A data processing method based on AI server
By receiving and processing image information on the AI server side, determining candidate targets and triggering corresponding camera units to identify them, and using preset hybrid models for direction and face recognition, the problem of waste of resources and bandwidth requirements of existing AI visual target recognition technology is solved. , achieving efficient and accurate target recognition.
Patent Information
- Application Number
- CN202411014761.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-26
AI Technical Summary
The existing AI visual object recognition technology requires real-time monitoring of all cameras in the area, resulting in waste of resources and large working bandwidth requirements.
By adopting the data processing method based on the AI server, by receiving the image information of the first imaging unit, determine whether there is a candidate target, and triggering the corresponding second imaging unit to collect and recognize images, and using a preset hybrid model for direction recognition and face Identify, save resources, and reduce bandwidth requirements.
It realizes the completion of target recognition under a smaller bandwidth, saves resources, improves the accuracy and efficiency of recognition, and avoids the need for all cameras to work simultaneously.
Smart Images

Figure CN118552909B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing technology, and in particular to a data processing method based on an AI server. Background Art
[0002] With the development of image processing technology, AI visual target recognition technology has also become a research hotspot. Target recognition, face recognition, target tracking, etc. are widely used in people's daily lives. AI visual recognition technology uses cameras to collect images or videos, obtain indoor visual information, such as the target's face information, and then recognize it. However, the existing AI visual target recognition technology requires all cameras in the area to monitor and collect in real time, and requires a large working bandwidth. Therefore, it is urgent to propose an improved technology. Summary of the invention
[0003] A brief overview of the present invention is provided below in order to provide a basic understanding of certain aspects of the present invention. It should be understood that this overview is not an exhaustive overview of the present invention. It is not intended to identify key or important parts of the present invention, nor is it intended to limit the scope of the present invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to a more detailed description discussed later.
[0004] In view of this, the present invention provides a data processing method based on an AI server to overcome the problem that the existing AI visual target recognition technology requires all cameras in the area to monitor and collect data in real time and requires a large working bandwidth.
[0005] According to one aspect of the present invention, a data processing method based on an AI server is provided, the method being used to be executed on an AI server side, the method comprising: receiving first image information from a first camera unit, the first image information comprising an image within a first area corresponding to the first camera unit; based on the first image information, determining whether there is a candidate target currently within the first area; in the case where it is determined that there is a candidate target currently within the first area, determining at least one target to be identified among the candidate targets, and triggering a second camera unit to capture an image of the at least one target to be identified to obtain second image information; wherein the second image information comprises an image within a second area corresponding to the second camera unit; based on the second image information, automatically capturing the at least one target to be identified motion recognition; the second camera unit includes N levels of second camera sub-units, N is an integer greater than 1; wherein, while receiving the acquisition results of each level of the second camera sub-unit, according to the acquisition results, a preset hybrid model is used to perform direction recognition processing, and the second camera sub-unit to be triggered next is determined according to the direction of travel of the identified target to be identified, and face recognition processing is performed; the training data used by the preset hybrid model includes a first type of training data and a second type of training data, the first type of training data is an image sequence containing local motion features, the local includes the head, arms or legs, and the second type of training data is an image sequence containing global motion features, the local corresponds to the head, arms or legs, and the global corresponds to part or the whole body; according to the result of the automatic recognition, a corresponding operation is performed.
[0006] Furthermore, the first camera unit includes at least one first camera module; each first camera module is used to acquire images at a predetermined frequency; the first camera unit is suitable for sending first image information to the AI server at a predetermined time interval, and the first image information includes an image sequence consisting of multiple frames of images.
[0007] Furthermore, in the step of determining whether there is a candidate target currently in the first area, the target movement information in the first area is extracted by subtracting adjacent frames of the image sequence, thereby realizing real-time detection of moving targets in the first area, so that the moving targets appearing in the first area are regarded as candidate targets.
[0008] Furthermore, the second camera unit includes at least one second camera module.
[0009] Furthermore, each second camera module is suitable for capturing images of a partial area or the entire area within the second area.
[0010] Furthermore, each second camera module is suitable for adjusting shooting parameters according to instructions from the AI server.
[0011] Furthermore, the shooting parameters include one or more of the following parameters: focus mode; focal length; shooting angle; shutter speed; frame rate; night vision function parameters; and compression format parameters.
[0012] Further, the second area includes N levels of second sub-areas; the i-th level second camera subunit corresponds to the i-th level second sub-area, wherein i∈{1,2,…,N}.
[0013] Furthermore, after the second camera subunit at the i-th level performs image acquisition, it is determined whether the at least one target to be identified has been fully identified; wherein, i∈{1,2,…,N-1}; if the at least one target to be identified has been fully identified, the image acquisition process of the second camera unit is terminated; if there are some targets that have not yet been identified among the at least one target to be identified, the second camera subunit at the i+1th level is triggered to continue to identify the some targets that have not yet been identified.
[0014] The data processing method based on an AI server of the present invention can determine in real time whether there is a candidate target in the first area based on the acquisition result of the first camera unit; when it is determined that there is a candidate target, at least one target to be identified is determined, and the second camera unit is triggered to perform image acquisition and recognition on the target to be identified, so that the target can be identified more accurately. The above-mentioned data processing method based on an AI server of the present invention can enable the camera modules in the first and second camera units to collaborate in acquisition, without requiring all cameras to work at the same time, saving resources, and being able to complete acquisition and recognition at a relatively small bandwidth.
[0015] These and other advantages of the present invention will become more apparent through the following detailed description of the best embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The present invention may be better understood by referring to the following description given in conjunction with the accompanying drawings, wherein the same or similar reference numerals are used throughout the drawings to represent the same or similar components. The accompanying drawings, together with the following detailed description, are incorporated in and form a part of this specification and are used to further illustrate the preferred embodiments of the present invention and to explain the principles and advantages of the present invention.
[0017] Figure 1 It is a flow chart showing a data processing method based on an AI server of the present invention.
[0018] Figure 2 is a flow chart illustrating one exemplary process for implementing the image acquisition sub-process.
[0019] Figure 3 is a flow chart showing one exemplary process of the travel direction identification subroutine.
[0020] Those skilled in the art will appreciate that the elements in the drawings are shown only for simplicity and clarity and are not necessarily drawn to scale. For example, the size of some elements in the drawings may be enlarged relative to other elements to help improve the understanding of the embodiments of the present invention. DETAILED DESCRIPTION
[0021] Exemplary embodiments of the present invention will be described below in conjunction with the accompanying drawings. For the sake of clarity and conciseness, not all features of the actual implementation are described in the specification. However, it should be understood that many implementation-specific decisions must be made in the process of developing any such actual implementation in order to achieve the developer's specific goals, such as meeting those constraints related to the system and business, and these constraints may vary from implementation to implementation. In addition, it should be understood that although the development work may be very complex and time-consuming, it is only a routine task for those skilled in the art who benefit from the content of the present invention.
[0022] It is also necessary to explain here that, in order to avoid obscuring the present invention due to unnecessary details, only the device structure and / or processing steps closely related to the scheme according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.
[0023] The present invention provides a data processing method based on an AI server, which is used to be executed on the AI server side, and the data processing method based on the AI server includes: receiving first image information from a first camera unit, the first image information including an image within a first area corresponding to the first camera unit; based on the first image information, determining whether there are candidate targets currently in the first area; when it is determined that there are candidate targets currently in the first area, determining at least one target to be identified among the candidate targets, and triggering a second camera unit to perform image acquisition on the at least one target to be identified to obtain second image information; wherein the second image information includes an image within a second area corresponding to the second camera unit; based on the second image information, automatically identifying the at least one target to be identified; and performing corresponding operations according to the result of the automatic identification.
[0024] Figure 1 The flowchart of the above-mentioned data processing method based on the AI server is shown.
[0025] In step S110, first image information from a first camera unit is received, where the first image information includes an image within a first area corresponding to the first camera unit.
[0026] The first area is, for example, an entrance area of a predetermined place, and the range of the entrance area can be set according to actual needs, or can be determined based on experience or through experiments.
[0027] As an example, the first camera unit includes at least one first camera module. Each first camera module is used to capture images at a predetermined frequency; the first camera unit is suitable for sending first image information to the AI server at a predetermined time interval, and the first image information includes an image sequence composed of multiple frames of images.
[0028] For example, in step S110, image acquisition may be performed at a predetermined frequency, such as 5 frames, 10 frames, or 20 frames per second, etc.
[0029] In addition, in step S110, the first image information may be acquired at predetermined time intervals, such as intervals of 1 second, 2 seconds, etc.
[0030] In this way, the first image information acquired in real time each time in step S110 includes a group of image sequences consisting of multiple frame images. For example, assuming that the first camera unit sends data to the AI server once per second and its acquisition frame rate is 10 frames / second, then the first image information received each time in step S110 includes a group of image sequences (or frame sequences) consisting of 10 frame images.
[0031] Next, in step S120, based on the first image information, it is determined whether there is a candidate target currently in the first area: if yes, execute step S130; otherwise, return to execute step S110 to re-acquire new first image information.
[0032] Among them, the candidate targets may be one or more.
[0033] For example, in the embodiment of the present invention, the existing frame difference method can be used to determine whether there is a candidate target in the first area. In this way, the target movement information is extracted by using the method of making a difference between adjacent frames of the image sequence, and the real-time detection of the moving target in the first area is realized, so that the moving target appearing in the first area is used as a candidate target.
[0034] In step S130, when it is determined that candidate targets currently exist in the first area, at least one target to be identified is determined among the candidate targets, and the second camera unit is triggered to capture an image of the at least one target to be identified to obtain second image information.
[0035] The second image information includes an image in the second area corresponding to the second camera unit. Similar to the first image information, each second image information also includes a set of image sequences consisting of multiple frames of images. The second camera unit can, for example, use the same frequency as the first camera unit to collect images, and can send data to the AI server at the same time interval as the first camera unit.
[0036] The second area may overlap or partially overlap with the first area, or the second area may not overlap with the first area at all.
[0037] For example, assuming that the entire monitoring area includes the entrance area and the lobby public area of the predetermined place, and the lobby public area is adjacent to the entrance area, when people enter the above-mentioned predetermined place, they need to first enter the entrance area, and then enter the lobby public area after passing through the entrance area. As described above, assuming that the first area is the above-mentioned entrance area, then in an embodiment of the present invention, the second area can be the above-mentioned lobby public area.
[0038] As an example, if there is no candidate target in the first area, collection and recognition are continued until a candidate target appears, and then the processing of step S130 is performed on the candidate target.
[0039] As an example, if a candidate target appears in the first area and there is only one candidate target, the process can automatically enter step S130, or the user can be provided with an option of whether to identify the candidate target and perform the corresponding operation based on the user's selection.
[0040] As an example, if candidate targets appear in the first area, and the candidate targets include two or more, there are two ways to process.
[0041] In the first way, for example, all the candidate targets that appear may be taken as targets to be identified, and then these targets to be identified may be subsequently processed.
[0042] In the second manner, for example, part of the candidate targets that appear may be used as targets to be identified. For example, a user may select at least one target to be identified from the candidate targets.
[0043] As an example, the second camera unit includes at least one second camera module. Each second camera module is suitable for capturing an image of a partial area or the entire area within the second area.
[0044] As an example, each second camera module can adjust shooting parameters according to the instructions of the AI server. The shooting parameters may include one or more of the following parameters: focus mode; focal length; shooting angle; shutter speed; frame rate; night vision function parameters; compression format parameters.
[0045] In addition, in the embodiment of the present invention, the first and / or second camera module can adopt a fixed focus lens according to the scene to be compatible with high temperature environment. The pixels of the first and / or second camera module can be 2 million pixels or 1.3 million pixels, for example, and the main code stream is 1920*1080 (1-60 frames / second), the secondary code stream is 704*576, 352*288 (1-60 frames / second), and the transmission rate is adjustable between 16kbps~20Mbps.
[0046] Thus, in step S140, at least one target to be identified is automatically identified based on the second image information.
[0047] As an example, the second camera unit may include N-level second camera sub-units, and the N-level second camera sub-units, for example, use a step-by-step recognition method to identify at least one target to be identified; the N-level second camera sub-units include 1st to Nth level second camera sub-units; N is an integer greater than 1.
[0048] In one example, in a step-by-step identification method, when a second camera sub-unit at a certain level is working, the second camera sub-units at other levels suspend collection and do not send data to the AI server, thereby greatly reducing the working bandwidth.
[0049] For example, for the kth second camera subunit among the N second camera subunits, the following may be used: Figure 2 The sub-process shown is used to implement image acquisition.
[0050] In step S210, after the k-th level second camera subunit performs image acquisition (k∈{1,2,…,N-1}), the currently unrecognized target (which may be one or more) is identified based on the image acquisition result of the k-th level second camera subunit. After the identification is completed, the judgment process of step S220 is entered, that is, whether the at least one target to be identified determined in step S130 has been fully identified.
[0051] If the at least one target to be identified has been identified (ie, “yes”), the image acquisition process of the second camera unit is terminated (step S230 ).
[0052] If there are some unrecognized targets among the at least one target to be recognized (i.e., "No"), the second camera subunit of the next level (i.e., the k+1 level) is triggered to continue to recognize the some unrecognized targets until all the targets to be recognized among the at least one target to be recognized are recognized. That is, by assigning k to k+1 (i.e., "k=k+1"), the loop process returns to step S210 to perform image acquisition and recognition at the next level.
[0053] As an example, the next level second camera subunit is non-fixed and selectable. For example, there are 3 second camera subunits (each second camera subunit may include one or more second camera modules), namely x1, x2 and x3. Assuming that x1 is currently the first level second camera subunit, and the first level second camera subunit has not identified all the targets to be identified, so the second level second camera subunit is to be started. In an embodiment of the present invention, according to the needs or the travel direction recognition results described later, the corresponding one can be selected from x2 and x3 as the second level second camera subunit.
[0054] In one example, it is assumed that the at least one target to be identified has three targets to be identified, for example, the first target, the second target and the third target. When the second camera unit is triggered to start image acquisition, the first-level second camera sub-unit is first triggered to acquire images, and target recognition is performed based on the image acquisition result of the first-level second camera sub-unit (i.e., the at least one target to be identified). After the recognition process is completed, it is assumed that the second target and the third target have been identified, and the first target has not yet been identified.
[0055] In this way, the second-level second camera subunit is then triggered to continue image acquisition, and target recognition is performed based on the image acquisition result of the second-level second camera subunit (i.e., the first target is recognized). After the second-level second camera subunit completes the recognition process, assuming that the first target has been recognized, the image acquisition and recognition process of the second camera unit is terminated.
[0056] In another example, it is assumed that the at least one target to be identified has three targets to be identified, for example, the first target, the second target and the third target. When the second camera unit is triggered to start image acquisition, the first-level second camera sub-unit is first triggered to acquire images, and target recognition is performed based on the image acquisition result of the first-level second camera sub-unit (i.e., the at least one target to be identified). After the recognition process is completed, it is assumed that the first target has been identified, and the second target and the third target have not yet been identified.
[0057] In this way, the second-level second camera subunit is then triggered to continue image acquisition, and target recognition is performed based on the image acquisition result of the second-level second camera subunit (i.e., the second target and the third target are recognized). After the second-level second camera subunit completes the recognition process, it is assumed that the third target has been recognized, and the second target has not yet been recognized.
[0058] Then, the third-level second camera unit is triggered to continue to acquire images, and target recognition is performed based on the image acquisition result of the third-level second camera unit (i.e., the second target is recognized). After the second-level second camera unit completes the recognition process, assuming that the second target has not been recognized (possibly because the second target has left the acquisition area corresponding to the third-level second camera unit), the fourth-level second camera unit is triggered to continue to acquire images, and target recognition is performed based on the image acquisition result of the fourth-level second camera unit (i.e., the second target is still recognized).
[0059] After the second camera subunit of the fourth level completes the recognition process, assuming that the second target has been recognized, the image acquisition and recognition process of the second camera subunit ends.
[0060] In addition, if there are still unrecognized targets after the second camera sub-units of level 1 to level N complete the recognition process, the information corresponding to the target to be recognized (such as images, etc.) can be saved, for example, sent to the control center for subsequent processing. After receiving the above information, the control center can, for example, prompt the staff to intercept, etc., or automatically switch the access level to a high security level (for example, authorization is required to unlock) at the preset access control point, etc.
[0061] In addition, as an example, the second area includes N-level second sub-areas, and the i-th level second camera sub-unit corresponds to the i-th level second sub-area, where i∈{1,2,…,N}.
[0062] In one example, the N-level second sub-regions can be obtained by dividing the second region into N second sub-regions that are adjacent to each other in sequence.
[0063] In another example, for example, a partial area in the second area may be determined as the corresponding i-th level second sub-area according to actual needs, and different second sub-areas may be partially overlapped.
[0064] As an example, while receiving the acquisition results of the second camera subunit at each level, the AI server performs direction recognition processing and face recognition processing according to the above acquisition results. Through the direction recognition processing, the AI server can identify the direction of travel of the target to be identified, and then perform face recognition on it in combination with the face recognition processing to determine its identity. In an embodiment of the present invention, the face recognition process can be implemented, for example, using an existing face recognition algorithm, which will not be described in detail here.
[0065] As an example, one can Figure 3 The traveling direction identification sub-process shown is used to identify the traveling direction of the target to be identified.
[0066] like Figure 3As shown, in the moving direction identification sub-process, step S310 is first executed to obtain a frame difference result based on the frame difference method processing result of step S120, thereby obtaining an image sequence after frame difference processing.
[0067] Next, in step S320, a second-order Volterra filter is performed on the image sequence after the frame difference processing, and the result after the second-order Volterra filter is used as the frame difference feature information.
[0068] In step S330, direction recognition is performed using a preset hybrid model based on the frame difference feature information.
[0069] Among them, the preset hybrid model uses a CNN (Convolutional Neural Network) and a ViT model (Vision Transformer, a deep learning model based on an attention mechanism) fused by a stage-level fusion method. In this hybrid model, CNN is used to obtain local feature information, and ViT is used to obtain global feature information. The two are fused, for example, by a stage-level fusion method.
[0070] As an example, a preset hybrid model may be trained in advance, and the training data set may include, for example, labeled historical data.
[0071] In the embodiment of the present invention, different from the prior art, the training data includes at least two categories, namely, a first category of training data and a second category of training data.
[0072] Each first-category training data is a set of image sequences containing local motion features, such as image sequences corresponding to local parts such as the head, arms, and legs.
[0073] For example, taking a set of first-category training data corresponding to a head as an example, assuming that the head in each frame of the first-category training data is a side face, and the face is facing the right side of the image (i.e., the left side is the back of the head), the label of this set of training data is rightward. The recognition of the side face can be achieved by using existing contour recognition, edge detection algorithms, etc., which will not be described here.
[0074] For example, taking a group of first-category training data corresponding to arms as an example, assuming that the body in each frame of the image in this group of first-category training data is also in profile, and the image is collected for one side of the arm of the target to be identified (including the hand or excluding the hand), the direction of travel is determined according to the bending direction of the arm, thereby determining the label of this group of training data (left row or right row).
[0075] In addition, each second type of training data is a set of image sequences containing global motion features, such as an image sequence corresponding to a part of the body (including at least multiple parts) or the whole body.
[0076] For example, the moving direction of the target to be identified is determined according to the change of its position in a place in a set of image sequences.
[0077] In another embodiment of the present invention, the training data may include a third type of training data in addition to the first and second types of training data. The third type of training data may include both local and global motion features, so that the trained model is more accurate.
[0078] According to an embodiment of the present invention, for example, the second camera unit to be triggered next can be determined based on the direction of travel of the identified target to be identified. On the one hand, it can be determined which level of second camera unit to be triggered as the second camera unit to be operated. In other words, if the current level is the k-th level second camera unit, it can be determined which sub-area corresponding to the second camera unit is the k+1-th level second camera unit; on the other hand, it can be determined which position or range of high-resolution cameras to be triggered to collect images for face recognition, so that all high-resolution cameras do not need to be triggered to work.
[0079] Under normal circumstances, the target to be identified first enters the entrance area and then enters the public area of the hall. For most people, assuming that the route is fixed, they are likely to walk in sequence from the 1st-level second sub-area corresponding to the 1st-level second camera sub-unit to the Nth-level second sub-area corresponding to the Nth-level second camera sub-unit.
[0080] However, in some scenarios, the N-level second sub-areas may not be adjacent to each other in sequence. For example, after entering the second-level second sub-area, the target to be identified may enter the third-level second sub-area or the fourth-level second sub-area; or the target to be identified may return along the original route, etc. In this way, according to the direction of travel of the currently identified target to be identified, the second sub-area it will enter next can be determined, and the second camera sub-unit corresponding to the second sub-area can be controlled to collect data, so as to timely and accurately track the target to be identified, and then obtain its face recognition result.
[0081] In one example, each level of the second camera sub-unit is provided with one or more first high-resolution cameras and one or more first low-resolution cameras, wherein the first low-resolution cameras are used to capture images for direction recognition processing, and the first high-resolution cameras are used to capture images for face recognition.
[0082] In another example, each level of the second camera subunit is provided with one or more second low-resolution cameras, one or more second high-resolution cameras, and one or more third high-resolution cameras. The second low-resolution camera is used to capture image information containing global motion features, and the second high-resolution camera is used to capture image information containing local motion features. The two are combined to input a trained preset hybrid model to perform travel direction recognition. The third high-resolution camera is used to collect images for face recognition.
[0083] Thus, see Figure 1 In step S150, according to the result of automatic identification in the above step S140, corresponding operations are performed, for example, identifying that its identity matches the identity of a preset user and maintaining the current security level; if a target to be identified is ultimately not identified, a corresponding higher level of security inspection measures is set for the target to be identified, or the security department is notified, etc.
[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention and the beneficial effects brought about by the present invention have been described in detail with reference to the aforementioned embodiments, ordinary technicians in the field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the claims of the present invention.
Claims
1. A data processing method based on an AI server, characterized in that: The data processing method based on the AI server is used to be executed on the AI server side, and the data processing method based on the AI server includes: receiving first image information from a first camera unit, wherein the first image information includes an image within a first area corresponding to the first camera unit; Based on the first image information, determining whether there is a candidate target currently in the first area by using a frame difference method; In the case where it is determined that there are currently candidate targets in the first area, at least one target to be identified is determined among the candidate targets, and a second camera unit is triggered to capture an image of the at least one target to be identified to obtain second image information; wherein the second image information includes an image in the second area corresponding to the second camera unit; Based on the second image information, the at least one target to be identified is automatically identified; the second camera unit includes N-level second camera subunits, N is an integer greater than 1; the N-level second camera subunits identify the at least one target to be identified in a step-by-step identification manner, and after the k-th level second camera subunit performs image acquisition, it is determined whether the at least one target to be identified has been fully identified; wherein, k∈{1,2,…,N-1}; if the at least one target to be identified has been fully identified, the image acquisition process of the second camera unit is terminated; if there are some unidentified targets among the at least one target to be identified, the k+1-th level second camera subunit is triggered to continue to identify the some unidentified targets; in the step-by-step identification manner, when one level of the second camera subunit is working, the other levels of the second camera subunits suspend acquisition and do not send data to the AI server; Wherein, while receiving the acquisition results of the second camera subunit of each level, direction recognition processing and face recognition processing are performed according to the acquisition results; the direction recognition processing and face recognition processing according to the acquisition results include: based on the frame difference results obtained by the frame difference method, obtaining an image sequence after frame difference processing and performing second-order Volterra filtering, using the results after the second-order Volterra filtering as frame difference feature information, based on the frame difference feature information, using a preset hybrid model to perform direction recognition processing, determining the second camera subunit to be triggered next according to the moving direction of the identified target to be identified, and performing face recognition processing; each level of the second camera subunit includes one or more second low-resolution cameras, one or more second high-resolution cameras and one or more third high-resolution cameras; the second low-resolution camera is used to capture image information containing global motion features, the second high-resolution camera is used to capture image information containing local motion features, and the third high-resolution camera is used to capture images for face recognition; The preset hybrid model adopts a CNN and ViT model fused by a stage-level fusion method, wherein CNN is used to obtain local feature information and ViT is used to obtain global feature information; the training data used by the preset hybrid model includes a first type of training data and a second type of training data, the first type of training data is an image sequence containing local motion features, and the second type of training data is an image sequence containing global motion features, the local corresponds to the head, arms or legs, and the global corresponds to a part or the whole body; according to the result of the automatic recognition, a corresponding operation is performed.
2. The data processing method based on the AI server according to claim 1, characterized in that: The first camera unit includes at least one first camera module; each first camera module is used to acquire images at a predetermined frequency; the first camera unit is suitable for sending first image information to the AI server at a predetermined time interval, and the first image information includes an image sequence consisting of multiple frames of images.
3. The data processing method based on the AI server according to claim 2, characterized in that: In the step of determining whether there is a candidate target currently in the first area, the target movement information in the first area is extracted by subtracting adjacent frames of the image sequence to achieve real-time detection of moving targets in the first area, so that the moving targets appearing in the first area are regarded as candidate targets.
4. The data processing method based on the AI server according to claim 1, characterized in that: The second camera unit includes at least one second camera module.
5. The data processing method based on the AI server according to claim 4, characterized in that: Each second camera module is suitable for capturing an image of a partial area or the entire area within the second area.
6. The data processing method based on the AI server according to claim 4 or 5, characterized in that: Each second camera module is suitable for adjusting shooting parameters according to instructions from the AI server.
7. The data processing method based on the AI server according to claim 6, characterized in that: The shooting parameters include one or more of the following parameters: focus mode; focal length; shooting angle; shutter speed; frame rate; night vision function parameters; and compression format parameters.
8. The data processing method based on the AI server according to claim 1, characterized in that: The second region includes N second sub-regions; The i-th level second camera sub-unit corresponds to the i-th level second sub-region, where i∈{1,2,…,N}.
