A method, system and dedicated player for edge-end multi-channel video stream desensitization and reversal

Abstract

The application discloses a kind of edge end multi-channel video stream desensitization and reverse method, system and special player, the original data of multi-channel video stream obtained by edge end is integrated and encoded, face detection is carried out to integrated data, face position and original data position are analyzed, face area data is copied and compressed and encrypted, face position, video stream ID, video stream start frame time, frame number and the like information are encoded, face area in integrated data is desensitized, and the integrated data after desensitization and the face data after encryption are stored, and the integrated data after desensitization is displayed on display terminal.The special player is used, and the integrated data after desensitization and face encryption are reversely operated, and are restored to original video data for playing, to meet the requirement of original video record, the application can reduce the demand of bandwidth and memory in the case of reaching desensitization requirement, and realize security.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, system, and dedicated player for desensitizing and reversing multi-channel video streams at the edge. Background Technology

[0002] Data processing includes cloud, edge, and terminal. The terminal refers to the camera acquisition device, the cloud refers to the remote server, and the edge refers to the processing terminal during video stream transmission.

[0003] Currently, in most urban and rural public areas of my country, such as shopping malls, office buildings, construction sites, roads, and residential communities, surveillance cameras are installed for security reasons. Property management companies in these areas typically install numerous cameras to record dynamic digital video within their jurisdiction. These companies have central control rooms that simultaneously play live feeds from dozens or even hundreds of cameras deployed across various areas. While this facilitates incident handling, it also raises privacy concerns for visitors, employees, and residents. The following are some commonly used solutions to address these issues:

[0004] 1. Regarding surveillance camera recordings, the most common method used by some property management companies is to directly store the raw video data. This method results in facial information not being anonymized, highlighting personal privacy issues.

[0005] 2. To address the issue of surveillance cameras infringing on personal privacy, some property management companies have adopted a dual-encoding and storage method for single-channel video streams, using both un-anonymized and anonymized video streams. This method doubles the amount of data processed during encoding. With the same number of video streams connected, the data transmission bandwidth requirement doubles, as does the storage space requirement. The biggest problem at this point is that it triggers a bandwidth bottleneck. Generally, the actual number of video streams connected can only reach about half of the chip's design capacity. Dual encoding is also performed on surveillance videos without faces, resulting in a significant waste of video encoding, bandwidth, and storage resources.

[0006] In order to protect privacy and save on hardware costs, some property management companies choose to store only anonymized data. While this achieves the goal of anonymizing faces and protecting privacy, it fails to meet the requirements for original video recording in some scenarios that require the use of original video, such as accident retrospective analysis and case handling.

[0007] Therefore, how to achieve desensitization and reversal of multiple video streams while reducing bandwidth and storage requirements is an urgent problem to be solved. Summary of the Invention

[0008] The purpose of this invention is to provide a method, system, and dedicated player for desensitizing and reversing multiple video streams acquired at the edge. The method integrates and encodes the raw data from multiple video streams acquired at the edge, performs face detection on the integrated data, parses the face location and the location of the original data, copies the face region data for compression and encryption, and encodes and encrypts information such as face location, video stream ID, video stream start frame time, and frame number. The face region in the integrated data is desensitized, and the desensitized integrated data and encrypted face data are stored. The desensitized integrated data is then displayed on a display terminal. A dedicated player is used to reverse the desensitized integrated data and face encryption, restoring the original video data for playback, thus meeting the requirements for original video recording. This method can reduce the demand for bandwidth and memory while achieving desensitization requirements and maintaining confidentiality.

[0009] Firstly, the above-mentioned objective of this invention is achieved through the following technical solution:

[0010] A method for desensitizing and reversing multiple video streams at the edge includes: marking each original video stream data; integrating multiple video streams to obtain integrated data; performing face detection in the integrated data; parsing the face location and mapping it to the corresponding location in the original video stream data; copying the face region data for compression and encryption to obtain encrypted data; desensitizing the face region in the original video stream data for use in terminal display; encoding face location-related information to obtain valid data; binding the valid data and encrypted data one-to-one; and performing a reverse operation on the desensitized original video stream data based on the valid data and encrypted data to obtain the recovered original video stream data.

[0011] The present invention is further configured to: perform position encoding on the original video stream data and record the timestamps corresponding to the starting frames of all video streams in the integrated data, for use in real-time face detection and desensitization of multiple video streams.

[0012] The present invention is further configured to: set a face detection threshold, detect faces and bodies in the integrated data, and mark the positions of face bounding boxes and body bounding boxes.

[0013] The present invention is further configured to: calculate the original video stream ID of the area where the face is located based on the face position, calculate the relative coordinates of the face in the original video stream, and calculate the real face position corresponding to the face coordinates mapped back to the original video stream.

[0014] The present invention is further configured to: copy the face region image in the original video stream data, compress it using a lossless compression method, symmetrically encrypt the compressed data, and store the encrypted data.

[0015] The present invention is further configured to: establish a multi-task deep learning neural network model, and use the trained multi-task deep learning neural network model to detect faces and pedestrians in the integrated data; face location-related information includes: the original video stream information where the face is located, the starting frame information, the original video stream frame number where the face is located, and the face location information, and bind the valid data with the encrypted data.

[0016] The present invention is further configured as follows: decoding the desensitized original video stream data, searching for the frame number in the original video stream data according to the valid data, obtaining all face data within the frame number, determining the position corresponding to each face data, decrypting the face data at each position, restoring it to its respective position, and obtaining the restored original video stream data.

[0017] Secondly, the above-mentioned objective of this invention is achieved through the following technical solution:

[0018] A multi-channel video stream desensitization and reversal system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method described in this application.

[0019] Thirdly, the above-mentioned objectives of this invention are achieved through the following technical solutions:

[0020] An edge-end multi-channel video stream desensitization and reversal device includes, in sequence, a multi-channel camera, a mixer, an intelligent central controller, a network video recorder, and a player or dedicated player. The multi-channel camera is used to acquire raw data from multiple video streams. The mixer is used to integrate and encode the raw data from the multiple video streams. The intelligent central controller is used to perform face detection in the integrated data, parse the face position, and map it to the corresponding position in the raw video stream data. The network video recorder is used to store the desensitized video and compressed encrypted face data. The player is used to play the desensitized integrated data. The dedicated player is used to reverse the desensitized integrated data and the encrypted face data, and play the restored raw video stream data.

[0021] Fourthly, the above-mentioned objective of this invention is achieved through the following technical solutions:

[0022] A dedicated player for edge-side multi-channel video streams includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the reversal method described in this application to reverse the desensitized data and obtain the restored original video stream data for playback.

[0023] Compared with the prior art, the beneficial technical effects of this application are as follows:

[0024] 1. This application integrates multiple video streams from the edge, encrypts and stores the face region in the integrated data separately, and performs desensitization processing on the face region in the integrated data, thereby achieving the purpose of protecting privacy;

[0025] 2. Furthermore, this application obtains the position, start frame time, and face position in the original video stream data of each video stream by encoding the original data of multiple video streams, and stores them separately to achieve reverse operation;

[0026] 3. Furthermore, this application calculates the location of the face in the partition and its relative position in the partition by detecting the face position in the integrated data, thereby calculating the real face position when the face is mapped back to the original video stream, so as to ensure that reverse operation can be achieved. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the desensitization and reversal process according to a specific embodiment of this application;

[0028] Figure 2 This is a schematic diagram of the desensitization logic flow of a specific embodiment of this application;

[0029] Figure 3 This is a schematic diagram of the reverse recovery process according to a specific embodiment of this application;

[0030] Figure 4 This is a schematic diagram showing the desensitized display result of a specific embodiment of this application. Detailed Implementation

[0031] The present invention will be further described in detail below with reference to the accompanying drawings.

[0032] This application discloses an edge-end multi-channel video stream desensitization and reversal device, such as... Figure 1 As shown, this includes cameras, mixers, AL-boxes, network video recorders (NVRs), players, or dedicated players.

[0033] The raw video streams acquired by multiple cameras are integrated and encoded by a mixer to obtain integrated data. The intelligent central controller performs pedestrian and face detection on the integrated data, calculates the location of the face in the video stream and its relative position, copies the face region data, establishes a one-to-one correspondence with the face position information, compresses and encrypts the face region data to obtain encrypted data, encodes the face position information to obtain valid data, and demasks the face positions in the video stream to obtain desensitized data. The network recorder stores the encrypted data, valid data, and desensitized data, and the player plays the desensitized data. A dedicated player performs reverse operations on the desensitized data based on the encrypted data and valid data to obtain the restored original video stream and plays it.

[0034] This application discloses a method for desensitizing and reversing multiple video streams at the edge. It obtains the original data of multiple video streams, performs data encoding and integration according to the quantity of the original data of different video streams, and considers the minimum face resolution support. The number of video streams includes multiple combinations such as 4, 9, 16, and 25. The original data of each video stream is positionally encoded.

[0035] The encoding method for the raw data of each video stream includes using two-dimensional coordinates, represented by CID(c, r), where CID represents the video stream ID, c represents the column coordinate, and r represents the row coordinate. Taking the integrated configuration of 16 video streams as an example, CID(3,2) represents the video stream in the 3rd column and the 2nd row.

[0036] After completing the video stream integration encoding, integrated data is obtained. The integrated data includes data from multiple video streams at the same time point, and records the timestamps corresponding to the starting frames of the original data of all video streams in the integrated data.

[0037] The timestamp also uses a two-dimensional coordinate system, denoted by TS(c,r), where TS represents the timestamp, and c and r represent the column and row, respectively.

[0038] In this application, each video stream is independent of the others, and the video stream integration process is a dynamic integration. Without increasing the memory usage of the frame buffer, the application can dynamically integrate each video stream, encode the position information and the initial frame time information. Based on the position information and the initial frame time information, the application can perform traceable data management of the video stream and complete the real-time desensitization and reversal of multiple video data streams.

[0039] For integrated data, a minimum face detection value is set according to the scale of the integrated video stream. The more integrated video streams there are, the smaller the minimum face detection value becomes. A face detection threshold is set to ensure that the lower limit of the face detection size is consistent when the detected face bounding boxes are restored to the original video stream.

[0040] In one specific embodiment of this application, the original video stream resolution is 1080P (1920*1080). The minimum detection capability of the algorithm at this resolution needs to be 160 pixels * 160 pixels. Therefore: in a configuration with 4 video streams integrated, the minimum face size is 160 / 2 = 80 pixels * 80 pixels; in a configuration with 9 video streams integrated, the minimum face size is 160 / 3 = 54 pixels * 54 pixels; in a configuration with 16 video streams integrated, the minimum face size is 160 / 4 = 40 pixels * 40 pixels; in a configuration with 25 video streams integrated, the minimum face size is 160 / 5 = 32 pixels * 32 pixels, and so on.

[0041] A multi-task deep learning neural network model is established to add pedestrian detection capability on the basis of face detection. The model is trained using video data with and without faces to obtain a well-trained multi-task deep learning neural network model. The trained multi-task deep learning neural network model is then used to analyze the integrated data to detect the position information of face bounding boxes and human bounding boxes. The human bounding box is used to limit the position of the face bounding box to reduce the false face detection rate and improve the face detection accuracy.

[0042] The detected face location is marked, the coordinates of the center point of the face bounding box are calculated, the partition location of the center coordinates of the face bounding box is determined, the video stream ID of the face bounding box is determined, the relative coordinate information of the face coordinates within the partition is calculated, and the face coordinates are mapped back to the corresponding real face location in the original video stream.

[0043] In one specific embodiment of this application, the position of the face box is represented by head_box(x,y,w,h), where x,y,w,h represent the x-axis coordinates and y-axis coordinates of the upper left corner of the face box predicted on the integrated image, and the width and height information of the face box, respectively.

[0044] The coordinates of the center point of the face bounding box are: center_x = x + w / 2, center_y = y + h / 2. Based on the coordinates of the center point, the column coordinates of the partition where the face is located are: cid_c = center_x / / sqrt(m), and the row coordinates are: cid_r = center_y / / sqrt(m).

[0045] Here, " / / " indicates quotient calculation, sqrt indicates square root function, and m is the current integration scale, which is equal to the number of video streams, such as one of {4, 9, 16, 25}. After determining the column coordinates cid_c and row coordinates cid_r of the partition where the face is located, the video stream ID of the face box can be determined as location CID(cid_c, cid_r). The starting frame timestamp of the location CID is found, and the frame number of the video frame where the face box is located is calculated by time superposition.

[0046] The relative coordinates of the face within this partition are as follows: the relative x-coordinate is rlt_x = x % (mux_w / sqrt(m)), the relative y-coordinate is rlt_y = y % (mux_h / sqrt(m)), the relative width is rlt_w = w, and the relative height is rlt_h = h, where mux_w represents the width in pixels of the integrated image, mux_h represents the height in pixels of the integrated image, and "%" indicates the remainder.

[0047] The real face coordinates mapped back to the original video stream correspond to the following real face positions: real_x = rlt_x *sqrt(m), real_y = rlt_y * sqrt(m), real_w = rlt_w * sqrt(m), real_h = rlt_h *sqrt(m).

[0048] The face region image in the original video stream data is copied and compressed using a lossless compression method. The compressed data is then symmetrically encrypted to obtain encrypted data. This ensures that the data information is not lost while the data size is significantly reduced, and the original data can be restored through inverse transformation.

[0049] Lossless compression methods include LZ77 and Huffman; encryption methods include base64.

[0050] The location CID, the timestamp of the starting frame of the video stream where the location CID is located, the frame number where the face frame is located, and the face location information are used as face location-related information, encoded to obtain valid data, and then bound one-to-one with the encrypted data.

[0051] After encryption and binding are completed, the face area in the original video stream data is desensitized to obtain a desensitized video stream, achieving the desensitization effect. Desensitization methods include mosaicking or blurring the face area.

[0052] Playing the de-identified video stream displays the de-identified video information, thus achieving privacy protection and resolving the conflict between surveillance camera recordings and facial privacy.

[0053] The de-identified video stream is digitally stored, and the encrypted data and the corresponding valid data are stored synchronously to facilitate reverse operation during subsequent recovery.

[0054] By performing lossless compression and symmetric encryption on the face region and then binding it with face location-related information, the demand for bandwidth and storage space is greatly reduced.

[0055] like Figure 2 As shown, the desensitization process includes the following steps:

[0056] S1. Access raw data from multiple video streams;

[0057] S2. Integrate the raw data from multiple video streams and encode the location information to obtain integrated data;

[0058] S3. Extract frames from the integrated data and detect pedestrians and faces in the frame images;

[0059] S4. Determine if a face is detected. If yes, proceed to the next step. If no, switch to the next frame and proceed to S3.

[0060] S5. Parse the face location information and map it to the corresponding position in the original video stream data;

[0061] S6. Copy the face region data from the video stream and perform lossless compression;

[0062] S7. Perform symmetric encryption on the compressed face data to obtain encrypted data;

[0063] S8. Encode the video stream ID, start frame timestamp, all frame numbers of the face position, and face position information to obtain valid data;

[0064] S9. Desensitize the face region in the original video stream data to obtain a desensitized video stream;

[0065] S10. Bind valid data and encrypted data, and store them together with the desensitized video stream.

[0066] S11. Play the desensitized video stream.

[0067] The de-identified video stream is reversed. The de-identified video ID is obtained from the de-identified video stream, the corresponding frame number is found, and the face data at different positions within the frame number is determined. The face data at each position is symmetrically decrypted, and the decrypted face data is losslessly decompressed to restore the original data of the face region. The video stream data of the de-identified video stream is decoded. The face data at different positions is restored to their corresponding positions. The same restoration work is performed on all frames in the video stream to obtain the restored original video stream data. The restored original video stream data is then played.

[0068] like Figure 3 As shown, reversing a de-identified video stream involves the following steps:

[0069] A1. Obtain de-identified video stream data;

[0070] A2. Determine the video stream ID;

[0071] A3. Decode the video stream;

[0072] A4. Based on the video stream ID, obtain the face data of all frames in the video stream ID and arrange them according to the frame number;

[0073] A5. Decrypt and losslessly decompress the facial data at different locations in each frame to restore the original facial region data;

[0074] A6. Restore the face region data at different locations within each frame to their corresponding positions according to the frame number;

[0075] A7. Restore the video streams of all paths to obtain the original data of the restored video streams;

[0076] A8. Play back the original data of the restored video stream.

[0077] like Figure 4 As shown in the diagram, the playback effect of the desensitized video stream is illustrated. The face area in the diagram has been desensitized, which effectively protects personal privacy.

[0078] This application discloses a dedicated player for multi-channel video streams at the edge, which performs reverse operations on the desensitized video stream based on valid data and encrypted data to obtain the original data of the recovered video stream and then plays it.

[0079] This application discloses an edge-side multi-channel video stream desensitization and reversal system. The system of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a real-time video stream desensitization program. When the processor executes the computer program, it implements the method in embodiment 1.

[0080] Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as: desensitization module, reversal module.

[0081] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the edge multi-channel video stream desensitization and reversal system. For example, the computer program can be divided into multiple modules, each with the following specific functions:

[0082] 1. Desensitization module, used to desensitize the raw video stream data;

[0083] 2. Reverse module, used to reverse the desensitized video stream to obtain the original data of the restored video stream.

[0084] The edge-side multi-channel video stream desensitization and reversal system can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The edge-side multi-channel video stream desensitization and reversal system may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above examples are merely examples of the edge-side multi-channel video stream desensitization and reversal system and do not constitute a limitation on the system. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the edge-side multi-channel video stream desensitization and reversal system may also include input / output devices, network access devices, buses, etc.

[0085] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the edge-side multi-channel video stream desensitization and reversal system, connecting various parts of the system via various interfaces and lines.

[0086] The memory can be used to store the computer program and / or modules. The processor implements various functions of the edge-side multi-channel video stream desensitization and reversal system by running or executing the computer program and / or modules stored in the memory, and by calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0087] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for desensitizing and reversing multiple video streams at the edge, characterized in that: This includes marking each original video stream data, integrating multiple video streams to obtain integrated data, performing face detection in the integrated data according to the quantity and scale of the original data from different video streams, parsing the face location and mapping it to the corresponding location in the original video stream data, copying the face region data for compression and encryption to obtain encrypted data, and desensitizing the face region in the original video stream data before using it for terminal display. Encode the face location information to obtain valid data, and bind the valid data to the encrypted data one-to-one; based on the valid data and the encrypted data, perform reverse operation on the desensitized original video stream data to obtain the recovered original video stream data. The integrated data includes data from multiple video streams at the same time node, and records the timestamps corresponding to the starting frames of all original video stream data in the integrated data. Set a face detection threshold, detect faces and bodies in the integrated data, and mark the positions of face and body bounding boxes. The face box position is represented by head_box(x, y, w, h), where x, y, w, and h represent the x-axis coordinates and y-axis coordinates of the predicted top-left corner of the face box on the integrated image, as well as the width and height information of the face box. The center point coordinates of the face box are: center_x = x + w / 2, center_y = y + h / 2. Based on the center point coordinates, the column coordinates of the partition where the face is located are: cid_c = center_x / / sqrt(m), and the row coordinates are: cid_r = center_y / / sqrt(m), where / / represents the quotient, sqrt represents the square root function, and m is the current integration scale. Based on the face location, calculate the original video stream ID of the area where the face is located, calculate the relative coordinates of the face in the original video stream, and calculate the real face position corresponding to the face coordinates mapped back to the original video stream. The real face position corresponding to the face coordinates mapped back to the original video stream is: real_x = rlt_x *sqrt(m), real_y = rlt_y * sqrt(m), real_w = rlt_w * sqrt(m), real_h = rlt_h *sqrt(m); A multi-task deep learning neural network model is established, and the trained multi-task deep learning neural network model is used to detect faces and pedestrians in the integrated data. The face location information includes: the original video stream information where the face is located, the starting frame information, the original video stream frame number where the face is located, and the face location information. The valid data is bound with the encrypted data.

2. The method for desensitizing multiple video streams at the edge according to claim 1, characterized in that: The original video stream data is positionally encoded, and the timestamps corresponding to the starting frames of all video streams in the integrated data are recorded to enable real-time face detection and de-identification of multiple video streams.

3. The method for desensitizing multiple video streams at the edge according to claim 1, characterized in that: The face region image in the original video stream data is copied, compressed using a lossless compression method, and the compressed data is symmetrically encrypted and stored.

4. The method for desensitizing multiple video streams at the edge according to claim 1, characterized in that: Decode the anonymized original video stream data, find the frame number in the original video stream data according to the valid data, obtain all face data within that frame number, determine the position corresponding to each face data, decrypt the face data at each position, restore it to its respective position, and obtain the restored original video stream data.

5. A system for desensitizing and reversing multiple video streams at the edge, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the method as described in any one of claims 1-4.

6. A device for desensitizing and reversing multiple video streams at the edge, characterized in that: The system includes, in sequence, multiple cameras, a mixer, a smart controller, a network recorder, and a player or dedicated player. The multiple cameras are used to acquire raw data from multiple video streams. The mixer is used to integrate and encode the raw data from the multiple video streams according to the quantity of raw data from different video streams. The smart controller is used to perform face detection in the integrated data, parse the face location, and map it to the corresponding location in the raw video stream data. The network recorder is used for storing de-identified video and compressed and encrypted face data. The player is used to play the desensitized integrated data, and the dedicated player is used to reverse the desensitized integrated data and face encryption data to play the restored original video stream data. The integrated data includes data from multiple video streams at the same time point and records the timestamps corresponding to the starting frames of all original video stream data in the integrated data. The intelligent central controller is also used for: Set a face detection threshold, detect faces and bodies in the integrated data, and mark the positions of face and body bounding boxes. The face box position is represented by head_box(x, y, w, h), where x, y, w, and h represent the x-axis coordinates and y-axis coordinates of the predicted top-left corner of the face box on the integrated image, as well as the width and height information of the face box. The center point coordinates of the face box are: center_x = x + w / 2, center_y = y + h / 2. Based on the center point coordinates, the column coordinates of the partition where the face is located are: cid_c = center_x / / sqrt(m), and the row coordinates are: cid_r = center_y / / sqrt(m), where / / represents the quotient, sqrt represents the square root function, and m is the current integration scale. Based on the face location, calculate the original video stream ID of the area where the face is located, calculate the relative coordinates of the face in the original video stream, and calculate the real face position corresponding to the face coordinates mapped back to the original video stream. The real face position corresponding to the face coordinates mapped back to the original video stream is: real_x = rlt_x *sqrt(m), real_y = rlt_y * sqrt(m), real_w = rlt_w * sqrt(m), real_h = rlt_h *sqrt(m); A multi-task deep learning neural network model is established, and the trained multi-task deep learning neural network model is used to detect faces and pedestrians in the integrated data. The face location information includes: the original video stream information where the face is located, the starting frame information, the original video stream frame number where the face is located, and the face location information. The valid data is bound with the encrypted data.

7. A dedicated player for multi-channel video streaming at the edge, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the method as described in claim 4 to reverse the desensitized data and obtain the restored original video stream data for playback.

Images