A file transmission method and system based on artificial intelligence

Abstract

The present invention provides an artificial intelligence-based file transmission method and system, which relates to the field of file transmission technology. The method includes determining the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and the speaker's appearance description text based on the speaker's speech video; using a position determination model to determine the position pointed by the speaker based on the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a panoramic image of the conference site; controlling a camera to take a picture of the position pointed by the speaker to obtain an image of the pointed area; using a personnel image matching model to match the image of the pointed conference person based on the speaker's appearance description text and the image of the pointed area; performing facial recognition on the image of the pointed conference person to determine the identity information of the pointed conference person; and based on the identity information of the pointed conference person, transmitting the file sent by the speaker to the pointed conference person. The method can quickly identify the identity of a specific conference person and improve file transmission efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of file transmission, and in particular to an artificial intelligence-based file transmission method and system. Background Art

[0002] In-person meetings are common across a variety of sectors, including business, education, government, and non-profit organizations. Sharing information and transferring files are crucial to ensuring all participants stay on the same page. However, when a speaker needs to send a file to a specific attendee, especially if they don't know all the attendees, the process often requires extra time and effort to verify the recipient's identity, resulting in inefficient file transfer.

[0003] In the current process, presenters may need to ask for the name, position, or other identifying information of specific meeting participants before manually sending files via email, instant messaging apps, or other file sharing services. This manual process of confirming identity and sending files can lead to inefficient file transfers and hinder meeting efficiency.

[0004] Therefore, how to quickly identify the identity of specific meeting participants and improve file transfer efficiency is a problem that needs to be solved. Summary of the Invention

[0005] The main technical problem solved by the present invention is how to quickly identify the identity of specific conference participants and improve file transmission efficiency.

[0006] According to the first aspect, the present invention provides a file transfer method based on artificial intelligence, including: identifying whether a speaker has said a preset sentence: if the speaker is identified as saying the preset sentence, obtaining the speaker's speech video and a panoramic image of the conference site; using a speech video processing model based on the speaker's speech video to determine the speaker's position, the speaker's facial orientation, the speaker's finger direction, and the speaker's appearance description text; using a position determination model based on the speaker's position, the speaker's facial orientation, the speaker's finger direction, and the panoramic image of the conference site to determine the position pointed by the speaker; controlling a camera to take a picture of the position pointed by the speaker to obtain an image of the pointed area; using a personnel image matching model based on the speaker's appearance description text and the pointed area image to match the image of the pointed conference person; performing face recognition on the image of the pointed conference person to determine the identity information of the pointed conference person; obtaining the file sent by the speaker; and based on the identity information of the pointed conference person, transferring the file sent by the speaker to the pointed conference person.

[0007] Furthermore, the method also includes: if the personnel image determination model does not match the image of the designated conference person, controlling the camera to reduce the shooting magnification to N times the shooting magnification of the designated area image, and obtaining the second shooting area image of the camera, where the value of N is between 0-1; generating the appearance description image of the speaker based on the appearance description text of the speaker using a diffusion model; determining the image of the designated conference person based on the second area shooting image of the camera and the appearance description image of the speaker; performing facial recognition on the image of the designated conference person to determine the identity information of the designated conference person; obtaining the file sent by the speaker; and transmitting the file sent by the speaker to the designated conference person based on the identity information of the designated conference person.

[0008] Furthermore, the speech video processing model is a recurrent neural network model, the input of the speech video processing model is the speech video of the speaker, and the output of the speech video processing model is the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a text description of the speaker's appearance.

[0009] Furthermore, the personnel image matching model is a residual neural network model, the input of the personnel image matching model is the appearance description text of the speaker and the image of the designated area, and the output of the personnel image matching model is the image of the designated conference person or the image of the designated conference person is not matched.

[0010] Furthermore, the value of N is 0.5.

[0011] According to a second aspect, the present invention provides an artificial intelligence-based file transfer system, comprising:

[0012] The recognition module is used to identify whether the speaker has spoken a preset sentence:

[0013] The first acquisition module is configured to acquire a video of the speaker's speech and a panoramic image of the conference site if the speaker is recognized to have spoken a preset sentence;

[0014] a speech video processing module, configured to determine the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a description of the speaker's appearance based on the speaker's speech video using a speech video processing model;

[0015] a position determination module, configured to determine the position pointed to by the speaker using a position determination model based on the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a panoramic image of the conference site;

[0016] A photographing module, configured to control a camera to take a photograph of the location pointed at by the speaker to obtain an image of the pointed area;

[0017] A person image matching module, configured to match an image of the designated conference person using a person image matching model based on the appearance description text of the speaker and the designated area image;

[0018] A face recognition module is used to perform face recognition on the image of the designated conference participant and determine the identity information of the designated conference participant;

[0019] The second acquisition module is used to obtain the file sent by the speaker;

[0020] The transmission module is used to transmit the file sent by the speaker to the designated conference participant based on the identity information of the designated conference participant.

[0021] Furthermore, the system is also used to: if the personnel image determination model does not match the image of the designated conference person, control the camera to reduce the shooting magnification to N times the shooting magnification of the designated area image, and obtain the second shooting area image of the camera, where the value of N is between 0-1; generate the appearance description image of the speaker based on the appearance description text of the speaker using a diffusion model; determine the image of the designated conference person based on the second area shooting image of the camera and the appearance description image of the speaker; determine the identity information of the designated conference person after performing face recognition on the image of the designated conference person; obtain the file sent by the speaker; and based on the identity information of the designated conference person, transfer the file sent by the speaker to the designated conference person.

[0022] Furthermore, the speech video processing model is a recurrent neural network model, the input of the speech video processing model is the speech video of the speaker, and the output of the speech video processing model is the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a text description of the speaker's appearance.

[0023] Furthermore, the personnel image matching model is a residual neural network model, the input of the personnel image matching model is the appearance description text of the speaker and the image of the designated area, and the output of the personnel image matching model is the image of the designated conference person or the image of the designated conference person is not matched.

[0024] Furthermore, the value of N is 0.5.

[0025] The present invention provides an artificial intelligence-based file transmission method and system, which includes identifying whether a speaker has spoken a preset sentence: if it is identified that the speaker has spoken the preset sentence, obtaining the speaker's speech video and a panoramic image of the conference site; using a speech video processing model based on the speaker's speech video to determine the speaker's position, the speaker's facial orientation, the speaker's finger direction, and the speaker's appearance description text; using a position determination model based on the speaker's position, the speaker's facial orientation, the speaker's finger direction, and the panoramic image of the conference site to determine the position pointed by the speaker; controlling a camera to take a picture of the position pointed by the speaker to obtain an image of the pointed area; using a personnel image matching model based on the speaker's appearance description text and the pointed area image to match the image of the pointed conference person; performing face recognition on the image of the pointed conference person to determine the identity information of the pointed conference person; obtaining the file sent by the speaker; and transferring the file sent by the speaker to the pointed conference person based on the identity information of the pointed conference person. This method can quickly identify the identity of a specific conference person and improve file transmission efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of an artificial intelligence-based file transfer method provided by an embodiment of the present invention;

[0027] Figure 2 A flowchart of an artificial intelligence-based file transfer method in a state where no image of a designated conference participant is matched, provided by an embodiment of the present invention;

[0028] Figure 3 A schematic diagram of an artificial intelligence-based file transfer system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In an embodiment of the present invention, there is provided Figure 1 An artificial intelligence-based file transmission method is shown, and the artificial intelligence-based file transmission method includes steps S1 to S9:

[0030] Step S1: Identify whether the speaker speaks a preset sentence.

[0031] In some embodiments, voice recognition technology can be used to identify whether the speaker has spoken a preset sentence. In some embodiments, the speaker's microphone audio can be transmitted to a control terminal in the conference, and the control terminal can use voice recognition technology to identify whether the speaker has spoken a preset sentence. For example, the preset sentence can be "Start file transfer."

[0032] Step S2: If it is recognized that the speaker has spoken a preset sentence, a video of the speaker's speech and a panoramic image of the conference site are obtained.

[0033] The speaker's speech video is a video recording of the speaker speaking at the meeting. The speaker's speech video includes the speaker's voice and image. In some embodiments, the speaker's speech video can be recorded by multiple cameras in the on-site meeting to obtain the speaker's speech video.

[0034] The panoramic image of the conference site is a panoramic image obtained by photographing the conference site.

[0035] Step S3: Determine the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a description of the speaker's appearance using a speech video processing model based on the speaker's speech video.

[0036] The speaker's position refers to the coordinates of the speaker's location in the video.

[0037] The speaker's facial orientation refers to the direction in which the speaker's face is facing.

[0038] The speaker's finger pointing refers to the direction in which the speaker's finger is pointing.

[0039] The speaker's appearance description is a description of the attendee's appearance. For example, the speaker's appearance description includes information such as hairstyle, clothing, and accessories. For example, the appearance description could be "I want to send this to the gentleman wearing a blue shirt, glasses, and blond hair."

[0040] The speech video processing model is a recurrent neural network model. The input of the speech video processing model is the speech video of the speaker, and the output of the speech video processing model is the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a text description of the speaker's appearance. The recurrent neural network model includes a recurrent neural network (RNN). The recurrent neural network model can process sequence data, capture sequence information, and output results based on the correlation between the previous and next data in the sequence. By processing the speaker's speech videos for consecutive time periods through the recurrent neural network model, it is possible to output features that comprehensively consider the correlation between the sequences at each time point, making the output features more accurate and comprehensive. The recurrent neural network includes a hidden layer and an output layer. Each neuron in the hidden layer receives the input and the hidden state of the previous moment as input at each time step, and generates the output of the current moment and the hidden state of the next moment. The recurrent neural network model can extract information about the speaker's position, facial orientation, finger pointing, and text description of appearance from the video data. Recurrent neural networks are an implementation of artificial intelligence.

[0041] In some embodiments, the speech video processing model includes a video segmentation layer, a text output layer, and a pointing information determination layer. Each of the video segmentation layer, the text output layer, and the pointing information determination layer comprises a recurrent neural network. The video segmentation layer takes as input the speaker's speech video, and outputs a video segment describing the speaker's appearance and a video segment showing the speaker's gestures and directions. The text output layer takes as input the video segment describing the speaker's appearance, and outputs a text description of the speaker's appearance. The pointing information determination layer takes as input the speaker's gestures and directions video segment, and outputs the speaker's position, facial orientation, and finger direction.

[0042] The video segmentation layer extracts appearance description and gesture pointing video segments from the speech video. The text output layer converts the appearance description video segments extracted by the video segmentation layer into text descriptions. The gesture pointing information determination layer receives the gesture pointing video segments and outputs the speaker's position, facial orientation, and finger pointing. Separating the speech video processing model into the video segmentation layer, text output layer, and gesture pointing information determination layer allows each layer to focus on a specific task, improving the efficiency and accuracy of the overall processing.

[0043] Step S4: Determine the position pointed by the speaker using a position determination model based on the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and the panoramic image of the conference site.

[0044] The speaker's pointed location refers to the speaker's pointed location determined by the location determination model based on the speaker's location, the speaker's facial orientation, the speaker's finger direction, and the panoramic image of the conference site. The speaker's pointed location indicates that the speaker intends to send a file to a person at the speaker's pointed location.

[0045] The position determination model is a residual neural network model. The input of the position determination model is the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a panoramic image of the conference site. The output of the position determination model is the position pointed by the speaker.

[0046] Residual neural networks introduce residual connections, which add the input and output signals together, allowing the network to more easily learn the residual information. Through residual connections, the model can directly learn the difference between the input and output, namely the residual, without having to completely relearn the mapping between input and output. This design helps alleviate the vanishing gradient problem and enables the training of very deep networks. Residual neural networks are an implementation method of artificial intelligence.

[0047] The residual neural network can process multiple input channels. For example, it can combine the speaker's position, facial orientation, and finger pointing information with the panoramic image and input them into the network for processing. It can then use the contextual information in the panoramic image of the conference site, combined with the speaker's posture and position, to more accurately infer the location the speaker is pointing to.

[0048] Step S5: controlling the camera to take a picture of the position pointed by the speaker to obtain an image of the pointed area.

[0049] After determining the location indicated by the speaker, the camera is controlled to take a directional image of the location, ensuring a clear image of the location. This provides a clear image of the location, helping to better understand and record the speaker's intended destination. The area indicated by the speaker includes multiple participants, so the image of the indicated area includes all of them.

[0050] Step S6: Match the image of the designated conference person using a personnel image matching model based on the appearance description text of the speaker and the designated area image.

[0051] The personnel image matching model is a residual neural network model. The input of the personnel image matching model is the appearance description text of the speaker and the image of the designated area. The output of the personnel image matching model is the image of the designated conference person or the image of the designated conference person is not matched.

[0052] The Residual Neural Network (ResNet) has powerful feature learning capabilities and can extract high-level, abstract feature representations from input text describing the appearance and images of the designated area. The model can understand the visual features of text descriptions and images and transform them into more recognizable and comparable feature representations.

[0053] Through training, the residual neural network model learns the correlations and interactions between different features, enabling pattern matching of input information. For example, the model can learn to match a text description of a speaker's appearance with images of multiple meeting participants in the area being referred to, identifying possible matches.

[0054] Step S7: performing face recognition on the image of the designated conference participant to determine the identity information of the designated conference participant.

[0055] Face recognition can be an existing face recognition algorithm, for example, face recognition can be principal component analysis, deep learning algorithm, etc.

[0056] The identity information of the conference participants referred to here refers to the specific identity information of the conference participants obtained after confirmation through facial recognition. The identity information of the conference participants referred to here includes name, position, etc.

[0057] Step S8: Obtain the file sent by the speaker.

[0058] In some embodiments, the speaker may transfer the file to the control terminal via email or a conference APP for the control terminal to obtain.

[0059] Step S9: Based on the identity information of the designated conference participant, the file sent by the speaker is transmitted to the designated conference participant.

[0060] After the identity information of the designated conference participants is determined, the control terminal can determine the specific person who should receive the file based on the conference participant list or identity information, and then transmit the file to the designated conference participants based on email, file sharing services, etc.

[0061] In some embodiments, if the personnel image matching model in step S6 does not match the image of the conference personnel, the Figure 2 The flowchart shown is used to transfer files. Figure 2 A flowchart of an artificial intelligence-based file transfer method provided by an embodiment of the present invention when the image of the designated conference participant is not matched is provided. The artificial intelligence-based file transfer method when the image of the designated conference participant is not matched comprises steps S21 to S26:

[0062] Step S21: If the personnel image determination model does not match the image of the designated meeting personnel, the camera is controlled to reduce the shooting magnification to N times the shooting magnification of the designated area image, and a second shooting area image of the camera is obtained, where the value of N is between 0 and 1.

[0063] If the person image determination model fails to match the specified meeting person image, it may indicate that the range of the specified meeting person image is too small and the search should be expanded. Therefore, the camera is controlled to reduce the shooting magnification to N times the shooting magnification of the specified area image and obtain a second camera shooting area image, where the value of N is between 0 and 1. When the value of N is 0.5, it means that the camera's shooting range is expanded to half of the original range, which ensures that a wider area is captured.

[0064] In some embodiments, the value of N is 0.5.

[0065] Step S22 : generating a speaker's appearance description image based on the speaker's appearance description text using a diffusion model.

[0066] The input to the diffusion model is a textual description of the speaker's appearance, and the output is an image describing the speaker's appearance. Diffusion models are a type of generative model that generates new samples by simulating the diffusion process of data. The core idea of ​​a diffusion model is to start with a simple noise distribution and gradually remove the noise to recover the distribution of the target data.

[0067] The diffusion model can generate an image of the speaker's appearance description that matches the speaker's appearance description text description.

[0068] Step S23 : determining the image of the conference participant based on the image captured by the second area of ​​the camera and the image describing the appearance of the speaker.

[0069] In some embodiments, the image of the designated conference person can be determined by a similarity calculation model. The similarity calculation model can be a residual neural network model. The input of the similarity calculation model is the image captured by the second area of ​​the camera and the appearance description image of the speaker, and the output of the similarity calculation model is the image of the designated conference person. Based on the residual neural network model, the image captured by the second area of ​​the camera and the appearance description image of the speaker can be used as input, and their feature representations can be learned through the residual neural network, and the similarity between the multiple images of the conference persons included in the image captured by the second area of ​​the camera and the appearance description image of the speaker can be calculated, so as to match the image of the designated conference person with the highest similarity, thereby determining the image of the designated conference person.

[0070] The diffusion model can generate visual images from text descriptions. Since meetings are often held in complex or crowded settings, cameras expand the captured area, increasing the size of the image of the indicated area. This can reduce the quality of the second image captured by the camera, and the image of the indicated area may not provide clear identification information. The generated appearance description image can be more directly compared with the actual image, which can improve matching accuracy. By generating an image, we can fully utilize the detailed information contained in the speaker's appearance description text, partially compensating for the subjectivity and inaccuracy of the text description, thereby improving the robustness and accuracy of the matching model.

[0071] Step S24: performing face recognition on the image of the designated conference participant to determine the identity information of the designated conference participant.

[0072] For step S24, please refer to step S7, which will not be repeated here.

[0073] Step S25: Obtain the file sent by the speaker.

[0074] For step S25, please refer to step S8, which will not be repeated here.

[0075] Step S26: Based on the identity information of the designated conference participant, the file sent by the speaker is transmitted to the designated conference participant.

[0076] For step S26, please refer to step S9, which will not be repeated here.

[0077] Based on the same inventive concept, Figure 3 A schematic diagram of an artificial intelligence-based file transfer system provided in an embodiment of the present invention includes:

[0078] The recognition module 31 is used to recognize whether the speaker has spoken a preset sentence:

[0079] A first acquisition module 32 is configured to acquire a video of the speaker's speech and a panoramic image of the conference site if the speaker is identified to have spoken a preset sentence;

[0080] a speech video processing module 33 for determining the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a description of the speaker's appearance based on the speaker's speech video using a speech video processing model;

[0081] a position determination module 34 for determining the position pointed to by the speaker using a position determination model based on the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and the panoramic image of the conference site;

[0082] A photographing module 35 is used to control a camera to take a picture of the location pointed by the speaker to obtain an image of the pointed area;

[0083] A person image matching module 36 is configured to match an image of the designated conference person based on the appearance description text of the speaker and the designated area image using a person image matching model;

[0084] A face recognition module 37 is used to perform face recognition on the image of the designated conference participant and determine the identity information of the designated conference participant;

[0085] A second acquisition module 38 is used to acquire the file sent by the speaker;

[0086] The transmission module 39 is configured to transmit the file sent by the speaker to the designated conference participant based on the identity information of the designated conference participant.

Claims

1. A file transfer method based on artificial intelligence, characterized in that: include: Identify whether the speaker utters a preset sentence; If the speaker is recognized to have spoken a preset sentence, the speaker's speech video and a panoramic image of the conference site are obtained; Determining the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and the speaker's appearance description based on the speaker's speech video using a speech video processing model, where the speaker's appearance description is the appearance description of the conference attendee to whom the speaker refers; Determine the position pointed by the speaker using a position determination model based on the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and the panoramic image of the conference site; Controlling a camera to take a picture of the location pointed by the speaker to obtain an image of the pointed area; Matching the image of the designated conference person using a personnel image matching model based on the appearance description text of the speaker and the designated area image; Performing facial recognition on the image of the designated conference participant to determine the identity information of the designated conference participant; Get the files sent by the spokesperson; Based on the identity information of the designated conference participant, the file sent by the speaker is transmitted to the designated conference participant.

2. The artificial intelligence-based file transfer method according to claim 1, wherein: The method further comprises: If the personnel image determination model does not match the image of the designated meeting personnel, the camera is controlled to reduce the shooting magnification to N times the shooting magnification of the designated area image, and a second shooting area image of the camera is obtained, where the value of N is between 0 and 1; Generate a speaker's appearance description image based on the speaker's appearance description text using a diffusion model, where the speaker's appearance description image is an appearance description image of the conference member referred to by the speaker; Determining the image of the conference person based on the image captured by the second area of ​​the camera and the image of the appearance description of the speaker; Performing facial recognition on the image of the designated conference participant to determine the identity information of the designated conference participant; Get the files sent by the spokesperson; Based on the identity information of the designated conference participant, the file sent by the speaker is transmitted to the designated conference participant.

3. The artificial intelligence-based file transfer method according to claim 1, wherein: The speech video processing model is a recurrent neural network model, the input of the speech video processing model is the speech video of the speaker, and the output of the speech video processing model is the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a text description of the speaker's appearance.

4. The artificial intelligence-based file transfer method according to claim 1, wherein: The personnel image matching model is a residual neural network model. The input of the personnel image matching model is the appearance description text of the speaker and the image of the designated area. The output of the personnel image matching model is the image of the designated conference person or the image of the designated conference person is not matched.

5. The artificial intelligence-based file transfer method according to claim 2, wherein: The value of N is 0.

5.

6. An artificial intelligence-based file transfer system, characterized in that: include: The recognition module is used to identify whether the speaker has spoken a preset sentence: The first acquisition module is configured to acquire a video of the speaker's speech and a panoramic image of the conference site if the speaker is recognized to have spoken a preset sentence; a speech video processing module, configured to determine, based on the speech video of the speaker, the speaker's facial orientation, the speaker's finger pointing, and a description of the speaker's appearance using a speech video processing model, wherein the description of the speaker's appearance is a description of the appearance of the conference participant to whom the speaker refers; a position determination module, configured to determine the position pointed to by the speaker using a position determination model based on the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a panoramic image of the conference site; A photographing module, configured to control a camera to take a photograph of the location pointed at by the speaker to obtain an image of the pointed area; A person image matching module, configured to match an image of the designated conference person using a person image matching model based on the appearance description text of the speaker and the designated area image; A face recognition module is used to perform face recognition on the image of the designated conference participant and determine the identity information of the designated conference participant; The second acquisition module is used to obtain the file sent by the speaker; The transmission module is used to transmit the file sent by the speaker to the designated conference participant based on the identity information of the designated conference participant.

7. The artificial intelligence-based file transfer system according to claim 6, wherein: The system is also used to: If the personnel image determination model does not match the image of the designated meeting personnel, the camera is controlled to reduce the shooting magnification to N times the shooting magnification of the designated area image, and a second shooting area image of the camera is obtained, where the value of N is between 0 and 1; Generate a speaker's appearance description image based on the speaker's appearance description text using a diffusion model, where the speaker's appearance description image is an appearance description image of the conference member referred to by the speaker; Determining the image of the conference person based on the image captured by the second area of ​​the camera and the image of the appearance description of the speaker; Performing facial recognition on the image of the designated conference participant to determine the identity information of the designated conference participant; Get the files sent by the spokesperson; Based on the identity information of the designated conference participant, the file sent by the speaker is transmitted to the designated conference participant.

8. The artificial intelligence-based file transfer system according to claim 6, wherein: The speech video processing model is a recurrent neural network model, the input of the speech video processing model is the speech video of the speaker, and the output of the speech video processing model is the speaker's position, the speaker's facial orientation, the speaker's finger pointing, and a text description of the speaker's appearance.

9. The artificial intelligence-based file transfer system according to claim 6, wherein: The personnel image matching model is a residual neural network model. The input of the personnel image matching model is the appearance description text of the speaker and the image of the designated area. The output of the personnel image matching model is the image of the designated conference person or the image of the designated conference person is not matched.

10. The artificial intelligence-based file transfer system according to claim 7, wherein: The value of N is 0.5.

Images