A multi-camera human action recognition method

Abstract

This invention discloses a multi-camera human motion recognition method, comprising: acquiring multiple sets of human motion videos using a multi-camera system; performing human motion recognition on the human motion video frames to generate a human motion decomposition feature queue and a human continuous motion recognition result queue; comparing whether the motion order in the human motion decomposition feature queue and the human continuous motion recognition result queue is consistent to obtain the human motion recognition result. This invention can obtain blind-spot-free motion recognition results for human motions; different joint features of a motion can be described using standardized coordinates from different cameras, achieving blind-spot-free recognition of motion joint features.

Description

Technical Field

[0001] This invention relates to the field of human motion recognition technology, and specifically to a multi-camera human motion recognition method. Background Technology

[0002] Existing human motion recognition systems mainly describe human motion by training human models on videos or images. However, during the recognition process, due to the incomplete recognition of human limbs by a single camera and omissions caused by limb occlusion, it is easy to encounter the phenomenon of being unable to recognize features when recognizing human motion. Therefore, it is necessary to supplement and enhance the information of single-camera human motion recognition in order to more accurately recognize continuous human motion. Summary of the Invention

[0003] To address the aforementioned shortcomings in the existing technology, this invention provides a multi-camera human motion recognition method.

[0004] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0005] A multi-camera human motion recognition method includes the following steps:

[0006] A multi-camera system was used to capture multiple sets of human motion videos;

[0007] Perform human motion recognition on human motion video frames to generate a human motion decomposition feature queue and a human continuous motion recognition result queue.

[0008] Compare the sequence of actions in the human motion decomposition feature queue and the human continuous motion recognition result queue to see if they are consistent. If they are, the sequence of decomposed actions meets the target; otherwise, the sequence of decomposed actions does not meet the target.

[0009] Furthermore, before acquiring multiple sets of human motion images using a multi-camera system, the following steps are included:

[0010] The standard start-up time method is used to synchronize the video recording time of the multi-camera system, and a ruler with multiple equal standard length sides is set at the center point of the monitored area;

[0011] The standard length side of the ruler is perpendicular to the lens axis of the camera being measured. The number of pixels on multiple equal sides of the ruler is measured within the monitoring field of view of the center camera of the camera being measured. The number of pixels on the standard length side of the first camera being measured is used as the benchmark. The ratio of the number of pixels on the standard length side of each camera to the number of pixels on the standard length side of the first camera being measured is used as the camera correlation coefficient. The coordinates of the center point of the ruler are recorded and used as the origin of the field of view of each camera.

[0012] Furthermore, the step of performing human motion recognition on human motion video frames to generate a human motion decomposition feature queue and a human continuous motion recognition result queue specifically includes the following steps:

[0013] The pixel displacement of a single joint position in two frames with the same frame interval is used as the main motion description factor. The standard coordinates of the joint points of each camera are determined based on the single joint pixel coordinates, the camera correlation coefficient, and the camera's field of view origin.

[0014] A two-dimensional description of human single-joint motion is generated based on the pixel variation and ratio of the two-dimensional coordinates of the joints.

[0015] Based on the starting time of each decomposed action, a two-dimensional description of the human body's decomposed actions is generated according to the two-dimensional description of the human body's single joint motion, which includes each joint point.

[0016] Based on the two-dimensional description of each decomposed human action in the continuous human action decomposition, a human action decomposition feature queue is generated.

[0017] A queue of continuous human motion recognition results is generated based on the motion time of human actions.

[0018] Furthermore, the step of generating a two-dimensional description of human single-joint motion based on the pixel variation and scaling ratio of the two-dimensional coordinates of the joints specifically includes the following steps:

[0019] Select the same frame from each camera, and determine the standard displacement of the single joint termination point of the human single joint movement based on the pixel distance between the two joint points of the starting frame.

[0020] Select the same frame from each camera, and determine the pixel displacement of the single joint termination point of the human single joint movement based on the pixel distance from the single joint start point to the single joint termination point in the starting frame.

[0021] A two-dimensional description of human single-joint motion is generated based on the human joint number, the start time of human motion, the end time of human motion, the coordinates of the start point of human single-joint motion, the standard displacement of the end point of human single-joint motion, and the pixel displacement of the end point of human single-joint motion.

[0022] Furthermore, the two-dimensional description of human single-joint motion is specifically represented as follows:

[0023] A(K0,t0,△t):{K0,CamID,t0,△t,(X0,Y0),(△STDX,△STDY),(△X,△Y)}

[0024] Where A(K0,t0,△t) represents a two-dimensional description of the movement of human joint K0 from the starting time point t0 to the ending time point △t, CamID represents the camera number, (X0,Y0) represents the coordinates of the starting point of the human single joint movement, (△STDX,△STDY) represents the standard displacement of the single joint ending point of the human single joint movement, and (△X,△Y) represents the pixel displacement of the single joint ending point of the human single joint movement.

[0025] Furthermore, the two-dimensional description of the human body's decomposed movements is specifically represented as follows:

[0026] ActBase(t0): {A(K0,t0,△t),A(K1,t0,△t),...,A(K 24 ,t0,△t)}

[0027] Where ActBase(t0) represents a two-dimensional description of the human body's decomposed movements at the starting time point t0, A(K0,t0,△t) represents a two-dimensional description of the movement of human joint K0 from the starting time point t0 to the ending time point △t, and A(K1,t0,△t) represents a two-dimensional description of the movement of human joint K1 from the starting time point t0 to the ending time point △t. 24 (t0, △t) represents the human body joint point K. 24 A two-dimensional description of the motion from the starting time point t0 to the ending time point Δt.

[0028] Furthermore, the human motion decomposition feature queue is specifically represented as follows:

[0029] ActArray:{ActNo,CamID,ActOrder,ActBase(t0)}

[0030] Where ActArray represents the human motion decomposition feature queue, ActNo represents the decomposition motion number, CamID represents the camera number, ActOrder represents the sequence number of the decomposition motion in the continuous motion, and ActBase(t0) represents the two-dimensional description of the human motion decomposition at the starting time point t0.

[0031] Furthermore, the queue of human continuous motion recognition results is specifically represented as follows:

[0032] ActResult:{ActNo,ActRealOrder,ActTime}

[0033] Wherein, ActResult represents the queue of human continuous motion recognition results, ActNo represents the decomposed motion number, ActRealOrder represents the sequence number of the decomposed motions in human continuous motion recognition, and ActTime represents the decomposed motion time.

[0034] Furthermore, the step of comparing whether the action order of the human motion decomposition feature queue and the human continuous motion recognition result queue is consistent specifically includes:

[0035] The human motion decomposition feature queue is sorted according to the order of decomposed motions, and the human continuous motion recognition result queue is sorted according to the order of motion recognition. The sorting results of the human motion decomposition feature queue and the order results of the human continuous motion recognition result queue are compared to see if they are consistent.

[0036] The present invention has the following beneficial effects:

[0037] This invention proposes a multi-camera human motion recognition method. By using multiple cameras arranged in a specific manner, it is possible to obtain blind-spot-free motion recognition results for human motion. The different joint features of a motion can be described using the standardized coordinates of different cameras, thus achieving blind-spot-free recognition of motion joint features. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating a multi-camera human motion recognition method according to the present invention.

[0039] Figure 2 This is a schematic diagram illustrating the configuration of each camera group containing one camera in this invention;

[0040] Figure 3 This is a schematic diagram illustrating the configuration of each camera group containing multiple cameras in this invention. Detailed Implementation

[0041] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0042] like Figures 1 to 3 As shown, this embodiment of the invention provides a multi-camera human motion recognition method, including the following steps S1 to S3:

[0043] S1. Employ a multi-camera system to capture multiple sets of human motion videos;

[0044] In an optional embodiment of the present invention, at least a plurality of cameras 1, 2, ..., N are set in the motion monitoring area, wherein the position and angle of each camera can be flexibly configured to facilitate the identification of specific decomposed actions.

[0045] Once the cameras are set up, the focal length and position of all cameras should be fixed and not changed. If changes are made, a second calibration is required.

[0046] Before this embodiment uses a multi-camera system to acquire multiple sets of human motion images, it includes the following:

[0047] The standard start-up time method is used to synchronize the video recording time of the multi-camera system, and a ruler with multiple equal standard length sides is set at the center point of the monitored area;

[0048] The standard length side of the ruler is perpendicular to the lens axis of the camera being measured. The number of pixels on multiple equal sides of the ruler is measured within the monitoring field of view of the center camera of the camera being measured. The number of pixels on the standard length side of the first camera being measured is used as the benchmark. The ratio of the number of pixels on the standard length side of each camera to the number of pixels on the standard length side of the first camera being measured is used as the camera correlation coefficient. The coordinates of the center point of the ruler are recorded and used as the origin of the field of view of each camera.

[0049] Specifically, in this embodiment, the camera recording time is synchronized before the motion monitoring begins. The synchronization method is the standard start-up time method, which uses the standard start-up time as the recording start time for multiple cameras, and the shooting frame rate of multiple cameras is the same. The shooting frame rate is set to no less than 24 frames per second.

[0050] An object 7, serving as a benchmark, is placed at the center point of the monitored area. Object 7 needs to have multiple sides of equal standard length, and can be a regular polygon such as an equilateral triangle, square, pentagon, or hexagon. The standard length sides must be perpendicular to the lens axis of the camera being measured. During measurement, the object can be rotated around the center point of the regular polygon as needed to ensure the standard length sides are perpendicular to the lens axis of the camera. The number of pixels (StdPix1, StdPix2, ..., StdPix) of the multiple equal sides of object 7 are measured within the monitoring field of view of the center camera of the camera being measured. n Using the standard length side pixel count StdPix1 of the first measured camera as a benchmark, the standard length side pixel count StdPix of each camera... n The ratio of the number of pixels on the standard length side of the first measured camera to the number of pixels on the second camera is expressed as the camera group correlation coefficient: Res n =StdPix n / StdPix1. Record the center point of object 7 as P0, and record them as (XT0, YT0), (XT1, YT1), ..., (XT1, YT1). n YT n Each of these points serves as the origin of the field of view for each camera.

[0051] The main camera is selected as the reference camera. Each camera in the auxiliary camera group has its correlation coefficient with the main camera calculated according to method 3.1, resulting in Res0, Res1, ..., Res n .

[0052] S2. Perform human motion recognition on human motion video frames to generate a human motion decomposition feature queue and a human continuous motion recognition result queue.

[0053] In an optional embodiment of the present invention, step S2 specifically includes the following steps:

[0054] S21. Using the pixel displacement of the single joint position in two frames with the same number of frames as the main action description factor, determine the standard coordinates of the joint points of each camera based on the single joint pixel coordinates, camera correlation coefficient and camera field of view origin.

[0055] Specifically, this embodiment uses 25 joint points of the human body as basic descriptive points to describe human movements. These 25 basic points include: 0: lower spine; 1: midpoint of spine; 2: neck; 3: head; 4: left shoulder joint; 5: left elbow joint; 6: left wrist joint; 7: left palm tip; 8: right shoulder joint; 9: right elbow joint; 10: right wrist joint; 11: right palm tip; 12: left hip joint; 13: left knee joint; 14: left ankle joint; 15: left toe; 16: right hip joint; 17: right knee joint; 18: right ankle joint; 19: right toe; 20: upper spine; 21: left middle finger tip; 22: left thumb tip; 23: right middle finger tip; 24: right thumb tip. The order and numbering of these 25 joint points can be arbitrary.

[0056] This embodiment uses frame images with the same number of frames from the same group of cameras to identify the position of each joint. OpenPose can be used for human joint motion recognition and position extraction. The pixel displacement of the single joint position in two frames with the same number of frames interval is used as the main motion description factor. The standard coordinates of the joint points of each camera are standardized using the following formula:

[0057] Xn=(X-XT n )*Res n Yn=(Y-YT) n )*Res n .

[0058] Where (Xn,Yn) are the standard coordinates of the joints, (X,Y) are the pixel coordinates of a single joint, and (XT) are the standard coordinates of the joints. n YT n Res is the origin of the camera's field of view. n This represents the camera correlation coefficient.

[0059] S22. Generate a two-dimensional description of human single-joint motion based on the pixel changes and scaling ratios of the two-dimensional coordinates of the joints; specifically including the following steps:

[0060] S221. Select the same frame from each camera and determine the standard displacement of the single joint termination point of the human single joint movement based on the pixel distance between the two joint points of the starting frame.

[0061] Specifically, in this embodiment, the standard displacement is a measurement method using the distance between an individual's body parts as the standard unit. The standard displacement metric value is based on the standard distance, which can be uniformly set as the distance from the cervical joint to the cephalic joint during movement. Alternatively, different standard distances can be used for different limb types, such as arms, hands, legs, and torsos. For example, for the arm, the distance from the elbow joint to the shoulder joint can be used as the standard distance. For the values ​​of (△STDX, △STDY, △STDZ), the standard metric for each frame needs to be obtained. When the standard metric is the pixel distance between two joints in the starting frame, the same number of frames from each camera group are used to calculate the standard displacement.

[0062] S222. Select the same frame from each camera, and determine the pixel displacement of the single joint termination point of the human single joint movement based on the pixel distance from the single joint start point to the single joint end point of the starting frame.

[0063] Specifically, in this embodiment, (△X, △Y) represents the pixel displacement of the single joint termination point of the action, and the pixel displacement is based solely on the pixel distance from the start point to the termination point in the monitoring image.

[0064] S223. Generate a two-dimensional description of human single-joint motion based on the human joint number, the start time of human motion, the end time of human motion, the coordinates of the start point of human single-joint motion, the standard displacement of the end point of human single-joint motion, and the pixel displacement of the end point of human single-joint motion.

[0065] Specifically, in this embodiment, Δt is marked with the number of frames between two frames multiplied by the frame rate, and t0 is marked with the number of frames in the starting frame multiplied by the frame rate. For the joint designated K0, its unit action (Act) is described as follows:

[0066] A(K0,t0,△t):{K0,CamID,t0,△t,(X0,Y0),(△STDX,△STDY),(△X,△Y)}

[0067] Where A(K0,t0,△t) represents a two-dimensional description of the movement of human joint K0 from the starting time point t0 to the ending time point △t, CamID represents the camera number, (X0,Y0) represents the coordinates of the starting point of the human single joint movement, (△STDX,△STDY) represents the standard displacement of the single joint ending point of the human single joint movement, and (△X,△Y) represents the pixel displacement of the single joint ending point of the human single joint movement.

[0068] S23. Based on the starting time of each decomposed action, generate a two-dimensional description of the human body's decomposed actions according to the two-dimensional description of the human body's single joint motion, which includes each joint point.

[0069] Specifically, in this embodiment, an action is decomposed into several decomposed actions ActBase(t), with the start time of each decomposed action being t0. The 25 joints of the human body are recorded using the single-joint motion description method in step S2 as a description of one decomposed action, represented as follows:

[0070] ActBase(t0): {A(K0,t0,△t),A(K1,t0,△t),...,A(K 24 ,t0,△t)}

[0071] Where ActBase(t0) represents a two-dimensional description of the human body's decomposed movements at the starting time point t0, A(K0,t0,△t) represents a two-dimensional description of the movement of human joint K0 from the starting time point t0 to the ending time point △t, and A(K1,t0,△t) represents a two-dimensional description of the movement of human joint K1 from the starting time point t0 to the ending time point △t. 24 (t0, △t) represents the human body joint point K. 24 A two-dimensional description of the motion from the starting time point t0 to the ending time point Δt.

[0072] S24. Based on the two-dimensional description of each decomposed human action in the continuous human action decomposition, generate a human action decomposition feature queue, specifically represented as follows:

[0073] ActArray:{ActNo,CamID,ActOrder,ActBase(t0)}

[0074] Where ActArray represents the human motion decomposition feature queue, ActNo represents the decomposition motion number, CamID represents the camera number, ActOrder represents the sequence number of the decomposition motion in the continuous motion, and ActBase(t0) represents the two-dimensional description of the human motion decomposition at the starting time point t0.

[0075] S25. Generate a queue of continuous human motion recognition results based on the motion time of human motions.

[0076] Specifically, in this embodiment, when a camera detects a decomposed action, the action time of the decomposed action is returned, and a queue of continuous human action recognition results is generated, which is specifically represented as follows:

[0077] ActResult:{ActNo,ActRealOrder,ActTime}

[0078] Wherein, ActResult represents the queue of human continuous motion recognition results, ActNo represents the decomposed motion number, ActRealOrder represents the sequence number of the decomposed motions in human continuous motion recognition, and ActTime represents the decomposed motion time.

[0079] In an optional embodiment of the present invention, this embodiment generates a queue of continuous human motion recognition results based on the action time of human motions, specifically as follows:

[0080] ActResult:{ActNo,ActRealOrder,ActTime,ActBase(t0),ActTimeAnylyRes,

[0081] ActAccuAnylyRes}

[0082] Wherein, ActResult represents the queue of human continuous action recognition results, ActNo represents the decomposed action number, ActRealOrder represents the sequence number of the decomposed action in human continuous action recognition, ActTime represents the decomposed action time, ActBase(t0) represents the two-dimensional description of the human decomposed action at the starting time point t0, ActTimeAnylyRes represents the decomposed action time evaluation coefficient, and ActAccuAnylyRes represents the decomposed action accuracy evaluation coefficient.

[0083] In an optional embodiment of the present invention, this embodiment further includes evaluating the decomposition actions after determining the sequentiality of the decomposition actions, specifically including:

[0084] The action time evaluation coefficient is calculated by comparing the actual Δt in ActBase(t0) of ActResult with the Δt of the same ActNo action in ActBase(t0) of ActArray. The time evaluation coefficient can be calculated as a ratio (Δt in ActResult / Δt in ActArray) or as a difference (Δt in ActResult - Δt in ActArray) and saved to ActTimeAnylyRes.

[0085] The decomposed action time evaluation coefficient is calculated based on the actual (△STDX, △STDY) in ActBase(t0) of the result ActResult and the (△STDX, △STDY) of the same ActNo action in ActBase(t0) of ActArray. The accuracy evaluation coefficient can be calculated as a ratio (△STDX in ActResult / △STDX in ActArray + △STDY in ActResult / △STDY in ActArray) / 2, or as a geometric difference. The formula is as follows: The calculated results are saved to ActAccuAnylyRes.

[0086] In an optional embodiment of the present invention, this embodiment further includes performing a comprehensive evaluation of the decomposed actions after the evaluation of the decomposed actions, specifically including:

[0087] The overall evaluation of the action is calculated based on the ActTimeAnylyRes and ActAccuAnylyRes values ​​for each decomposed action in ActResult. The evaluation method can be either the average of ActTimeAnylyRes or the average of ActAccuAnylyRes.

[0088] S3. Compare the sequence of actions in the human motion decomposition feature queue and the human continuous motion recognition result queue to see if they are consistent. If they are, the sequence of decomposed actions meets the target; otherwise, the sequence of decomposed actions does not meet the target.

[0089] In an optional embodiment of the present invention, step S3 specifically includes:

[0090] The human motion decomposition feature queue is sorted according to the order of decomposed motions, and the human continuous motion recognition result queue is sorted according to the order of motion recognition. The sorting results of the human motion decomposition feature queue and the order results of the human continuous motion recognition result queue are compared to see if they are consistent.

[0091] This 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 will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0094] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0095] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A multi-camera human motion recognition method, characterized in that, Includes the following steps: A multi-camera system was used to capture multiple sets of human motion videos; Perform human motion recognition on human motion video frames to generate a human motion decomposition feature queue and a human continuous motion recognition result queue. Specifically, the following steps are included: The pixel displacement of a single joint position in two frames with the same frame interval is used as the main motion description factor. The standard coordinates of the joint points of each camera are determined based on the single joint pixel coordinates, the camera correlation coefficient, and the camera's field of view origin. A two-dimensional description of human single-joint motion is generated based on the pixel variation and ratio of the two-dimensional coordinates of the joints. Based on the starting time of each decomposed action, a two-dimensional description of the human body's decomposed actions is generated according to the two-dimensional description of the human body's single joint motion, which includes each joint point. Based on the two-dimensional description of each decomposed human action in the continuous human action decomposition, a human action decomposition feature queue is generated. Generate a queue of continuous human motion recognition results based on the motion time of human actions. Compare the sequence of actions in the human motion decomposition feature queue and the human continuous motion recognition result queue to see if they are consistent. If they are, the sequence of decomposed actions meets the target; otherwise, the sequence of decomposed actions does not meet the target.

2. The multi-camera human motion recognition method according to claim 1, characterized in that, Before acquiring multiple sets of human motion images using a multi-camera system, the following steps are included: The standard start-up time method is used to synchronize the video recording time of the multi-camera system, and a ruler with multiple equal standard length sides is set at the center point of the monitored area; The standard length side of the ruler is perpendicular to the lens axis of the camera being measured. The number of pixels on multiple equal sides of the ruler is measured within the monitoring field of view of the center camera of the camera being measured. The number of pixels on the standard length side of the first camera being measured is used as the benchmark. The ratio of the number of pixels on the standard length side of each camera to the number of pixels on the standard length side of the first camera being measured is used as the camera correlation coefficient. The coordinates of the center point of the ruler are recorded and used as the origin of the field of view of each camera.

3. The multi-camera human motion recognition method according to claim 1, characterized in that, The process of generating a two-dimensional description of human single-joint motion based on the pixel variation and scaling ratio of the two-dimensional coordinates of joint points specifically includes the following steps: Select the same frame from each camera, and determine the standard displacement of the single joint termination point of the human single joint movement based on the pixel distance between the two joint points of the starting frame. Select the same frame from each camera, and determine the pixel displacement of the single joint termination point of the human single joint movement based on the pixel distance from the single joint start point to the single joint termination point in the starting frame. A two-dimensional description of human single-joint motion is generated based on the human joint number, the start time of human motion, the end time of human motion, the coordinates of the start point of human single-joint motion, the standard displacement of the end point of human single-joint motion, and the pixel displacement of the end point of human single-joint motion.

4. The multi-camera human motion recognition method according to claim 1, characterized in that, The two-dimensional description of human single-joint motion is specifically represented as follows: A(K0,t0,△t):{K0,CamID,t0,△t,(X0,Y0),(△STDX,△STDY),(△X,△Y)} Where A(K0,t0,△t) represents a two-dimensional description of the movement of human joint K0 from the starting time point t0 to the ending time point △t, CamID represents the camera number, (X0, Y0) represents the coordinates of the starting point of the human single joint movement, (△STDX,△STDY) represents the standard displacement of the single joint ending point of the human single joint movement, and (△X,△Y) represents the pixel displacement of the single joint ending point of the human single joint movement.

5. The multi-camera human motion recognition method according to claim 1, characterized in that, The two-dimensional description of the human body's decomposition movements is specifically represented as follows: ActBase(t0):{ A(K0,t0,△t), A(K1,t0,△t),..., A(K 24 ,t0,△t)} Where ActBase(t0) represents a two-dimensional description of the human body's decomposed movements at the starting time point t0, A(K0,t0,△t) represents a two-dimensional description of the movement of human joint K0 from the starting time point t0 to the ending time point △t, and A(K1,t0,△t) represents a two-dimensional description of the movement of human joint K1 from the starting time point t0 to the ending time point △t. 24 (t0, △t) represents the human body joint point K. 24 A two-dimensional description of the motion from the starting time point t0 to the ending time point Δt.

6. The multi-camera human motion recognition method according to claim 1, characterized in that, The human motion decomposition feature queue is specifically represented as follows: ActArray:{ActNo,CamID,ActOrder, ActBase(t0)} Where ActArray represents the human motion decomposition feature queue, ActNo represents the decomposition motion number, CamID represents the camera number, ActOrder represents the sequence number of the decomposition motion in the continuous motion, and ActBase(t0) represents the two-dimensional description of the human motion decomposition at the starting time point t0.

7. The multi-camera human motion recognition method according to claim 1, characterized in that, The queue of human continuous motion recognition results is specifically represented as follows: ActResult:{ActNo,ActRealOrder,ActTime} Wherein, ActResult represents the queue of human continuous motion recognition results, ActNo represents the decomposed motion number, ActRealOrder represents the sequence number of the decomposed motions in human continuous motion recognition, and ActTime represents the decomposed motion time.

8. The multi-camera human motion recognition method according to claim 1, characterized in that, The comparison of whether the action order of the human action decomposition feature queue and the human continuous action recognition result queue is consistent specifically includes: The human motion decomposition feature queue is sorted according to the order of decomposed motions, and the human continuous motion recognition result queue is sorted according to the order of motion recognition. The sorting results of the human motion decomposition feature queue and the order results of the human continuous motion recognition result queue are compared to see if they are consistent.

Images