Driving video tamper-proofing detection method based on video frame fingerprints

Abstract

The invention discloses a driving video tamper-proofing detection method based on video frame fingerprints, which relates to the technical field of video security, and comprises the following steps: continuously acquiring original video frame data through a camera, preprocessing each video frame in the acquisition process, and generating video frame data corresponding to each video frame; and calculating the video frame data of each video frame, taking a calculation result as a video frame fingerprint, associating the current video frame fingerprint with the previous video frame fingerprint, generating an association value, and forming a chain type fingerprint structure. According to the driving video tamper-proofing detection method based on the video frame fingerprints, frame-level tamper-proofing detection of the driving video is realized, comprehensive analysis can be carried out on the video frames according to the video frame fingerprints and image structure features, whether the video frames are tampered or not is judged, and the judgment accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of video security technology, and in particular to a method for tamper-proof detection of vehicle video based on video frame fingerprints. Background Technology

[0002] With the development of in-vehicle electronic technology and AI forgery technology, the need to ensure the authenticity of dashcam videos, as the core evidence for traffic accident investigation, is becoming increasingly urgent. Currently, the industry mainly uses three types of technologies for video tamper-proofing and authenticity verification: digital watermarking, traditional video fingerprinting, and blockchain-based evidence storage. However, each of these three technologies has the following drawbacks:

[0003] (1) Digital watermarking technology: This technology achieves copyright protection or tamper detection by embedding invisible or visible identification information (watermark) into video data. The main problem with this technology is that its anti-tampering capability depends on the video data itself. The watermark information is stored inside the video file. Once an attacker processes the video through transcoding, recompression, filtering, or special watermark removal attacks, the watermark may be destroyed or erased and become difficult to detect, making it difficult to find out whether a particular frame in the video has been tampered with.

[0004] (2) Traditional video fingerprinting technology: This technology aims to generate a compact and unique "fingerprint" (or hash value) for video content for content recognition, retrieval or integrity verification. Existing technologies mostly focus on generating a global fingerprint for the entire video or key frame sequence, but this method is difficult to achieve frame-level detection.

[0005] (3) Blockchain and evidence storage technology: This technology uses the immutable and traceable characteristics of blockchain to store the hash value, timestamp and other metadata of video files on a distributed ledger to prove that the file has not been modified as a whole after a certain point in time. However, this detection method can only detect whether the entire video file has been modified as a whole after it has been stored as evidence. It is also difficult to achieve frame-level video anti-tampering detection.

[0006] Therefore, in order to achieve frame-level anti-tampering detection of driving videos, this invention proposes a driving video anti-tampering detection method based on video frame fingerprints to solve the above-mentioned technical problems. Summary of the Invention

[0007] This invention proposes a method for tamper-proof detection of driving video based on video frame fingerprints. It realizes frame-level tamper-proof detection of driving video and can comprehensively analyze video frames based on video frame fingerprints and image structural features to determine whether the video frames have been tampered with, thereby improving the accuracy of the judgment.

[0008] A method for preventing tampering of vehicle video based on video frame fingerprinting includes the following steps:

[0009] The camera continuously captures raw video frame data, and each video frame is preprocessed during the acquisition process to generate video frame data corresponding to each video frame.

[0010] The video frame data of each video frame is calculated, the calculation result is used as the video frame fingerprint, and the current video frame fingerprint is associated with the previous video frame fingerprint to generate an association value. Several association values ​​form a chain fingerprint structure.

[0011] The video frame fingerprint and the associated value are sent to the security chip, and the private key stored inside the security chip is used to digitally sign the video frame fingerprint and the associated value to generate a video frame authentication signature with a timestamp.

[0012] The video frame fingerprint, the chained fingerprint structure, and the video frame authentication signature are encrypted as encrypted data, and the encrypted data is stored in the hardware security module.

[0013] During video playback and uploading, a verification video frame fingerprint is generated for each video frame of the video to be verified. A corresponding verification chain fingerprint structure is generated using the verification video frame fingerprint. The corresponding encrypted data is read from the hardware security module and decrypted and signed using the corresponding public key in the security chip.

[0014] If signature verification fails, the verification video frame fingerprint is verified to be the same as the original video frame fingerprint, and the verification chain fingerprint structure is verified to be the same as the chain fingerprint structure. If one of them is different, the video image is analyzed, and the video is determined to be tampered with based on the analysis results.

[0015] Furthermore, the method for preprocessing each video frame during the acquisition process to generate video frame data corresponding to each video frame is as follows:

[0016] For each video frame, grayscale processing is performed, and the image structure features are extracted. The image structure features include average brightness, brightness standard deviation, gradient magnitude mean, gray-level co-occurrence matrix contrast, HSV saturation mean, and RGB histogram feature values. The above image structure features are then integrated into an intra-frame feature vector.

[0017] For a video frame, the association features with its previous and next frames are calculated, and the association features are integrated into an inter-frame association feature vector. The association features include brightness difference, texture similarity and color consistency.

[0018] The camera's unique identifier, timestamp information, and location information are used as device metadata. The intra-frame feature vector, inter-frame correlation feature vector, and device metadata are fused to form a fused dataset. The fused dataset is used as the video frame data for each video frame.

[0019] Furthermore, the fused dataset is formed as follows:

[0020] The eigenvalues ​​of the intra-frame feature vectors are normalized and converted into an 8-byte floating-point array D1;

[0021] The eigenvalues ​​of the inter-frame correlation feature vector are normalized and converted into an 8-byte floating-point array D2;

[0022] Convert the device metadata into a string byte array D3;

[0023] The fused dataset is obtained by concatenating the byte streams of D1, D2, and D3.

[0024] Furthermore, the chain fingerprint structure is formed as follows:

[0025] A hash calculation is performed on the video frame data of each video frame, and the calculation result is used as the video frame fingerprint of the current video frame;

[0026] The video frame fingerprint of the second frame in the video is concatenated with the video frame fingerprint of the first frame, and then hashed again to obtain the associated hash value of the second frame.

[0027] The method for calculating the associated hash value of video frames starting from the third frame is as follows: the fingerprint of the current video frame is concatenated with the associated hash value of the previous frame and then hashed again to generate the associated hash value of the current frame. Several associated hash values ​​together form the chain fingerprint structure.

[0028] The associated hash value of the first frame of the video is its own video frame fingerprint.

[0029] Furthermore, the method for generating the timestamped video frame authentication signature is as follows:

[0030] The security chip is internally configured with an asymmetric encryption key pair, wherein the asymmetric encryption key pair includes a private key and a public key, and the private key is not exportable and is protected by hardware.

[0031] The video frame fingerprint, timestamp, and chained fingerprint structure are used as the original signature, and the private key is used to generate the video frame authentication signature.

[0032] Furthermore, the decryption and signature verification using the corresponding public key in the security chip are performed as follows:

[0033] The encrypted data of the corresponding frame stored in the hardware security module is read, and the encrypted data is decrypted by the security chip to obtain the video frame fingerprint, the chain fingerprint structure, and the video frame authentication signature.

[0034] The validity of the video frame authentication signature is verified using the public key in the security chip. Then, the fingerprint of each video frame is compared with the verified video frame fingerprint, and the chain fingerprint structure is compared with the verified chain fingerprint structure to determine whether the original video has been tampered with compared with the currently played-back or uploaded video.

[0035] Furthermore, if the video frame authentication signature of a certain frame is invalid, then it is verified whether the video frame fingerprint of that frame, the associated hash value in the chain fingerprint structure, and the verified video frame fingerprint and the associated hash value in the verification chain fingerprint structure are the same.

[0036] When the video frame fingerprints of a certain frame are different, the video frame data of that frame is analyzed;

[0037] The chain fingerprint structure is analyzed frame by frame to determine whether it has been tampered with.

[0038] Furthermore, the method for analyzing the video frame data of this frame is as follows:

[0039] Calculate the intra-frame feature vector and inter-frame correlation feature vector of the frame, obtain the device metadata of the frame, and use the above as real-time data;

[0040] The real-time data is compared with the video frame data stored in the hardware security module. If the device metadata is the same and the differences in intra-frame feature vectors and inter-frame correlation feature vectors are not greater than the threshold, the frame is determined to be unaltered. If the device metadata is different, or the differences in intra-frame feature vectors and inter-frame correlation feature vectors are greater than the threshold, the frame is determined to be tampered with.

[0041] Furthermore, the frame-by-frame analysis is performed as follows:

[0042] Starting from the first frame of the video, the original associated hash value of each frame is compared with the subsequently generated associated hash value until the first frame with a different associated hash value is found. The video frame data of that frame is then analyzed to determine whether the frame has been tampered with.

[0043] Once it is determined that a frame has been tampered with, the original associated hash value of that frame is used as the standard to recalculate the associated hash values ​​of all subsequent frames. If the recalculated associated hash values ​​of the other frames are the same as the original associated hash values, then there are no other tampered frames in the video; if they are different, then frame-by-frame analysis is performed again until all tampered frames in the video are found.

[0044] One or more technical solutions proposed in this application have at least the following technical effects:

[0045] (1) This scheme can perform hash calculation on the video frame data of each video frame to obtain the hash value of each video frame, and use it as the fingerprint of the video frame. Based on this, the associated hash value between two adjacent frames is generated to form a chain fingerprint structure. The above method can not only check the fingerprint of each frame, but also check whether the associated hash value in the chain fingerprint structure has changed. This allows for double checking of each frame in the video in two different ways, which helps to improve the accuracy of the check.

[0046] (2) The video frame fingerprint in this scheme is calculated by fusing intra-frame feature vector, inter-frame correlation feature vector and device metadata. It is related to different aspects of the image data. Therefore, during the inspection process, the above image data can be analyzed to further determine whether the video frame has been tampered with. For a certain frame in the video, even if the attacker tampers with the video frame fingerprint of that frame in the subsequent process, it is difficult to make the image data of that frame completely consistent with the original video frame data. Therefore, by combining the analysis of image data with the comparison of video frame fingerprint in the above way, it helps to make a more accurate judgment.

[0047] (3) The chain fingerprint structure in this scheme is composed of associated hash values. The associated hash value of each frame is calculated based on the associated hash value of the previous frame. Therefore, even if the attacker tampers with the fingerprint of one frame of the video frame, it is difficult to tamper with the associated hash values ​​of all video frames. This makes the chain fingerprint structure more difficult to tamper with, which helps to further improve the accuracy of the inspection. Attached Figure Description

[0048] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0050] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0051] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0052] See Figure 1 This invention proposes a method for tamper-proof detection of vehicle video based on video frame fingerprinting, comprising the following steps:

[0053] The dashcam continuously collects raw video frame data through its camera. During the collection process, each video frame is preprocessed to generate video frame data corresponding to each video frame.

[0054] The video frame data of each video frame is calculated, and the calculation result is used as the video frame fingerprint. The current video frame fingerprint is associated with the previous video frame fingerprint to generate an association value. Several of the association values ​​form a chain fingerprint structure.

[0055] The video frame fingerprint and associated value are sent to the security chip, and the private key stored inside the security chip is used to digitally sign the video frame fingerprint and associated value to generate a video frame authentication signature with a timestamp; in this embodiment, the security chip is integrated into the camera.

[0056] The video frame fingerprint, chain fingerprint structure, and video frame authentication signature are encrypted and stored in the hardware security module as encrypted data.

[0057] During video playback and uploading, a verification video frame fingerprint is generated for each video frame of the video to be verified. The verification video frame fingerprint is used to generate a corresponding verification chain fingerprint structure. The corresponding encrypted data is read from the hardware security module and decrypted and signed using the corresponding public key in the security chip.

[0058] If signature verification fails, the system checks whether the fingerprint of the verified video frame is the same as the fingerprint of the original video frame, and whether the verified chain fingerprint structure is the same as the chain fingerprint structure. If any of these are different, the video image is analyzed, and based on the analysis results, it is determined whether the video has been tampered with.

[0059] This embodiment encrypts the video frame fingerprint, chained fingerprint structure, and video frame authentication signature using a hardware encryption algorithm, specifically the AES-256 symmetric encryption algorithm. The hardware security module mentioned above is an encryption IC that has passed FIPS 140-2 Level 3 security certification. It integrates a physical attack detection grid, voltage / temperature / clock anomaly monitoring circuits, and an active shielding layer, providing physical tamper resistance. This hardware security module only allows access from the security chip through a secure channel (ISO / IEC 7816 protocol), making it difficult for external devices to read directly.

[0060] The collected driving video (video frame data) can be stored on a hard drive or other devices, but not in the hardware security module, so there is a risk of it being tampered with.

[0061] Furthermore, during the acquisition process, each video frame is preprocessed to generate the corresponding video frame data as follows:

[0062] For each video frame, grayscale processing is performed, and image structural features are extracted. These features include average brightness (arithmetic mean of the brightness of all pixels within the frame), brightness standard deviation (the degree of deviation of pixel brightness within the frame from the average brightness), gradient magnitude mean (average gradient intensity of edge pixels within the frame, used to reflect texture clarity), gray-level co-occurrence matrix contrast (the degree of local variation of pixel grayscale values ​​within the frame, reflecting the coarseness or fineness of texture), HSV saturation mean (average color vibrancy within the frame), and RGB histogram feature values ​​(the degree of overlap between the color histograms of the current frame and the preceding and following frames, used to reflect the consistency of color distribution). These image structural features are then integrated into an intra-frame feature vector. The extraction method for these image structural features is common knowledge in this field and will not be detailed here.

[0063] For a video frame, calculate the association features with its previous and next frames, and integrate the association features into an inter-frame association feature vector. The association features include brightness difference (the brightness change amplitude between the current frame and the adjacent frames), texture similarity (the cosine similarity of the gradient features between the current frame and the adjacent frames), and color consistency (the intersection of the color histograms of the current frame and the adjacent frames).

[0064] The camera's unique identifier, timestamp information, and location information are used as device metadata. Intra-frame feature vectors, inter-frame correlation feature vectors, and device metadata are fused to form a fused dataset, which is then used as the video frame data for each video frame.

[0065] Furthermore, the fused dataset is formed as follows:

[0066] The eigenvalues ​​of the intra-frame feature vectors are normalized to the [0,1] interval and converted into an 8-byte floating-point array D1;

[0067] The eigenvalues ​​of the inter-frame correlation feature vector are normalized to the [0,1] interval and converted into an 8-byte floating-point array D2;

[0068] Convert the device metadata into a string byte array D3;

[0069] By concatenating the byte streams of D1, D2, and D3, a fused dataset is obtained.

[0070] Furthermore, the chain fingerprint structure is formed as follows:

[0071] A hash calculation is performed on the video frame data of each video frame, and the calculation result is used as the video frame fingerprint of the current video frame; in this embodiment, the SHA-256 algorithm is used for calculation.

[0072] The video frame fingerprint of the second frame in the video is concatenated with the video frame fingerprint of the first frame, and then hashed again to obtain the associated hash value of the second frame.

[0073] The calculation method for the associated hash value of video frames starting from the third frame is as follows: the fingerprint of the current video frame is concatenated with the associated hash value of the previous frame and then hashed again to generate the associated hash value of the current frame. Several associated hash values ​​together form a chain fingerprint structure.

[0074] The associated hash value of the first frame of the video is its own video frame fingerprint.

[0075] For each frame in the video, the fingerprint verification logic for each frame is as follows: generate the current video frame fingerprint and compare it with the video frame fingerprint stored in the hardware security module. If the comparison results are different, it means that the image content of the current frame in the video has been tampered with. However, there is a very small but possible vulnerability in this case: the attacker tampers with the video frame fingerprint stored in the hardware security module, that is, the original video frame fingerprint. For example: (1) the attacker modifies the image content of a certain frame; (2) the attacker recalculates the video frame fingerprint of the tampered frame; (3) the video frame fingerprint (the original video frame fingerprint) stored in the hardware security module is replaced with the recalculated video frame fingerprint (the video frame fingerprint recalculated after the image content is tampered with). In this way, during subsequent verification, the video frame fingerprint calculated in real time will be exactly the same as the video frame fingerprint stored in the hardware security module, thus misjudging that "the frame has not been tampered with".

[0076] Therefore, to overcome the above situation, this application establishes a chain fingerprint structure, so that each frame in the video not only has a corresponding video frame fingerprint, but also an associated hash value, that is, "another fingerprint containing the associated hash value of the previous frame". Therefore, when an attacker tampers with a frame in the video, during the subsequent verification (detection) process, the associated hash values ​​of other frames after that frame will change, that is, they will be different from the associated hash values ​​stored in the hardware security module. If the attacker wants to tamper with the associated hash values ​​of other frames after that frame, they must recalculate all the associated hash values ​​of other frames after that frame (because the associated hash value of each frame is related to the associated hash value of the previous frame). For hours and tens of thousands of frames of video recorded by a dashcam, it is almost impossible for an attacker to forge the entire chain fingerprint structure. In this way, the difficulty of tampering by attackers is increased, and the accuracy of verification is improved.

[0077] Furthermore, the method for generating a timestamped video frame authentication signature is as follows:

[0078] The security chip is equipped with an asymmetric encryption key pair, which includes a private key and a public key. The private key cannot be exported and is protected by hardware.

[0079] The video frame fingerprint, timestamp, and chained fingerprint structure are used as the original signature, and the private key is used to generate the video frame authentication signature.

[0080] Furthermore, the decryption and signature verification methods using the corresponding public key in the security chip are as follows:

[0081] The encrypted data of the corresponding frame stored in the hardware security module is read, and the encrypted data is decrypted through the security chip to obtain the video frame fingerprint, chain fingerprint structure, and video frame authentication signature.

[0082] The validity of the video frame authentication signature is verified using the public key in the security chip. Then, the fingerprint of each video frame is compared with the verified video frame fingerprint, and the chain fingerprint structure is compared with the verified chain fingerprint structure to determine whether the original video has been tampered with compared with the video being played back or uploaded now.

[0083] Furthermore, when the video frame authentication signature of a certain frame is invalid, the associated hash value in the video frame fingerprint and chain fingerprint structure of that frame is verified to be the same as the associated hash value in the verified video frame fingerprint and chain fingerprint structure.

[0084] When the video frame fingerprints of a certain frame are different, the video frame data of that frame is analyzed;

[0085] The chain fingerprint structure is analyzed frame by frame to determine whether it has been tampered with.

[0086] Dashcam videos are typically stored using compressed encoding formats (such as H.264 and H.265). Intra-frame prediction errors and quantization losses during encoding, or insufficient temporary computing resources during decoding, can all lead to slight deviations in frame data during reading and writing, resulting in changes to the hash value. This deviation is a normal loss at the algorithm level and not due to human tampering. Therefore, to avoid the aforementioned misjudgments, after verifying the video frame fingerprint, it is necessary to also examine the image of the frame to further determine whether the frame has been tampered with. This involves a preliminary judgment (whether the video frame fingerprints are the same) and a more advanced judgment (analyzing the image to determine if it has changed).

[0087] Furthermore, the video frame data of this frame is analyzed as follows:

[0088] Calculate the intra-frame feature vector and inter-frame correlation feature vector of the frame, obtain the device metadata of the frame, and use the above as real-time data;

[0089] The real-time data is compared with the video frame data stored in the hardware security module. If the device metadata is the same and the differences in intra-frame feature vectors and inter-frame correlation feature vectors are not greater than the threshold, the frame is determined to be untampered. If the device metadata is different, or the differences in intra-frame feature vectors and inter-frame correlation feature vectors are greater than the threshold, the frame is determined to be tampered with.

[0090] Since dashcams record continuously, the captured images will change continuously. That is, each frame in the video is related to the previous and next frames. Therefore, this solution introduces inter-frame correlation feature vectors. When a certain frame is significantly different from the images of its predecessor and successor, the image of that frame is very likely to have been tampered with.

[0091] Preferably, if a frame is determined to be tampered with, the above analysis can be performed on the five frames before and after it, which helps to confirm whether the tampering range is a single frame or multiple frames.

[0092] Furthermore, the frame-by-frame analysis method is as follows:

[0093] Starting from the first frame of the video, the original associated hash value of each frame is compared with the subsequently generated associated hash value until the first frame with a different associated hash value is found. The video frame data of that frame is then analyzed to determine whether the frame has been tampered with.

[0094] Once it is determined that a frame has been tampered with, the original associated hash value of that frame is used as the standard to recalculate the associated hash values ​​of all subsequent frames. If the recalculated associated hash values ​​of the other frames are the same as the original associated hash values, then there are no other tampered frames in the video; if they are different, then frame-by-frame analysis is performed again until all tampered frames in the video are found.

[0095] Therefore, the modified frames in the entire video can be identified using the above method. Compared with single-frame fingerprint verification, the chain fingerprint structure relies more heavily on the correlation between adjacent frames, increasing the cost of tampering for attackers. This makes it more difficult for attackers to modify all the associated hash values ​​of the entire video, increasing the difficulty of tampering and making the tampering easier to detect. Those skilled in the art can choose to use single-frame fingerprint verification or a chain fingerprint structure for verification according to actual needs.

[0096] The above description is merely an exemplary embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention specification and drawings under the technical concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for tamper-proof detection of vehicle video based on video frame fingerprinting, characterized in that, Includes the following steps: The camera continuously captures raw video frame data, and each video frame is preprocessed during the acquisition process to generate video frame data corresponding to each video frame. The video frame data of each video frame is calculated, the calculation result is used as the video frame fingerprint, and the current video frame fingerprint is associated with the previous video frame fingerprint to generate an association value. Several association values ​​form a chain fingerprint structure. The video frame fingerprint and the associated value are sent to the security chip, and the private key stored inside the security chip is used to digitally sign the video frame fingerprint and the associated value to generate a video frame authentication signature with a timestamp. The video frame fingerprint, the chained fingerprint structure, and the video frame authentication signature are encrypted as encrypted data, and the encrypted data is stored in the hardware security module. During video playback and uploading, a verification video frame fingerprint is generated for each video frame of the video to be verified. A corresponding verification chain fingerprint structure is generated using the verification video frame fingerprint. The corresponding encrypted data is read from the hardware security module and decrypted and signed using the corresponding public key in the security chip. If signature verification fails, the verification video frame fingerprint is verified to be the same as the original video frame fingerprint, and the verification chain fingerprint structure is verified to be the same as the chain fingerprint structure. If one of them is different, the video image is analyzed, and the video is determined to be tampered with based on the analysis results.

2. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 1, characterized in that, The method for preprocessing each video frame during the acquisition process to generate video frame data corresponding to each video frame is as follows: For each video frame, grayscale processing is performed, and the image structure features are extracted. The image structure features include average brightness, brightness standard deviation, gradient magnitude mean, gray-level co-occurrence matrix contrast, HSV saturation mean, and RGB histogram feature values. The above image structure features are then integrated into an intra-frame feature vector. For a video frame, the association features with its previous and next frames are calculated, and the association features are integrated into an inter-frame association feature vector. The association features include brightness difference, texture similarity and color consistency. The camera's unique identifier, timestamp information, and location information are used as device metadata. The intra-frame feature vector, inter-frame correlation feature vector, and device metadata are fused to form a fused dataset. The fused dataset is used as the video frame data for each video frame.

3. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 1, characterized in that, The fused dataset is formed in the following way: The eigenvalues ​​of the intra-frame feature vectors are normalized and converted into an 8-byte floating-point array D1; The eigenvalues ​​of the inter-frame correlation feature vector are normalized and converted into an 8-byte floating-point array D2; Convert the device metadata into a string byte array D3; The fused dataset is obtained by concatenating the byte streams of D1, D2, and D3.

4. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 2, characterized in that, The chain fingerprint structure is formed as follows: A hash calculation is performed on the video frame data of each video frame, and the calculation result is used as the video frame fingerprint of the current video frame; The video frame fingerprint of the second frame in the video is concatenated with the video frame fingerprint of the first frame, and then hashed again to obtain the associated hash value of the second frame. The method for calculating the associated hash value of video frames starting from the third frame is as follows: the fingerprint of the current video frame is concatenated with the associated hash value of the previous frame and then hashed again to generate the associated hash value of the current frame. Several associated hash values ​​together form the chain fingerprint structure. The associated hash value of the first frame of the video is its own video frame fingerprint.

5. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 2, characterized in that, The method for generating a timestamped video frame authentication signature is as follows: The security chip is internally configured with an asymmetric encryption key pair, wherein the asymmetric encryption key pair includes a private key and a public key, and the private key is not exportable and is protected by hardware. The video frame fingerprint, timestamp, and chained fingerprint structure are used as the original signature, and the private key is used to generate the video frame authentication signature.

6. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 4, characterized in that, The decryption and signature verification methods using the corresponding public key in the security chip are as follows: The encrypted data of the corresponding frame stored in the hardware security module is read, and the encrypted data is decrypted by the security chip to obtain the video frame fingerprint, the chain fingerprint structure, and the video frame authentication signature. The validity of the video frame authentication signature is verified using the public key in the security chip. Then, the fingerprint of each video frame is compared with the verified video frame fingerprint, and the chain fingerprint structure is compared with the verified chain fingerprint structure to determine whether the original video has been tampered with compared with the currently played-back or uploaded video.

7. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 6, characterized in that, If the video frame authentication signature of a certain frame is invalid, then verify whether the video frame fingerprint of that frame, the associated hash value in the chain fingerprint structure, and the verified video frame fingerprint and the associated hash value in the verification chain fingerprint structure are the same. When the video frame fingerprints of a certain frame are different, the video frame data of that frame is analyzed; The chain fingerprint structure is analyzed frame by frame to determine whether it has been tampered with.

8. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 7, characterized in that, The method for analyzing the video frame data of this frame is as follows: Calculate the intra-frame feature vector and inter-frame correlation feature vector of the frame, obtain the device metadata of the frame, and use the above as real-time data; The real-time data is compared with the video frame data stored in the hardware security module. If the device metadata is the same and the differences in intra-frame feature vectors and inter-frame correlation feature vectors are not greater than the threshold, the frame is determined to be unaltered. If the device metadata is different, or the differences in intra-frame feature vectors and inter-frame correlation feature vectors are greater than the threshold, the frame is determined to be tampered with.

9. The method for tamper-proof detection of vehicle video based on video frame fingerprinting according to claim 7, characterized in that, The frame-by-frame analysis method is as follows: Starting from the first frame of the video, the original associated hash value of each frame is compared with the subsequently generated associated hash value until the first frame with a different associated hash value is found. The video frame data of that frame is then analyzed to determine whether the frame has been tampered with. Once it is determined that a frame has been tampered with, the original associated hash value of that frame is used as the standard to recalculate the associated hash values ​​of all subsequent frames. If the recalculated associated hash values ​​of the other frames are the same as the original associated hash values, then there are no other tampered frames in the video; if they are different, then frame-by-frame analysis is performed again until all tampered frames in the video are found.

Images