure 1 A schematic diagram showing a process flow of a vehicle video detection method according to an embodiment of the present invention is shown;

[0039] Figure 2 A schematic flow chart of a vehicle video detection method according to another embodiment of the present invention is shown;

[0040] Figure 3 A schematic flow chart of a vehicle video detection method according to another embodiment of the present invention is shown;

[0041] Figure 4 A schematic flow chart of a vehicle video detection method according to another embodiment of the present invention is shown;

[0042] Figure 5 A schematic flow chart of a vehicle video detection method according to another embodiment of the present invention is shown;

[0043] Figure 6 A schematic flow chart of a vehicle video detection method according to another embodiment of the present invention is shown;

[0044] Figure 7 A schematic diagram showing a process flow of a vehicle video detection method according to a specific embodiment of the present invention is shown;

[0045] Figure 8 A schematic flow chart of a vehicle video detection method according to a complete embodiment of the present invention is shown;

[0046] Fig. 9 A schematic flow chart of a vehicle video detection method according to another complete embodiment of the present invention is shown;

[0047] Fig.10 A schematic block diagram of a vehicle video detection device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0048] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0050] Refer to the following Figures 1 to 10 A vehicle video detection method, a vehicle video detection device, and a computer-readable storage medium according to an embodiment of the present invention are described.

[0051] Embodiment 1:

[0052] like Figure 1 As shown, in one embodiment of the present invention, a vehicle video detection method is provided, comprising:

[0053] Step S102, obtaining a vehicle video file, and inputting the vehicle video file into a detection model;

[0054] Step S104, extracting feature information and sequence information corresponding to the feature information from the vehicle video file through the detection model;

[0055] Step S106, determining the authentication result of the vehicle video file according to the feature information and the sequence information through the detection model.

[0056] In this embodiment, after obtaining the vehicle video file, the vehicle video file is input into the detection model for detection. The detection model includes a three-dimensional grouped convolution layer, a two-dimensional feature extraction layer, a long short-term memory network layer and a classification layer. The feature information in the vehicle video file is extracted by the three-dimensional grouped convolution layer and the two-dimensional feature extraction layer. The sequence information corresponding to the feature information in the vehicle video file is extracted by the three-dimensional grouped convolution layer, wherein the feature information extracted by the three-dimensional grouped convolution layer is part of the feature information of the video file, and the extracted sequence information is also the local sequence information corresponding to the part of the feature information. Compared with the three-dimensional convolution in the related art, the three-dimensional grouped convolution in the present application has the effect of small calculation amount and high model performance. The long short-term memory network layer and the classification layer can analyze the sequence information and feature information to determine the authentication result of the vehicle video file. The sequence information in the vehicle video file can be effectively extracted by the detection model structure, and the authentication result of the vehicle video file is determined by the sequence information and feature information, thereby improving the accuracy of the vehicle authentication result. The above-mentioned vehicle video detection method is executed on the server side, that is, the above-mentioned vehicle video detection method can be configured on the server side. By uploading the vehicle video file to the server by the user, the server can obtain the vehicle video file sent by the user, thereby realizing online detection of the vehicle video file, avoiding manual authentication of the vehicle video file, and saving manpower and material resources.

[0057] Among them, step S104 and step S106 are both steps executed by detecting the model.

[0058] It can be understood that applying the above-mentioned vehicle video detection method to the online car-hailing registration process enables users to complete the video authentication step without human-computer interaction, thereby improving user registration efficiency and further enhancing user experience.

[0059] Embodiment 2:

[0060] like Figure 2 As shown, in another embodiment of the present invention, a vehicle video detection method is provided, comprising:

[0061] Step S202, obtaining a vehicle video file, and inputting the vehicle video file into a detection model;

[0062] Step S204, intercepting a frame sequence of a set length from the vehicle video file by using the detection model;

[0063] Step S206, extracting feature information in the frame sequence and sequence information corresponding to the feature information through the detection model;

[0064] Step S208, determining the authentication result of the vehicle video file according to the feature information and sequence information through the detection model.

[0065] In this embodiment, when extracting feature information and sequence information of a vehicle video file, a partial frame sequence is intercepted from the acquired vehicle video file, and only the feature information and sequence information of the intercepted partial frame sequence is extracted. There is no need to extract the entire acquired vehicle video file, thereby greatly reducing the amount of calculation and improving the performance of the model.

[0066] It can be understood that the present application utilizes the three-dimensional grouping convolutional layer in the detection model to group and intercept the vehicle video files, extract feature information, and extract sequence information. For example, if the video frame at frame 8 is intercepted, the video frames at frames 6, 7, 9, and 10 will be intercepted at the same time, thereby realizing the interception of a partial frame sequence.

[0067] like Figure 3 As shown, in the above embodiment, the step of extracting feature information from the frame sequence specifically includes:

[0068] Step S302, extracting video features from the frame sequence;

[0069] Step S304: convert the video features into two-dimensional features, and extract feature information from the two-dimensional features.

[0070] In this embodiment, the video features in the frame sequence are first extracted through the three-dimensional grouping convolution layer, and then the video features are converted into two-dimensional features through the two-dimensional feature extraction layer, and the feature information in the two-dimensional features is extracted. Among them, the three-dimensional grouping convolution layer extracts a large number of video features, and the paired video features are extracted again through the two-dimensional feature extraction, which achieves the goal of reducing the feature amount while ensuring the accuracy of the feature information, thereby further improving the performance of the detection model, reducing the hardware requirements, and achieving the effect of cost saving.

[0071] It can be understood that the two-dimensional feature extraction layer in the detection model contains 60 convolutional layers, and a large number of 3×3, 1×3 and 3×1 convolution kernels are used in the convolutional layers. While ensuring high accuracy, a large amount of network inference time is reduced, thereby improving the efficiency of the detection model.

[0072] Embodiment three:

[0073] like Figure 4 As shown, in another embodiment of the present invention, a vehicle video detection method is provided, comprising:

[0074] Step S402, obtaining a vehicle video file, and inputting the vehicle video file into a detection model;

[0075] Step S404, extracting feature information and sequence information corresponding to the feature information from the vehicle video file through the detection model;

[0076] Step S406, determining the authentication score of the vehicle video file according to the feature information and sequence information through the detection model;

[0077] Step S408: determining the authentication result according to the authentication score through the detection model.

[0078] In this embodiment, the long short-term memory network layer and the classification layer in the detection model can determine the authentication score of the vehicle video file according to the feature information and sequence information, and determine the authentication result according to the obtained authentication score. It is realized that the detection model can automatically authenticate the vehicle video file and automatically output the authentication result, and the authentication result is determined by the authentication score, which has high accuracy and configurability.

[0079] like Figure 5 As shown, in the above embodiment, the step of determining the authentication score of the vehicle video file according to the feature information specifically includes:

[0080] Step S502, parsing the characteristic information according to the sequence information to generate authentication information of the vehicle video file;

[0081] Step S504: classify the authentication information to generate an authentication score for the vehicle video file.

[0082] In this embodiment, the feature information is parsed through the sequence information by the long short-term memory network layer to generate the authentication information of the vehicle video file, and the authentication information is input into the classification layer for category judgment, thereby generating the authentication score of the vehicle video file.

[0083] It is understandable that the long short-term memory network layer can learn sequence information and convert the sequence information into a basis for judging whether the vehicle video file is fake. The convolutional neural network in the related art has limited learning ability for video sequence information. The long short-term memory network layer is based on the context information of the sequence and can fully learn the relationship between video sequences. The relationship between feature information can be detected based on the sequence information.

[0084] It can be understood that the detection results generated by the long short-term memory network layer are input into the classification layer for classification to obtain a relatively intuitive authentication score.

[0085] like Figure 6As shown, in any of the above embodiments, the step of determining the authentication result according to the authentication score specifically includes:

[0086] Step S602, obtaining a preset score;

[0087] Step S604: determining the authentication result of the vehicle video file according to the numerical relationship between the preset score and the authentication score.

[0088] In this embodiment, the preset score is compared with the authentication score generated by the classification layer, the numerical relationship between the preset score and the authentication score generated by the classification layer is determined, and the authentication result of the vehicle video file is determined according to the numerical relationship. By setting the preset score, the authentication standard can be adjusted.

[0089] It is understandable that staff can set preset scores based on actual needs to adjust the authentication standards for vehicle video files.

[0090] Embodiment 4:

[0091] like Figure 7 As shown, in a specific embodiment of the present invention, a vehicle video detection method is provided, comprising:

[0092] Step S702, obtaining a vehicle video file;

[0093] Step S704, intercepting a frame sequence of a set length in the vehicle video file;

[0094] Step S706, extracting video features in the frame sequence and sequence information corresponding to the video features;

[0095] Step S708, converting the video features into two-dimensional features, and extracting feature information from the two-dimensional features;

[0096] Step S710, parsing the characteristic information according to the sequence information to generate authentication information of the vehicle video file;

[0097] Step S712, classifying the authentication information to generate an authentication score for the vehicle video file;

[0098] Step S714, obtaining a preset score;

[0099] Step S716, determining whether the authentication score is less than a preset score, if the determination result is yes, executing step S718, if the determination result is no, executing step S720;

[0100] Step S718, outputting the audit passing information;

[0101] Step S720, output audit failure information.

[0102] In this embodiment, the detection model is deployed in the server, and the user uploads the vehicle video file. The three-dimensional grouping convolution layer in the detection model intercepts part of the video frames, and obtains the frame sequence and sequence information of the part of the video frames, extracts the video features of the frame sequence, and then enters the two-dimensional feature extraction layer. The two-dimensional feature extraction layer can extract the video features into feature information of two-dimensional features, thereby reducing the computational complexity of the server and improving the authentication efficiency. The feature information and sequence information are input into the long short-term memory network layer for analysis, and the authentication information obtained by the analysis is input into the classification layer for classification to obtain an authentication score. The authentication score is compared with the preset score obtained by the preset numerical relationship, thereby determining whether the vehicle video file is a real file.

[0103] Among them, steps S704 to S716 are performed in the detection model.

[0104] Among them, when the authentication score is greater than or equal to the preset score, the vehicle video file is considered to be a false video file; when the authentication score is less than the preset score, the vehicle video file is considered to be a real video file.

[0105] After the authentication result is determined according to the preset score, the authentication result is sent back to the user end, prompting the user whether he needs to re-upload the vehicle video file for re-authentication.

[0106] Embodiment five:

[0107] like Figure 8 As shown, in a complete embodiment of the present invention, a vehicle video detection method is provided, comprising:

[0108] Step S802, reading the vehicle video uploaded by the user;

[0109] Step S804, the model predicts the vehicle video;

[0110] Step S806, determining whether the authentication score is less than a preset score, if the determination result is yes, executing step S808, if the determination result is no, executing step S810;

[0111] Step S808: The review is successful, and the user is prompted that the vehicle video review has passed;

[0112] Step S810: The review fails, and the user is prompted that the vehicle video review has not passed.

[0113] In this embodiment, the vehicle video pre-labeling layer pulls 50,000 real vehicle videos and 50,000 fake vehicle videos for manual labeling. The training model uses the labeled data set to train the detection model offline; the vehicle video uploaded by the user is verified. When used online, the vehicle video file uploaded by the user is transmitted to the server, the model is predicted, and the vehicle video authentication score is output. If the authentication score is higher than the preset score, the review fails and the result is fed back to the user; if the authentication score is less than the threshold, the review passes.

[0114] like Fig. 9 As shown, in the above embodiment, the detection model 900 includes: a video sequence input layer 902, a 3D grouped convolution layer 904 (three-dimensional grouped convolution base layer), a 2D feature extraction layer 906 (two-dimensional feature extraction layer), an LSTM layer 908 (long short-term memory network layer) and a classification layer 910.

[0115] The vehicle video file is input to the detection model 900, and the detection model 900 outputs a counterfeit detection score. InputVideo (input vehicle video file).

[0116] The 3D grouped convolution layer 904 can quickly and effectively extract local sequence information of the video, the 2D feature extraction layer 906 processes the image and sequence, extracts features, and finally the LSTM layer 908 processes the features and sequence information, and finally outputs the fake score of the vehicle video through the classification layer. This model structure can effectively extract the sequence information of the video and improve the accuracy of distinguishing the authenticity of the vehicle video.

[0117] Video sequence input layer, video sequence information is extremely important for determining the authenticity of the video. This paper designs a video sequence reading method. For the input video, a frame sequence of a specified length is randomly intercepted, and then the frame sequence is sent to the 3D grouped convolution layer 904. The 3D grouped convolution layer 904 can extract features from the intercepted video sequence and bring in local sequence information. Compared with the traditional 3D convolution layer, the 3D grouped convolution layer 904 can greatly reduce the amount of calculation, improve the model performance, and effectively extract the sequence information of the video.

[0118] 2D feature extraction layer 906, this application designs an improved 2D feature extraction layer 906 based on resnet50 (neural network), which contains 60 convolutional layers, and a large number of 3×3, 1×3 and 3×1 convolutional kernels are used in the convolutional layers, which reduces a lot of network inference time while ensuring high accuracy. Compared with the original resnet50, the 2D feature extraction layer 906 improves the server performance, thereby saving machine costs.

[0119] LSTM layer 908, sequence information is an important basis for judging whether the vehicle video file is fake. The 3D group convolution layer 904 retains the video sequence information, and the LSTM layer 908 learns the sequence information and converts the sequence information into a basis for judging whether the video is fake. The traditional CNN (convolutional neural network) has limited learning ability for video sequence information. The LSTM layer 908 is based on the context information of the sequence and can fully learn the relationship between video sequences. And it can generate vehicle video authentication information based on sequence information and feature information.

[0120] The classification layer 910 uses a fully connected layer to classify the information output by the LSTM layer 908 to determine whether the input video is a fake video.

[0121] Embodiment six:

[0122] like Fig.10 As shown, in another embodiment of the present invention, a vehicle video detection device 1000 is provided, which includes: a memory 1002 , a processor 1004 , and a computer program stored in the memory 1002 and executable on the processor 1004 .

[0123] The computer program is executed by the processor 1004 to implement the steps of the vehicle video detection method as described in any one of the above technical solutions; the vehicle video detection device has all the beneficial effects of the vehicle video detection method as described in any one of the above technical solutions, which will not be repeated here.

[0124] Embodiment seven:

[0125] In another embodiment of the present invention, a computer-readable storage medium is provided, on which a control program is stored, and when the control program is executed by a processor, the steps of the vehicle video detection method in any of the above embodiments are implemented. The computer-readable storage medium has all the beneficial effects of the vehicle video detection method in any of the above embodiments, which will not be repeated here.

[0126] In the present invention, the terms "first", "second", and "third" are used for descriptive purposes only and should not be understood as indicating or implying relative importance; the term "plurality" refers to two or more, unless otherwise clearly defined. The terms "installed", "connected", "connected", "fixed", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; "connected" can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0127] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by terms such as “upper”, “lower”, “left”, “right”, “front” and “back” are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or unit referred to must have a specific direction, be constructed and operated in a specific orientation, and therefore, cannot be understood as a limitation on the present invention.

[0128] In the description of this specification, the description of the terms "one embodiment", "an embodiment", "a specific embodiment", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0129] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A vehicle video detection method, It is characterized in that include: Obtaining a vehicle video file, and inputting the vehicle video file into a detection model; Extracting feature information and sequence information corresponding to the feature information from the vehicle video file through the detection model, and determining an authentication result of the vehicle video file according to the feature information and the sequence information; The step of extracting the characteristic information and the sequence information corresponding to the characteristic information in the vehicle video file specifically includes: intercepting a frame sequence of a set length in the vehicle video file; Partial feature information in the frame sequence and local sequence information corresponding to the partial feature information are extracted through a three-dimensional grouped convolutional layer.

2. The vehicle video detection method according to claim 1, It is characterized in that The step of extracting part of the feature information in the frame sequence specifically includes: Extracting video features from the frame sequence; The video features are converted into two-dimensional features, and part of feature information in the two-dimensional features is extracted.

3. The vehicle video detection method according to claim 2, It is characterized in that The step of determining the authentication result of the vehicle video file according to the partial feature information and the local sequence information specifically includes: Determine a counterfeit identification score of the vehicle video file according to the partial feature information and the local sequence information; The authentication result is determined according to the authentication score.

4. The vehicle video detection method according to claim 3, It is characterized in that The step of determining the authentication score of the vehicle video file according to the partial feature information specifically includes: Parsing the partial feature information according to the local sequence information to generate authentication information of the vehicle video file; The authentication information is classified to generate an authentication score for the vehicle video file.

5. The vehicle video detection method according to claim 4, It is characterized in that The step of determining the authentication result according to the authentication score specifically includes: Get preset ratings; The authentication result of the vehicle video file is determined according to a numerical relationship between the preset score and the authentication score.

6. The vehicle video detection method according to claim 5, It is characterized in that The step of determining the authentication result of the vehicle video file according to the numerical relationship between the preset score and the authentication score specifically includes: The authentication score is greater than or equal to the preset score, and the vehicle video file is determined to be a false video file; The authentication score is smaller than the preset score, and the vehicle video file is determined to be a real video file.

7. The vehicle video detection method according to any one of claims 1 to 6, It is characterized in that Before the step of obtaining the vehicle video file, the method further includes: Obtaining an initial video file, and annotating the initial video file to obtain a video dataset; Training the video data set to obtain the detection model; The initial video file includes: a real video file and a fake video file.

8. A vehicle video detection device, It is characterized in that include: A memory, a processor, and a computer program stored on the memory and executable on the processor; The computer program is executed by the processor to implement the steps of the vehicle video detection method according to any one of claims 1 to 7.

9. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a vehicle video detection program, and when the vehicle video detection program is executed by a processor, the steps of the vehicle video detection method according to any one of claims 1 to 7 are implemented.