CN106874827A - Video frequency identifying method and device - Google Patents

Abstract

The present disclosure relates to a video recognition method and device, by acquiring a target video; dividing the target video according to a first preset frame interval to obtain multiple video clips; according to a second preset frame interval, from each video clip extract the first frame image; extract the first frame image containing face information to obtain the second face frame image; based on the preset recognition model, identify the face identity in the second face frame image, Determine the face identification contained in the second human face frame image; according to the correspondence between the human face identification and the second human face frame image, according to the correspondence between the second human face frame image and the video clip to which the second human face frame image belongs relationship to form a tripartite correspondence table of face identifiers, second face frame images, and video clips. Therefore, according to the determined correspondence between the video clips and the face identifiers, the user can be pushed a video clip containing only appearances of actors he wishes to watch.

Description

Video recognition method and device

technical field

The present disclosure relates to the technical field of image processing, and in particular to a video recognition method and device.

Background technique

With the rapid development of society and the continuous advancement of science and technology, the information that people can touch has shown a geometric growth. People increasingly need to use information search technology to dig out effective information from massive amounts of information.

The current information search technology has a better effect on text search, and can quickly locate articles containing user-preset keywords. It can be searched by dragging the video progress button or pressing the video fast forward key, which is time-consuming and laborious, and the positioning is not accurate.

Contents of the invention

In order to solve the problem in the prior art that the actor's appearance time period in the video cannot be located, the present disclosure provides a video recognition method and device, by dividing the video into segments, and performing face recognition in the divided video segments, Determine the corresponding relationship between video clips and face identities, so that users can push video clips containing only the actors they want to watch. This method is effective and fast to realize face recognition, face video clip positioning, and improve user viewing Video user experience.

The present disclosure provides a video recognition method and device, and the technical solution is as follows:

According to a first aspect of an embodiment of the present disclosure, a video recognition method is provided, including:

Get the target video;

According to the first preset frame interval, the target video is divided to obtain a plurality of video clips;

extracting a first frame image from each of the video clips according to a second preset frame interval;

Detecting whether the first frame image contains human face information, extracting the first frame image containing human face information, and obtaining the second human face frame image;

Based on the preset identification model, identify the face identity in the second frame image of human face, and determine the face identity contained in the second frame image of human face;

According to the corresponding relationship between the face identification and the second human face frame image, according to the corresponding relationship between the second human face frame image and the video clip to which the second human face frame image belongs, the person is formed. A three-party correspondence table of the face identifier, the frame image of the second human face, and the video clip.

According to a second aspect of an embodiment of the present disclosure, a video recognition device is provided, including:

The first obtaining module is used to obtain the target video;

A first division module, configured to divide the target video according to a first preset frame interval to obtain a plurality of video clips;

The second dividing module is used to extract the first frame image from each of the video clips according to the second preset frame interval;

A detection module, configured to detect whether the first frame image contains face information, extract the first frame image containing face information, and obtain a second face frame image;

A recognition module, configured to recognize the face identity in the second face frame image based on a preset recognition model, and determine the face identity contained in the second face frame image;

A matching module, configured to, according to the correspondence between the face identification and the second face frame image, and according to the correspondence between the second face frame image and the video segment to which the second face frame image belongs and forming a tripartite correspondence table of the face identifier, the second face frame image, and the video clip.

The method and device provided by the embodiments of the present disclosure may include the following beneficial effects: by acquiring the target video; dividing the target video according to the first preset frame interval to obtain multiple video clips; according to the second preset frame interval, from Extract the first frame image from each video clip; detect whether the first frame image contains face information, extract the first frame image containing face information, and obtain the second face frame image; based on the preset recognition model , identify the face identity in the second face frame image, and determine the face identifier contained in the second face frame image; according to the corresponding relationship between the face identifier and the second face frame image, according to the second face The corresponding relationship between the frame image and the video segment to which the second human face frame image belongs forms a tripartite correspondence table of the face identifier, the second human face frame image, and the video segment. Thereby, according to the corresponding relationship between the determined video clips and the face identification, the user can push the video clips that only include the appearance of the actors they want to watch. This method effectively and quickly realizes face recognition, face video clip positioning, and improves The user experience of the user watching the video.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description serve to explain the principles of the disclosure.

Fig. 1 is a flowchart of a video recognition method shown according to an exemplary embodiment;

Fig. 2 is a flowchart of a video recognition method shown according to another exemplary embodiment;

Fig. 3 is a schematic diagram of a video division method of the embodiment shown in Fig. 2;

Fig. 4 is a flowchart of a video recognition device according to an exemplary embodiment;

Fig. 5 is a flow chart of a video recognition device according to another exemplary embodiment.

By means of the above-mentioned drawings, certain embodiments of the present disclosure have been shown and will be described in more detail hereinafter. These drawings and written description are not intended to limit the scope of the disclosed concept in any way, but to illustrate the disclosed concept for those skilled in the art by referring to specific embodiments.

detailed description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatuses and methods consistent with aspects of the present disclosure as recited in the appended claims.

Fig. 1 is a flow chart of a video recognition method shown according to an exemplary embodiment. As shown in Fig. 1, the video recognition method of this embodiment can be applied to the video server of the video provider and can also be applied to the video receiving party In the terminal (client device), the following example is applied to a video server, and the method of this embodiment includes the following steps:

The video processing method includes the following steps:

In step 101, a target video is acquired.

Specifically, video is essentially connected by a series of static images. Generally speaking, when continuous image changes exceed 24 frames per second, according to the principle of persistence of vision, the human eye cannot distinguish a single static image. It looks like a smooth and continuous visual effect, and such a continuous picture is called a video. Recognition of actors appearing in the target video can be realized by performing face image recognition on the continuous frame images constituting the target video.

In step 102, the target video is divided according to the first preset frame interval to obtain multiple video clips.

Specifically, as mentioned above, the reason why a video stream can be composed of frames of static images is mainly because human eyes have limited ability to recognize rapidly changing single static images. Going up can be a smooth and continuous visual effect. Therefore, the target video can be divided into individual video clips according to the number of static pictures contained in a certain interval interval, and from the perspective of the user's viewing effect of the video, the first preset frame interval can be in minutes, such as 0.5 Minutes, 1 minute, so that when the video clips containing the appearance of the user's favorite actor are extracted from the target video, the fluidity of each video segment is better, without the sense of jumping and abruptness of each frame.

In step 103, the first frame image is extracted from each video clip according to the second preset frame interval.

Specifically, even after the complete video is segmented, the number of frame images contained in each video segment is still very large. As mentioned above, a second of video can contain dozens of frames of static images. Face recognition is performed on each frame of image in each video clip, which requires a huge amount of computation and a low recognition rate. Therefore, some specific frame images can be extracted in each video clip, and these specific frame images are scanned to obtain the facial feature information contained in the image. The extraction of specific frame images can be based on the processing performance of the processor. To divide, if the processing performance of the processor is high, the second preset frame interval can be smaller, because the first frame image may contain face information, or may not contain face information, if the second preset frame interval is small then Increased the probability of extracting frame images containing human face information from video clips. Preferably, the first preset frame interval is greater than the second preset frame interval.

In step 104, it is detected whether the first frame image contains human face information, and the first frame image containing human face information is extracted to obtain a second human face frame image.

Specifically, detecting whether the first frame image contains face information refers to searching the image with a certain strategy to determine whether it contains face information, where the face information can be single face information or multiple face information, and calibrate the position where the face information appears in the frame image to confirm the coordinate position of each face information in the frame image. The first frame image is screened, and the first frame image containing human face information is extracted to obtain a second human face frame image.

In step 105, based on a preset recognition model, the identity of the face in the second frame image of the face is identified, and the identity of the face contained in the second frame image of the face is determined.

Specifically, in the prior art, there are a variety of algorithms for identifying the face identity in the image. Based on different algorithms, different recognition models can be obtained. For example, a large number of face pictures are collected as sample data, and artificial neural The network trains the sample data to obtain a neural network model with artificial intelligence learning ability, and then uses the trained artificial neural network model to recognize the face image to be recognized to obtain the recognition result. The trained artificial neural network model is the preset recognition model. After preprocessing all the second face frame images, they are input into the preset recognition model as input data, and the recognition results of the face images appearing in each second face frame image can be obtained, that is, the second person The face identifier included in the face frame image, where the face identifier can be the name of an actor in the video.

In step 106, according to the corresponding relationship between the human face identification and the second human face frame image, according to the corresponding relationship between the second human face frame image and the video segment to which the second human face frame image belongs, a human face identification, a second Three-party correspondence table of face frame images and video clips.

Specifically, through the tripartite correspondence table, a specific face identification can be quickly located to obtain video clips containing the specific face identification, so that these video clips containing the specific face identification can be extracted and Continuous playback, so as to achieve the purpose of users only watching video clips of their favorite actors.

In this embodiment, by acquiring the target video; according to the first preset frame interval, the target video is divided to obtain multiple video clips; according to the second preset frame interval, the first frame image is extracted from each video clip ; Detect whether the first frame image contains human face information, extract the first frame image containing human face information, and obtain the second human face frame image; based on the preset recognition model, the people in the second human face frame image Identify the face identity, determine the face identification contained in the second human face frame image; according to the corresponding relationship between the human face identification and the second human face frame image, according to the belonging The corresponding relationship of the video clips forms a tripartite correspondence table of the face identification, the second face frame image, and the video clips. Thereby, according to the corresponding relationship between the determined video clips and the face identification, the user can push the video clips that only include the appearance of the actors they want to watch. This method effectively and quickly realizes face recognition, face video clip positioning, and improves The user experience of the user watching the video.

Fig. 2 is a flow chart of a video recognition method according to another exemplary embodiment. As shown in Fig. 2, the video processing method of this embodiment can be applied to a video server of a video provider or to receive video In the terminal (client device) of the party, the following is used in a video server as an example, the method of this embodiment includes the following steps:

In step 201, a target video is acquired.

In step 202, the target video is divided according to the first preset frame interval to obtain multiple video clips.

In step 203, the first frame image is extracted from each video clip according to the second preset frame interval.

Wherein, the first preset frame interval is greater than the second preset frame interval. Preferably, the second preset frame interval is 5 frames of static images.

In step 204, it is detected whether the first frame image contains human face information, and the first frame image containing human face information is extracted to obtain a second human face frame image.

In step 205, the target face identification corresponding to the target video is obtained from the description information of the target video.

Specifically, the description information refers to the program introduction for the target video, which usually includes the cast list of the main actors in the video, and the names of the actors in the cast list can be used as face identification, and the target video is identified according to the face identification, to determine which video clips contain the face identification. In a frame image, there may be multiple face logos, and each face logo is calibrated. Compared with only the specified face logo in the image, the processing efficiency of the two is very different. Therefore, by acquiring the target face identification, the efficiency of locating the target face in the target video can be accelerated.

In step 206, a first preset recognition model corresponding to the target face identifier is called from the recognition model database according to the target face identifier.

In step 207, based on the first preset recognition model, the identity of the face in the second face frame image is identified, the third face frame image is determined in the second face frame image, and the third face frame image It is a second human face frame image containing the target human face identification.

Specifically, the first preset recognition model is a targeted recognition model, which can identify a given target face more specifically. For example, a recognition model is obtained by training 100,000 photos of 100 celebrities. After training, the recognition model can quickly recognize other photos of the 100 celebrities. It is also possible to use 100,000 or other photos of 10 celebrities or 1 celebrity to train the corresponding recognition model. Generally speaking, under the same training conditions, the narrower the scope of application of the recognition model, the better its performance on face recognition. The higher the accuracy. Therefore, by calling the first preset recognition model corresponding to the target face identifier from the recognition model database according to the target face identifier, and based on the specific first preset recognition model, the second face frame image Identifying the identity of the face, so that the third frame image of the face containing the identity of the target face can be determined in the second frame image of the face. Therefore, the recognition accuracy of the target face is improved.

In step 208, according to the corresponding relationship between the target human face identification and the third human face frame image, according to the corresponding relationship between the third human face frame image and the video segment to which the third human face frame image belongs, form the target human face identification, A tripartite correspondence table of the third face frame image and the video clip.

Optionally, before step 206, calling the first preset recognition model corresponding to the target face identifier from the recognition model database according to the target face identifier, it may also include:

According to the target face identification, the target human face picture data packet corresponding to the target human face identification is transferred from the picture database;

A target face picture data package is used as a training sample, and a first preset recognition model corresponding to the target face identification is obtained through training.

Specifically, the target face picture data package contains a preset number of face images corresponding to the target face identification, that is, training samples; generally speaking, the more training samples, the greater the recognition accuracy of the recognition model. The higher, but the specific number of training samples also needs to be determined according to the properties of the algorithm used. Among them, the training algorithm can use a deep convolutional neural network.

Optionally, after forming the tripartite correspondence table of the face identification, the second face frame image, and the video clip, it also includes:

Receive the video push request sent by the terminal, the video push request includes: the face ID to be pushed;

Search the three-party correspondence table according to the face ID to be pushed, and push the video clip corresponding to the face ID to be pushed to the terminal.

Specifically, the user can install the application APP suitable for the video recognition method on the terminal (mobile phone, PAD, etc.), input the name of the actor he wants to watch, and the cloud will analyze the target video in advance according to the name of the actor. The three-party correspondence table determines the video segment corresponding to the actor's name for the user, and pushes it to the terminal, so that the user can only watch the segment in the target video in which his favorite actor appears, improving the viewing effect.

The following example illustrates how the video recognition method locates a specific actor in a video: Please refer to FIG. 3. In FIG. As shown, be divided into 6 video clips; Obtain the main actor information contained in this target video from the descriptive information of this target video (such as cast and crew introduction, film introduction etc.); (for example include " Yang Ying ", " Deng Chao ", "Zheng Kai", etc.); in the database, the face picture data package corresponding to the actor's identification is called, for example, a large number of pictures about "Yang Ying" are called; a large number of pictures about "Yang Ying" are used as training samples, A recognition model that can recognize whether the target face is "Yang Ying" is obtained through training. Because each video clip is composed of frame images one by one, and generally speaking, a second of video contains dozens of frames. According to the definition of the video, the number of frame images contained in high-definition video is more, so If face detection is performed on each frame of the target video, resources are wasted, and the efficiency is not high; at the same time, for users to watch the appearance screen of their favorite actors, it is more reasonable to use minutes as the truncation unit, and video jumps to be performed in seconds. affect the viewing experience. Therefore, the length of the video segment can preferably be positioned at half a minute or one minute. For the first frame image extracted from the video clip and used to detect the face information, it is not necessary to check every frame, and the second preset frame interval with a preset step size can be used for extraction and detection. As shown in B in FIG. 3 , a certain number of frame images are extracted from each video clip as the first frame image B to be detected. Face detection is performed on the extracted first frame image B, and the detection algorithm can use the AdaBoost iterative algorithm, which can effectively improve the detection efficiency of the face image and improve the detection accuracy at the same time. As shown in C in FIG. 3 , the second human face frame image C detected to contain the human face image in the first frame image B is extracted for face recognition. Input the second face frame image C into the respective recognition models of "Yang Ying", "Deng Chao" and "Zheng Kai" obtained before for recognition, and obtain the third face frame image D1, D2, D3, D4, wherein D1 contains "Yang Ying", D2 contains "Deng Chao", D3 contains "Yang Ying" and "Deng Chao", D4 contains "Deng Chao", "Zheng Kai" and other recognition results. As shown in Table 1: the corresponding relationship between the third human face frame image and the video clip is determined, and a tripartite correspondence table of the target face identifier, the third human face frame image and the video clip is formed.

Table 1. The three-party correspondence table between the target face identification, the third face frame image and the video clip

If a push request is received that the user chooses to watch a video clip with "Yang Ying" appearing, the video clips A2 and A3 can be played continuously for the user, thereby quickly locating the video of the actor he wants to see for the user.

To sum up, this embodiment divides the video into segments, and in the divided video segments, establishes a specific face recognition model for a specific face, and based on the specific face recognition model, detects the specific person in each video segment. Face recognition can effectively improve the recognition efficiency, and users can quickly push video clips that only contain the actors they want to watch. This method effectively and quickly realizes face recognition, face video clip positioning, and improves users' viewing experience. experience.

The following are device embodiments of the present disclosure, which can be used to implement the method embodiments of the present disclosure. For details not disclosed in the disclosed device embodiments, please refer to the disclosed method embodiments.

Fig. 4 is a flow chart of a video recognition device according to an exemplary embodiment. As shown in Fig. 4, the video recognition device can be implemented as part or all of an electronic device through software, hardware or a combination of the two. The video processing device may include:

The first acquiring module 41 is configured to acquire the target video. The first division module 42 is configured to divide the target video according to the first preset frame interval to obtain multiple video clips. The second dividing module 43 is used for extracting the first frame image from each video segment according to the second preset frame interval. The detection module 44 is used to detect whether the first frame image contains human face information, extract the first frame image containing human face information, and obtain the second human face frame image. The identification module 45 is configured to identify the identity of the face in the second frame image of the face based on the preset identification model, and determine the identity of the face contained in the frame image of the second face. Matching module 46, is used for according to the corresponding relation of face mark and the second frame image of face, according to the corresponding relation of the video segment that the frame image of the second face and the frame image of the second face belong to, form face mark, the first frame image of face The tripartite correspondence table of two face frame images and video clips.

In this embodiment, by acquiring the target video; according to the first preset frame interval, the target video is divided to obtain multiple video clips; according to the second preset frame interval, the first frame image is extracted from each video clip ; Detect whether the first frame image contains human face information, extract the first frame image containing human face information, and obtain the second human face frame image; based on the preset recognition model, the people in the second human face frame image Identify the face identity, determine the face identification contained in the second human face frame image; according to the corresponding relationship between the human face identification and the second human face frame image, according to the belonging The corresponding relationship of the video clips forms a tripartite correspondence table of the face identification, the second face frame image, and the video clips. Thereby, according to the corresponding relationship between the determined video clips and the face identification, the user can push the video clips that only include the appearance of the actors they want to watch. This method effectively and quickly realizes face recognition, face video clip positioning, and improves The user experience of the user watching the video.

Fig. 5 is a flow chart of a video recognition device according to another exemplary embodiment. The video recognition device may be implemented as part or all of an electronic device through software, hardware or a combination of the two. Based on the above device embodiment, the first preset frame interval is greater than the second preset frame interval.

Optionally, the video recognition device also includes:

The second acquiring module 47 is configured to acquire the target face identification corresponding to the target video from the description information of the target video.

Correspondingly, the recognition module 45 includes:

The call sub-module 451 is configured to call the first preset recognition model corresponding to the target face mark from the recognition model database according to the target face mark.

The identification sub-module 452 is configured to identify the face identity in the second face frame image based on the first preset identification model.

The determination sub-module 453 is configured to determine a third human face frame image in the second human face frame image, and the third human face frame image is the second human face frame image including the target human face identification.

Correspondingly, the matching module 46 is specifically configured to, according to the corresponding relationship between the target face identification and the third human face frame image, and according to the corresponding relationship between the third human face frame image and the video segment to which the third human face frame image belongs, to form A tripartite correspondence table of the target face identification, the third face frame image, and the video clip.

Optionally, the video recognition device also includes:

The picture acquisition module 48 is used for calling the target human face picture data package corresponding to the target human face identification from the picture database according to the target human face identification.

The training module 49 is configured to use the target face picture data package as a training sample to train to obtain a first preset recognition model corresponding to the target face identification.

Optionally, the video recognition device also includes:

The receiving module 50 is configured to receive the video push request sent by the terminal, and the video push request includes: the face identification to be pushed.

The search module 51 is used to search in the three-party correspondence table according to the face identification to be pushed, and push the video clip corresponding to the face identification to be pushed to the terminal.

Regarding the apparatus in the foregoing embodiments, the specific manner in which each module executes operations has been described in detail in the embodiments related to the method, and will not be described in detail here.

Other embodiments of the present disclosure will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, and these modifications, uses or adaptations follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not disclosed in the present disclosure . The specification and examples are to be considered exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It should be understood that the present disclosure is not limited to the precise constructions which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims (10)

1. A video recognition method, characterized in that the method comprises: Get the target video; According to the first preset frame interval, the target video is divided to obtain a plurality of video clips; extracting a first frame image from each of the video clips according to a second preset frame interval; Detecting whether the first frame image contains human face information, extracting the first frame image containing human face information, and obtaining the second human face frame image; Based on the preset identification model, identify the face identity in the second frame image of human face, and determine the face identity contained in the second frame image of human face; According to the corresponding relationship between the face identification and the second human face frame image, according to the corresponding relationship between the second human face frame image and the video clip to which the second human face frame image belongs, the person is formed. A three-party correspondence table of the face identifier, the frame image of the second human face, and the video clip. 2. The method according to claim 1, wherein the first preset frame interval is greater than the second preset frame interval. 3. The method according to claim 1, characterized in that, based on the preset recognition model, the identity of the face in the second frame image of the face is identified, and the identity of the face in the frame image of the second face is determined. Before the included face ID, also include: Acquire the target face identification corresponding to the target video from the description information of the target video; Correspondingly, the recognition of the face identity in the second face frame image based on the preset recognition model, and determining the face identity contained in the second face frame image include: Call the first preset recognition model corresponding to the target face identifier from the recognition model database according to the target face identifier, and based on the first preset recognition model, frame the image of the second face Identify the identity of the face in the frame image, determine the third frame image of the face in the second frame image of the face, and the frame image of the third face is the second frame image of the face that contains the target face identification ; Correspondingly, according to the correspondence between the face identification and the second face frame image, according to the correspondence between the second face frame image and the video clip to which the second face frame image belongs , forming the tripartite correspondence table of the face identification, the second face frame image, and the video clip includes: Forming the A three-party correspondence table of the target face identifier, the third face frame image, and the video clip. 4. The method according to claim 3, wherein, before calling the first preset recognition model corresponding to the target face mark from the recognition model database according to the target face mark, further include: According to the target human face identification, the target human face picture data package corresponding to the target human face identification is called from the picture data base; Using the target face picture data package as a training sample, the first preset recognition model corresponding to the target face identifier is obtained through training. 5. The method according to any one of claims 1 to 4, characterized in that, after forming the tripartite correspondence table of the face identifier, the second face frame image, and the video clip, further include: Receive the video push request sent by the terminal, the video push request includes: the face identification to be pushed; Searching the tripartite correspondence table according to the face ID to be pushed, and pushing the video clip corresponding to the face ID to be pushed to the terminal. 6. A video recognition device, characterized in that the device comprises: The first obtaining module is used to obtain the target video; A first division module, configured to divide the target video according to a first preset frame interval to obtain a plurality of video clips; The second dividing module is used to extract the first frame image from each of the video clips according to the second preset frame interval; A detection module, configured to detect whether the first frame image contains face information, extract the first frame image containing face information, and obtain a second face frame image; A recognition module, configured to recognize the face identity in the second face frame image based on a preset recognition model, and determine the face identity contained in the second face frame image; A matching module, configured to, according to the correspondence between the face identification and the second face frame image, and according to the correspondence between the second face frame image and the video segment to which the second face frame image belongs and forming a tripartite correspondence table of the face identifier, the second face frame image, and the video clip. 7. The device according to claim 6, wherein the first preset frame interval is greater than the second preset frame interval. 8. The device according to claim 6, further comprising: The second obtaining module is used to obtain the target face identification corresponding to the target video from the description information of the target video; Correspondingly, the identification module includes: The calling submodule is used to call the first preset recognition model corresponding to the target face mark from the recognition model database according to the target face mark; An identification submodule, configured to identify the face identity in the second face frame image based on the first preset identification model; A determining submodule, configured to determine a third face frame image in the second face frame image, where the third face frame image is a second face frame image containing the target face identification; Correspondingly, the matching module is specifically configured to: according to the corresponding relationship between the target face identifier and the third face frame image, according to the third face frame image and the third face frame image The corresponding relationship of the attributed video clips forms a tripartite correspondence table of the target face identifier, the third human face frame image, and the video clips. 9. The device according to claim 8, further comprising: The picture acquisition module is used to retrieve the target face picture data packet corresponding to the target face mark from the picture database according to the target face mark; A training module, configured to use the target face picture data package as a training sample to train to obtain the first preset recognition model corresponding to the target face identification. 10. The device according to any one of claims 6-9, characterized in that the device further comprises: The receiving module is used to receive the video push request sent by the terminal, and the video push request includes: the face identification to be pushed; A search module, configured to search in the tripartite correspondence table according to the face ID to be pushed, and push the video clip corresponding to the face ID to be pushed to the terminal.