CN108229251A - A kind of action identification method and device - Google Patents

Abstract

The present invention relates to smart home field more particularly to a kind of action identification methods and device, and in order to solve the problems, such as that in the prior art blocked because of camera sight cannot accurately carry out action recognition, this method is：First extract the video image in preset time period, determine the area match value of the face in each frame picture that video image includes and apart from match value, obtain area fitting value set and apart from fitting value set, it is fitted in value set from distance, according to preset rules, first is extracted apart from match value and second distance match value, value set is fitted based on area, determine extraction first apart from match value and second distance match value meet preset condition when, it determines that target action occurs in preset time period, in this way, only need the face of detection behavioral agent, can execution in real time identification, it effectively prevents blocking because of camera sight, and it can not determine behavioral agent four limbs and trunk, and then the problem of accurate action recognition can not be carried out.

Description

An action recognition method and device

technical field

The invention relates to the field of smart home, in particular to an action recognition method and device.

Background technique

With the development of smart technology, the smart home industry is also constantly developing, and in smart homes, cameras are essential, for example, through cameras to identify the actions of behavior subjects in home activities to achieve indoor security monitoring , personal security monitoring, personal entertainment and other purposes.

Under the existing technology, when performing action recognition through the camera, it is necessary to extract the whole body features of the subject based on the video images provided by the camera, especially the features of the trunk and lower limbs, and then obtain the movement pattern of the subject by analyzing the features, and then The mapping relationship between video image content and behavior type is established, and based on the mapping relationship, action recognition can be performed.

However, in smart homes, action recognition is usually applied to real-time scenes and has a certain timeliness. Moreover, the actions of actors in real-time scenes have certain complexity and diversity. A series of complex algorithms for motion recognition, the timeliness will be greatly reduced, and it is difficult to meet the application needs of smart homes.

Moreover, in the action recognition method in the prior art, it is necessary to extract the whole body features of the behavior subject, especially the features of the trunk and lower limbs. up, sit down on the edge of the sofa, or stand up), not applicable.

Therefore, it is necessary to design a new action recognition method to overcome the above defects.

Contents of the invention

Embodiments of the present invention provide an action recognition method and device, which are used to solve the problem in the prior art that action recognition cannot be performed accurately due to the occlusion of the camera line of sight.

The specific technical scheme that the embodiment of the present invention provides is as follows:

A method for action recognition, comprising:

Extract video images within a preset time period;

Respectively determine the area fitting value and the distance fitting value of the face in each frame picture included in the video image, and obtain the area fitting value set and the distance fitting value set;

Extracting a first distance fitting value and a second distance fitting value from the distance fitting value set according to preset rules;

When it is determined that the first distance fitting value and the second distance fitting value satisfy a preset condition based on the set of area fitting values, it is determined that a target action occurs within the preset time period.

Optionally, determining the area fitting value and distance fitting value of the face in any frame included in the video image includes:

Perform face detection on any frame of the video image to determine the face area;

Based on the face area, determine the coordinate position of the face in the picture;

Based on the coordinate position, calculate the area value and the distance value of the human face respectively, and the distance value is the distance between the human face and the lower horizontal line of the picture;

Based on the area value and the distance value, an area fitting value corresponding to the area value and a distance fitting value corresponding to the distance value are respectively determined.

Optionally, the extraction of the first distance fitting value and the second distance fitting value from the distance fitting value set according to a preset rule includes:

Recording each distance fitting value in the distance fitting value set to form a discrete distance fitting value curve;

determining a pair of peaks and troughs closest to the starting point of the preset time period;

Extracting the peak and the trough, and determining the peak as the first distance fitting value and the trough as the second distance fitting value respectively.

Optionally, when it is determined that the first distance fitting value and the second distance fitting value satisfy a preset condition based on the set of area fitting values, it is determined that the target action occurs within the preset time period, include:

Respectively determine a first frame moment corresponding to the first distance fitting value and a second frame moment corresponding to the second distance fitting value;

Extracting a first area fitting value corresponding to the second frame moment from the set of area fitting values;

determining that the absolute difference between the second frame time and the first frame time is between a preset first frame time threshold and a preset second frame time threshold; and,

determining the absolute difference between the second distance fitting value and the first distance fitting value, which is between twice the first area fitting value and three times the first area fitting value time;

It is determined that the target action occurs within the preset time period.

Optionally, after determining that the target action occurs within the preset time period, the method further includes:

If the second distance fitting value is higher than the first distance fitting value, then determining that the target action is to stand up;

If the second distance fitting value is lower than the first distance fitting value, it is determined that the target action is sitting down.

An action recognition device, comprising:

a first extraction unit, configured to extract video images within a preset time period;

The first determination unit is configured to respectively determine an area fitting value and a distance fitting value of a human face in each frame of the video image, and obtain a set of area fitting values and a set of distance fitting values;

The second extraction unit is configured to extract a first distance fitting value and a second distance fitting value from the distance fitting value set according to preset rules;

The second determination unit is configured to, based on the set of area fitting values, determine that when the first distance fitting value and the second distance fitting value satisfy a preset condition, determine that the occurrence occurs within the preset time period target action.

Optionally, when determining the area fitting value and the distance fitting value of the face in any frame included in the video image, the first determination unit is used for:

Perform face detection on any frame of the video image to determine the face area;

Based on the face area, determine the coordinate position of the face in the picture;

Based on the coordinate position, calculate the area value and the distance value of the human face respectively, and the distance value is the distance between the human face and the lower horizontal line of the picture;

Based on the area value and the distance value, an area fitting value corresponding to the area value and a distance fitting value corresponding to the distance value are respectively determined.

Optionally, when the first distance fitting value and the second distance fitting value are extracted from the distance fitting value set according to preset rules, the second extraction unit is used for:

Recording each distance fitting value in the distance fitting value set to form a discrete distance fitting value curve;

determining a pair of peaks and troughs closest to the starting point of the preset time period;

Extracting the peak and the trough, and determining the peak as the first distance fitting value and the trough as the second distance fitting value respectively.

Optionally, based on the set of area fitting values, when it is determined that the first distance fitting value and the second distance fitting value meet a preset condition, it is determined that when a target action occurs within the preset time period , the second determining unit is used for:

Respectively determine a first frame moment corresponding to the first distance fitting value and a second frame moment corresponding to the second distance fitting value;

Extracting a first area fitting value corresponding to the second frame moment from the set of area fitting values;

determining that the absolute difference between the second frame time and the first frame time is between a preset first frame time threshold and a preset second frame time threshold; and,

determining the absolute difference between the second distance fitting value and the first distance fitting value, which is between twice the first area fitting value and three times the first area fitting value time;

It is determined that the target action occurs within the preset time period.

Optionally, after determining that the target action occurs within the preset time period, the second determining unit is further configured to:

If the second distance fitting value is higher than the first distance fitting value, then determining that the target action is to stand up;

If the second distance fitting value is lower than the first distance fitting value, it is determined that the target action is sitting down.

In the embodiment of the present invention, the video image within the preset time period is first extracted, the area fitting value and the distance fitting value of the face in each frame included in the video image are determined, and the area fitting value set and the distance fitting value set are obtained. The combined value set, from the distance fitted value set, according to the preset rules, extract the first distance fitted value and the second distance fitted value, based on the area fitted value set, determine the extracted first distance fitted value and When the second distance fitting value satisfies the preset condition, it is determined that the target action occurs within the preset time period. In this way, the action recognition can be completed in real time only by detecting the face of the subject, effectively avoiding the occlusion of the camera line of sight. However, it is impossible to determine the limbs and torso of the subject, and thus cannot perform accurate action recognition.

Description of drawings

Fig. 1 is a structural diagram of an action recognition system in an embodiment of the present invention;

FIG. 2 is a flowchart of an action recognition method in an embodiment of the present invention;

Fig. 3 is an example coordinate position map in the embodiment of the present invention

Fig. 4 is an example graph in the embodiment of the present invention

Fig. 5 is a diagram of an action recognition device in an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

In order to solve the low timeliness of the existing technology for action recognition of the subject in real-time scenarios, and in the home scenario, for scenarios where the lower limbs are covered (such as sitting down or standing up at the dining table, sitting on the edge of the sofa) down or stand up), the problem that motion recognition cannot be performed accurately, in the embodiment of the present invention, a method of motion recognition is designed, the method is to first extract video images within a preset time period, and determine each The area fitting value and the distance fitting value of the face in a frame are obtained, and the area fitting value set and the distance fitting value set are obtained, and the first distance fitting value is extracted from the distance fitting value set according to the preset rules. The combined value and the second distance fitting value are determined based on the area fitting value set, and when the extracted first distance fitting value and the second distance fitting value satisfy a preset condition, it is determined that the target action occurs within a preset time period.

Referring to Fig. 1, in the embodiment of the present invention, a smart camera is used to monitor the monitored area in real time, and the video image in the monitored area is transmitted to the human body motion analysis platform in real time, and the human body motion analysis platform is used for the received video. The image is analyzed and processed in real time to identify the target action of the subject in the monitored area. Among them, the smart camera can transmit the video image to the human motion analysis platform through a wireless network or a wired network.

Specifically, as shown in Figure 2, the human motion analysis platform performs the specific process of motion recognition based on the received video images as follows:

Step 200: The human motion analysis platform extracts video images within a preset time period.

Specifically, the human body motion analysis platform extracts video images within the preset time period according to the preset time period.

The preset time period can be adjusted according to actual needs. For example, if the action of the subject in the video image within half an hour is recognized, then the human motion analysis platform can extract the video image within half an hour. If the movement of the subject in the video image within 10 minutes is recognized, then the human motion analysis platform can extract the video image within 10 minutes.

The preset time period can also be divided according to the ability of the human motion analysis platform to process video images. For example, if the capacity of the human motion analysis platform to process video images at one time is 1T, according to the resolution of the video images to be extracted, the calculation can accommodate time period S, and use this time period S as the preset time period.

In the embodiment of the present invention, there is no limitation on the way of setting the preset time period.

Step 210: The human body motion analysis platform performs face detection on each frame contained in the extracted video image, and determines the face area.

Specifically, the human motion analysis platform splits the extracted video images into frame-by-frame pictures according to the frame format, and uses the face detection algorithm to detect the face of each frame of the picture, and detects the face The face in each frame of the picture is marked, and the face area is determined. Preferably, in the embodiment of the present invention, the face is marked with a box.

Step 220: The human body motion analysis platform determines the coordinate position of the face in each frame based on the face area in each frame.

Specifically, the human motion analysis platform has determined the face area in each frame of the detected face, and determined the coordinate position of the face in each frame based on the determined face area in each frame , as shown in Figure 3 for details.

For example, it has been determined that a human face is detected in the Tth frame picture, and the human face area, that is, the small box, is established as a coordinate point with the upper left vertex of the Tth frame picture, and it is determined to include the human face area (small box) The coordinate position of each vertex of the box), the coordinate group of the small box is {X1 _T , X2 _T , Y1 _T , Y2 _T }, and determine the maximum width X _f and maximum height Y _f of the picture.

Step 230: The human motion analysis platform calculates the area value and the distance value of the face based on the coordinate position of the face in each frame, where the distance value is the distance between the face and the lower horizontal line of the frame.

Specifically, the human motion analysis platform has determined the coordinate position of the human face in each frame of the detected human face, and based on the coordinate position of the human face, the area value and the distance value of the human face are respectively calculated.

Continue to use the above-mentioned example, referring to Fig. 3, calculate the area value and the distance value of the human face in the T-th frame picture, preferably, in the embodiment of the present invention, use the following formula to calculate the area of the human face in the T-th frame picture Value S _T :

S _T ＝ Y2 _T -Y1 _T

Wherein, Y2 _T and Y1 _T are the coordinate values of the face area in the T-th frame on the Y coordinate.

Preferably, in the embodiment of the present invention, the following formula is used to calculate and obtain the distance value H _T of the human face:

H _T ＝Y _f -(Y1 _T +Y2 _T )/2

Among them, Y2 _T and Y1 _T are the coordinate values of the face area in the T-th frame on the Y coordinate, and Y _f is the maximum height value of the T-th frame.

Step 240: Based on the area value and distance value of the face obtained in each frame, the human motion analysis platform determines the corresponding area fitting value and distance fitting value, and obtains a set of area fitting values and a distance fitting value gather.

Specifically, in step 230, the area value and distance value of the face in each frame obtained by the human motion analysis platform are determined based on the corresponding current frame moment. However, when the subject is performing an action, such as , squatting, is usually continuous, therefore, the area value and distance value of the face in the corresponding frame at the current frame time are usually also the area value and distance value of the face in the frame corresponding to the previous frame time, It is even related to the area value and distance value of the face in the picture corresponding to the first two frame moments.

Further, after obtaining the area value and distance value of the human face of a certain frame of picture, combined with the area value and distance value of the human face in the picture corresponding to the previous frame time of the picture, in the picture corresponding to the previous two frame time The area value and distance value of the face, and the area value and distance value of the face in the frame corresponding to the first three frame moments are calculated to obtain the area fitting value and the distance fitting value of the face in the frame.

Preferably, in the embodiment of the present invention, still taking the T-th frame as an example, the following formula is used to obtain the distance fitting value H _T of the human face in the frame corresponding to the T-th frame:

Among them, H _T is the distance value of the human face in the picture corresponding to the T frame time, H _T-1 is the distance value of the human face in the picture corresponding to the (T-1) frame time, that is, the previous T frame time The distance value of the face in the picture corresponding to the frame time, H _T-2 is the distance value of the face in the picture corresponding to the (T-2)th frame time, that is, the distance value of the face in the picture corresponding to the first two frame times of the T-th frame time The distance value of the human face, H _T-3 is the distance value of the human face in the picture corresponding to the (T-3) frame time, that is, the distance value of the human face in the picture corresponding to the first three frame times of the T-th frame time.

Preferably, in the embodiment of the present invention, still taking the T-th frame as an example, the following formula is used to obtain the fitting value SC _T of the face area in the frame corresponding to the T-th frame:

Among them, S _T is the area value of the face in the picture corresponding to the T-th frame time, SC _T-1 is the distance fitting value of the face in the picture corresponding to the (T-1) frame time, that is, the T-th frame time The area value of the face in the frame corresponding to the previous frame moment.

So far, after the above steps, the distance fitting value and the area fitting value of the face in the frame corresponding to the Tth frame time are obtained. The distance fitting value and area fitting value of the face, for each frame of the picture that has been determined to have a human face, the distance fitting value and area fitting value of the respective face are obtained by the above method, and the obtained several distances The fitted values form a set of fitted distance values, and the obtained several fitted values of area form a set of fitted values of area.

For example, assuming that it is currently determined that a human face is detected in frames 8-10, if the distance fitting value and area fitting value of the face in the 8th frame are HC ₈ and SC ₈ respectively, if the 9th frame The distance fitting value and area fitting value of the face in are HC ₉ and SC ₉ respectively, if the distance fitting value and area fitting value of the face in the tenth frame are HC ₁₀ and SC ₁₀ respectively, then , the obtained distance fitting value set is {HC ₈ , HC ₉ , HC ₁₀ }, and the obtained area fitting value set is {SC ₈ , SC ₉ , SC ₁₀ }.

Step 250: Based on the obtained distance fitting value set, the human motion analysis platform extracts a first distance fitting value and a second distance fitting value according to preset rules.

Specifically, the human motion analysis platform records each distance fitting value in the obtained distance fitting value set to form a discrete distance fitting value curve, and determines the pair closest to the starting point of the preset time period in the curve The peak and the trough are extracted, the peak and the trough are extracted, the peak is determined as the first distance fitting value, and the trough is determined as the second distance fitting value.

For example, assuming that in step 210, the video image is split into m frames, the distance fitting values corresponding to each of the m frames are: HC _T-m+1 , HC _T-m+2 , ..., HC _{T- 2} , HC _T-1 , HC _T , the above m distance fitting values are combined into a discrete distance fitting value curve, wherein the frame time corresponding to HC _T is the starting point of the preset time period, as shown in Figure 4 for details .

Further, it is determined that in the distance fitting value curve, a pair of peaks and troughs closest to the T frame time, corresponding to Figure 4, can be determined (T-1) frame time and (T-2) frame time corresponding to each The distance fitting value is the closest pair of peaks and troughs, that is, determine the distance fitting value HC corresponding to the (T-1) frame time _T-1 is the peak, and determine the distance fitting value corresponding to the (T-2) frame time HC _T-2 for the trough.

Step 260: Based on the obtained set of area fitting values, the human motion analysis platform determines that the target action occurs within a preset time period when the extracted first distance fitting value and second distance fitting value satisfy a preset condition.

Specifically, respectively determine the first frame time corresponding to the first distance fitting value and the second frame time corresponding to the second distance fitting value, and extract the first frame time corresponding to the second frame time from the obtained area fitting value set An area fitting value to determine the absolute difference between the second frame time and the first frame time, which is located between the preset first frame time threshold and the second second frame time threshold, and determine the second distance When the absolute difference between the fitted value and the first distance fitted value is between twice the first area fitted value and three times the first area fitted value, it is determined at a preset time The target action occurs within the segment.

In the embodiment of the present invention, the preset first frame time threshold and the second frame time threshold are selected experience values of the human body when performing the target action, and are obtained through long-term training. Similarly, when judging the second distance fitting value and For the absolute difference between the first distance fitting values, the selected "double" and "triple" are also obtained during long-term training to perform the target action, and can be adjusted dynamically.

Preferably, in the embodiment of the present invention, the following formula is used to determine whether the target action occurs within the preset time period:

t _min ≤ |LK| ≤ t _max ; and,

2×SC _L ≤|HC _L -HC _K |≤3×SC _L

Wherein, in Formula 1, t _min is the preset first frame time threshold, t _max is the preset second frame time threshold, K is the frame time corresponding to the wave peak (the first distance fitting value), and L is the wave trough (the second distance fitting value) corresponds to the frame moment.

Wherein, in Formula 2, H _K is the first distance fitting value, _HCL is the second distance fitting value, and _SCL is the first area fitting value.

Furthermore, after it is determined that the target action is taken by the behavior subject, the relationship between the second distance fitting value and the first distance fitting value can be used to further judge whether the target action is standing up or sitting down, if the second distance fitting value When the fitting value of the second distance is higher than the fitting value of the first distance, it is determined that the target action is to stand up; if the fitting value of the second distance is lower than the fitting value of the first distance, it is determined that the target action is to sit down.

Preferably, in the embodiment of the present invention, after determining that the target action occurs, the following formula is used to further determine whether the target action is standing up or sitting down:

Wherein, HC _K is the fitting value of the first distance, and _HCL is the fitting value of the second distance.

Referring to Fig. 5, in the embodiment of the present application, the action recognition device includes at least a first extraction unit 50, a first determination unit 51, a second extraction unit 52 and a second determination unit 53, wherein,

The first extracting unit 50 is used to extract video images within a preset time period;

The first determining unit 51 is configured to respectively determine an area fitting value and a distance fitting value of a human face in each frame of the video image, and obtain a set of area fitting values and a set of distance fitting values;

The second extraction unit 52 is configured to extract a first distance fitting value and a second distance fitting value from the distance fitting value set according to preset rules;

The second determining unit 53 is configured to determine that the first distance fitting value and the second distance fitting value meet a preset condition based on the set of area fitting values, and determine that within the preset time period Target action occurs.

Optionally, when determining the area fitting value and distance fitting value of the face in any frame included in the video image, the first determining unit 51 is configured to:

Perform face detection on any frame of the video image to determine the face area;

Based on the face area, determine the coordinate position of the face in the picture;

Based on the coordinate position, calculate the area value and the distance value of the human face respectively, and the distance value is the distance between the human face and the lower horizontal line of the picture;

Based on the area value and the distance value, an area fitting value corresponding to the area value and a distance fitting value corresponding to the distance value are respectively determined.

Optionally, when the first distance fitting value and the second distance fitting value are extracted from the distance fitting value set according to preset rules, the second extracting unit 52 is configured to:

Recording each distance fitting value in the distance fitting value set to form a discrete distance fitting value curve;

determining a pair of peaks and troughs closest to the starting point of the preset time period;

Extracting the peak and the trough, and determining the peak as the first distance fitting value and the trough as the second distance fitting value respectively.

Optionally, based on the set of area fitting values, when it is determined that the first distance fitting value and the second distance fitting value meet a preset condition, it is determined that when a target action occurs within the preset time period , the second determining unit 53 is used for:

Respectively determine a first frame moment corresponding to the first distance fitting value and a second frame moment corresponding to the second distance fitting value;

Extracting a first area fitting value corresponding to the second frame moment from the set of area fitting values;

determining that the absolute difference between the second frame time and the first frame time is between a preset first frame time threshold and a preset second frame time threshold; and,

determining the absolute difference between the second distance fitting value and the first distance fitting value, which is between twice the first area fitting value and three times the first area fitting value time;

It is determined that the target action occurs within the preset time period.

Optionally, after determining that the target action occurs within the preset time period, the second determining unit 53 is further configured to:

If the second distance fitting value is higher than the first distance fitting value, then determining that the target action is to stand up;

If the second distance fitting value is lower than the first distance fitting value, it is determined that the target action is sitting down.

In the embodiment of the present invention, the video image within the preset time period is first extracted, the area fitting value and the distance fitting value of the face in each frame included in the video image are determined, and the area fitting value set and the distance fitting value set are obtained. The combined value set, from the distance fitted value set, according to the preset rules, extract the first distance fitted value and the second distance fitted value, based on the area fitted value set, determine the extracted first distance fitted value and When the second distance fitting value satisfies the preset condition, it is determined that the target action occurs within the preset time period. In this way, the action recognition can be completed in real time only by detecting the face of the subject, effectively avoiding the occlusion of the camera line of sight. However, it is impossible to determine the limbs and torso of the subject, and thus cannot perform accurate action recognition. At the same time, it also effectively avoids the problem of being unable to perform accurate action recognition due to the difference in height of the subject and the distance of the camera.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods, systems, or computer program products. Accordingly, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

While preferred embodiments of the invention have been described, additional changes and modifications to these embodiments can be made by those skilled in the art once the basic inventive concept is appreciated. Therefore, it is intended that the appended claims be construed to cover the preferred embodiment as well as all changes and modifications which fall within the scope of the invention.

Apparently, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. In this way, if the modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and equivalent technologies, the present invention also intends to include these modifications and variations.

Claims (10)

1. An action recognition method, characterized in that, comprising: Extract video images within a preset time period; Respectively determine the area fitting value and the distance fitting value of the face in each frame picture included in the video image, and obtain the area fitting value set and the distance fitting value set; Extracting a first distance fitting value and a second distance fitting value from the distance fitting value set according to preset rules; When it is determined that the first distance fitting value and the second distance fitting value satisfy a preset condition based on the set of area fitting values, it is determined that a target action occurs within the preset time period. 2. The method according to claim 1, wherein determining the area fitting value and the distance fitting value of the face in any one frame of the picture included in the video image comprises: Perform face detection on any frame of the video image to determine the face area; Based on the face area, determine the coordinate position of the face in the picture; Based on the coordinate position, calculate the area value and the distance value of the human face respectively, and the distance value is the distance between the human face and the lower horizontal line of the picture; Based on the area value and the distance value, an area fitting value corresponding to the area value and a distance fitting value corresponding to the distance value are respectively determined. 3. The method according to claim 1, wherein the first distance fitting value and the second distance fitting value are extracted from the distance fitting value set according to preset rules, comprising: Recording each distance fitting value in the distance fitting value set to form a discrete distance fitting value curve; determining a pair of peaks and troughs closest to the starting point of the preset time period; Extracting the peak and the trough, and determining the peak as the first distance fitting value and the trough as the second distance fitting value respectively. 4. The method according to any one of claims 1-3, wherein, based on the set of area fitting values, it is determined that the first distance fitting value and the second distance fitting value satisfy a preset conditions, determine that the target action occurs within the preset time period, including: Respectively determine a first frame moment corresponding to the first distance fitting value and a second frame moment corresponding to the second distance fitting value; Extracting a first area fitting value corresponding to the second frame moment from the set of area fitting values; determining that the absolute difference between the second frame time and the first frame time is between a preset first frame time threshold and a preset second frame time threshold; and, determining the absolute difference between the second distance fitting value and the first distance fitting value, which is between twice the first area fitting value and three times the first area fitting value time; It is determined that the target action occurs within the preset time period. 5. The method according to claim 4, further comprising: If the second distance fitting value is higher than the first distance fitting value, then determining that the target action is to stand up; If the second distance fitting value is lower than the first distance fitting value, it is determined that the target action is sitting down. 6. An action recognition device, characterized in that it comprises: a first extraction unit, configured to extract video images within a preset time period; The first determination unit is configured to respectively determine an area fitting value and a distance fitting value of a human face in each frame of the video image, and obtain a set of area fitting values and a set of distance fitting values; The second extraction unit is configured to extract a first distance fitting value and a second distance fitting value from the distance fitting value set according to preset rules; The second determination unit is configured to, based on the set of area fitting values, determine that when the first distance fitting value and the second distance fitting value satisfy a preset condition, determine that the occurrence occurs within the preset time period target action. 7. The device according to claim 6, wherein when determining the area fitting value and the distance fitting value of the human face in any frame of the video image, the first determining unit is used to : Perform face detection on any frame of the video image to determine the face area; Based on the face area, determine the coordinate position of the face in the picture; Based on the coordinate position, calculate the area value and the distance value of the human face respectively, and the distance value is the distance between the human face and the lower horizontal line of the picture; Based on the area value and the distance value, an area fitting value corresponding to the area value and a distance fitting value corresponding to the distance value are respectively determined. 8. The device according to claim 6, wherein when the first distance fitting value and the second distance fitting value are extracted from the distance fitting value set according to preset rules, the The second extraction unit is used for: Recording each distance fitting value in the distance fitting value set to form a discrete distance fitting value curve; determining a pair of peaks and troughs closest to the starting point of the preset time period; Extracting the peak and the trough, and determining the peak as the first distance fitting value and the trough as the second distance fitting value respectively. 9. The device according to any one of claims 6-8, wherein, based on the set of area fitting values, it is determined that the first distance fitting value and the second distance fitting value satisfy a preset condition, when it is determined that the target action occurs within the preset time period, the second determining unit is used to: Respectively determine a first frame moment corresponding to the first distance fitting value and a second frame moment corresponding to the second distance fitting value; Extracting a first area fitting value corresponding to the second frame moment from the set of area fitting values; determining that the absolute difference between the second frame time and the first frame time is between a preset first frame time threshold and a preset second frame time threshold; and, determining the absolute difference between the second distance fitting value and the first distance fitting value, which is between twice the first area fitting value and three times the first area fitting value time; It is determined that the target action occurs within the preset time period. 10. The device according to claim 9, wherein after determining that the target action occurs within the preset time period, the second determining unit is further configured to: If the second distance fitting value is higher than the first distance fitting value, then determining that the target action is to stand up; If the second distance fitting value is lower than the first distance fitting value, it is determined that the target action is sitting down.