Data collection method and system based on 5G gateway

Abstract

The present invention relates to the field of air conditioning engineering technology, and discloses a data collection method and system based on a 5G gateway, comprising: within a first preset time period, collecting image data of a sample office space through a 5G gateway at a fixed time interval t; performing semantic segmentation on the collected image data to obtain training samples, and constructing a portrait workstation feature extraction model based on the training samples; within a second preset time period, extracting portrait workstation features of employees in a target office space through the portrait workstation feature extraction model; determining the employee's posture based on the employee's portrait workstation features, and judging the employee's comfort based on the employee's posture; adjusting the wind speed and temperature of each air outlet in the office space based on the employee's comfort; improving the intelligence level of the air conditioning system and improving the comfort of the office environment.

Description

Technical Field

[0001] The present invention relates to the field of air conditioning engineering technology, and more specifically, to a data acquisition method and system based on a 5G gateway. Background Art

[0002] Existing central air conditioning systems in offices are usually controlled by fixed temperature settings or by manual temperature adjustment. However, due to the difference in temperature perception among each employee, some employees feel too cold or too hot. This traditional control method cannot be adjusted according to individual needs.

[0003] With the rapid development of 5G gateway technology, more and more devices are connected to the Internet of Things. With its high speed, low latency and ability to connect large-scale devices, 5G gateways provide technical support for real-time data collection and transmission. Through the 5G gateway, image data of employees' workstations is collected in real time to determine their current comfort level with the ambient temperature, so as to intelligently control the air speed, direction and temperature of the central air conditioner to ensure that each employee is in a suitable environment. Summary of the invention

[0004] The present invention provides a data collection method and system based on a 5G gateway to solve the technical problems raised in the background technology.

[0005] First, the data collection method based on 5G gateway includes:

[0006] Step 101, within a first preset time period, the office space is divided into a plurality of sub-areas according to the number of workstations in the office space, and image data of each sub-area is continuously collected through a camera connected to a 5G gateway at a fixed time interval t;

[0007] Step 102, semantic segmentation is performed on the collected image data to obtain training samples, and a portrait workstation feature extraction model is trained based on the training samples;

[0008] The collected image data is used as the input of the portrait workstation feature extraction model, and the training samples are used as sample labels of the collected image data; the portrait workstation feature extraction model is used to identify and semantically segment the portrait workstation features in the image data;

[0009] Step 103, within a second preset time period, extracting portrait workstation features of employees in the target office space through a portrait workstation feature extraction model; determining the employee's posture according to the employee's portrait workstation features, and judging the employee's comfort level according to the employee's posture;

[0010] The target space is divided into sub-areas according to the number of workstations in the target space, image data of each sub-area of ​​the target space within a second preset time period is collected, and the heat dissipation behavior of the employees is determined, specifically including:

[0011] Gray-scale the image data collected at fixed time intervals t in the sub-region of the target space within the second preset time period to obtain gray-scale image data;

[0012] pass Determine the change value of the xth row and yth column in the ath sub-region picture data at adjacent time points; wherein, Indicates the change value of the xth row and yth column in the ath sub-region image data at adjacent time points; Indicates the ath sub-region In the grayscale image at the moment Line The pixel value of the column; Indicates the ath sub-region In the grayscale image at the moment Line The pixel value of the column; Indicates the preset motion threshold;

[0013] pass Compare the changed areas in the image data at several adjacent time points to determine whether the changed areas are the same area; wherein, Indicates at time and time Changing areas and times in image data and time The intersection of the changed area in the image data and the pixels above the basic boundary; It represents the operation of taking pixels above the basic boundary in the recognition area; Represents the intersection operation;

[0014] pass Determine the periodicity of motion occurring in the same spatial region; Indicates that the sub-area is collected at each time point in the second preset time period. The number of reciprocating motions that occur in the same area of ​​the frame image; express Frame picture The index of the number of intervals; Represents the indicator function, which is used to The judgment result is converted to 1 or 0; represents a preset threshold area;

[0015] The image data collected in the second preset time period is scaled, cut and stacked to obtain an enhanced image of the sub-region of the target space at each moment in the second preset time period;

[0016] Extracting portrait position features from the enhanced image of the target space through the portrait position feature extraction model;

[0017] Determine the basic boundary of each portrait workstation feature. The basic boundary is the part where the portrait outline and the workstation outline are repeatedly marked.

[0018] Set the upper and lower bounds based on the basic boundary. The upper and lower bounds are parallel boundaries on both sides of the basic boundary. The distances between the upper and lower bounds and the basic boundary are custom parameters.

[0019] The portrait contour is intercepted by the upper and lower bounds, and the fuzzy curvature of the intercepted portion of the portrait contour away from the workstation and the convergence degree of the portrait and the workstation within the second preset time period are calculated;

[0020] The comfort level of the employee is determined based on the heat dissipation behavior of the employee in the second preset time period, the fuzzy curvature of the intercepted portion of the portrait contour away from the workstation, and the degree of convergence between the portrait and the workstation. The comfort level is calculated as follows:

[0021] The fuzzy curvature of the portion of the employee's portrait contour away from one end of the workstation in the second preset time period is classified and determined by a preset classification threshold; if the fuzzy curvature of the portion of the employee's portrait contour away from one end of the workstation in the second preset time period is greater than or equal to the posture threshold, the employee is determined to be in a crouching posture, otherwise the employee is determined to be in a stretching posture;

[0022] If the employee is in a curled-up posture, the degree of closeness between the employee's portrait and the workstation in the second preset time period is classified and determined by using a preset closeness threshold; if the degree of closeness between the employee's portrait and the workstation in the second preset time period is ≥ the closeness threshold, the employee's comfort level is determined to be the first comfort level, otherwise the employee's comfort level is determined to be the second comfort level;

[0023] If the employee is in a stretching posture, the employee's heat dissipation behavior is determined by a preset behavior threshold; if , it is determined that the employee has heat dissipation behavior, and the comfort level of the employee is the third comfort level; otherwise, the comfort level of the employee is the second comfort level;

[0024] Step 104, dynamically adjusting the wind speed and temperature of the air outlet of the office space based on the employee comfort level, as follows:

[0025] Determine the coverage and wind direction trajectory of the air outlet;

[0026] Setting the wind speed and temperature of the air outlet during the second preset time period as the initial wind speed and initial temperature of the air outlet;

[0027] When the air outlet sweeps over the non-workstation area, the initial wind speed and initial temperature are output;

[0028] When the air outlet sweeps over the Rth workstation area, if the comfort level of the employee at the workstation during the second preset time period is the second comfort level, the air outlet outputs the initial wind speed and initial temperature;

[0029] If the air outlet sweeps over the Uth workstation area, and if the comfort level of the employee at the workstation during the second preset time period is the first comfort level or the third comfort level, the initial wind speed and initial temperature are weighted and adjusted using a preset linear adjustment factor;

[0030] Collect image data of the target office space within the next preset time period, determine the comfort level of each employee, and repeatedly weight the initial wind speed and initial temperature according to the employee's comfort level and the preset linear adjustment factor; repeat this process until the comfort level of all employees is the second comfort level.

[0031] Furthermore, the image data of each sub-area is continuously collected at a fixed time interval t through a camera connected to the 5G gateway, including:

[0032] Divide the sample office space into an equal number of sub-areas according to the number of workstations in the sample office space;

[0033] Continuously collect image data of each sub-area at fixed time intervals t through a camera connected to the 5G gateway within a first preset time period;

[0034] The image data collected by the camera within the first preset time period is scaled, cut and stacked to obtain enhanced images of several sub-areas at each moment within the first preset time period.

[0035] Furthermore, semantic segmentation is performed on the collected image data to obtain training samples, and a portrait workstation feature extraction model is trained based on the training samples, including:

[0036] The part of the enhanced image where the employee is located at the workstation is defined as a recognition area, and a first annotation is performed on the outline of the recognition area;

[0037] The part of the enhanced image where the employee is not at the workstation is regarded as an invalid background, and the background is filled with a second annotation;

[0038] Classify and label the human figure contour and workstation contour in the recognition area;

[0039] If part of the contour of the recognition area, part of the portrait contour and part of the workstation contour overlap, then the part of the contour is labeled multiple times;

[0040] Annotate the enhanced image to obtain a semantic matrix;

[0041] Rotate, scale and mirror the semantic matrix and the enhanced image synchronously to obtain an extended matrix and an extended image;

[0042] The extended image and the enhanced image are jointly constructed into a sample set, and the semantic matrix and the extended matrix are jointly constructed into a label set. The deep learning model is trained based on the sample set and the label set to obtain the portrait workstation feature extraction model.

[0043] Further, calculating the fuzzy curvature of the intercepted portion of the employee portrait contour away from one end of the workstation within the second preset time period includes:

[0044] The calculation formula for the fuzzy curvature of the intercepted portion of the portrait contour of the enhanced image at a moment in the second preset time period away from one end of the workstation is as follows:

[0045] ;

[0046] ;

[0047] ;

[0048] ;

[0049] in, Indicates the fuzzy curvature of the intercepted portion of the portrait contour away from one end of the workstation at a moment in the second preset time period; Represents the local curvature of the i-th pixel point of the intercepted part of the portrait contour away from the workstation; and They respectively represent the blurred horizontal and vertical coordinates of the pixel points intercepted by the portrait outline after smoothing; Represents the local curvature weight parameter of the i-th pixel; Indicates the number of pixels of the intercepted part of the portrait outline away from the workstation; It represents the area of ​​the triangle constructed by three adjacent pixels of the part of the portrait outline away from the workstation; , and Respectively represent the Euclidean distance between the corresponding three pixels; and are the original horizontal and vertical coordinates of the i-th pixel point of the intercepted part of the portrait contour away from the workstation; and Represents the two ends of the smoothing window, and smoothes the i-th pixel within the window range; Indicates the first pixels; Indicates the first Pixels and pixels The Euclidean distance of

[0050] According to the fuzzy curvature of the intercepted portion of the portrait contour away from one end of the workstation at each moment in the second preset time period, the discrete data is removed to obtain the effective fuzzy curvature;

[0051] The average value of the effective fuzzy curvature is calculated to obtain the fuzzy curvature of the intercepted portion of the employee portrait contour away from one end of the workstation within the second preset time period.

[0052] Furthermore, the degree of convergence of the portraits within the second preset time period is calculated, and the calculation formula is as follows:

[0053] pass Determine the degree of convergence of the portraits within a second preset time period; wherein, Indicates the degree of convergence of the portrait; Indicates the number of pixels of the intercepted part of the portrait outline; The number of pixels that represent the base boundary.

[0054] In the second aspect, a data collection system based on a 5G gateway is applied to the method described, including:

[0055] Data collection module: used to divide the office space into several sub-areas according to the number of workstations in the office space within a first preset time period, and continuously collect image data of each sub-area through a camera connected to the 5G gateway at a fixed time interval t;

[0056] Data processing module: used to process the image data separately to obtain grayscale image data and enhanced images;

[0057] Portrait station extraction module: used to perform semantic segmentation on the collected image data, obtain training samples, and train the portrait station feature extraction model based on the training samples;

[0058] Posture comfort determination module: used to extract the portrait workstation features of employees in the target office space through the portrait workstation feature extraction model within the second preset time period; determine the employee's posture according to the employee's portrait workstation features, and determine the employee's comfort according to the employee's posture;

[0059] Air conditioning outlet adjustment module: used to dynamically adjust the wind speed and temperature of the office space outlet based on employee comfort.

[0060] The beneficial effects of the present invention are: through data collection based on 5G gateway, combined with semantic segmentation and deep learning models, employees' posture, heat dissipation behavior and other information are monitored in real time, their comfort level is automatically judged, and the wind speed and temperature are dynamically adjusted to gradually adjust to the comfort state of each employee, thereby effectively solving the problem of difficult to achieve personalized needs, improving the intelligence level of the air-conditioning system, and improving the comfort of the office environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of the data collection method based on 5G gateway of the present invention. DETAILED DESCRIPTION

[0062] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0063] like Figure 1 As shown, the data collection method based on the 5G gateway includes:

[0064] Step 101, within a first preset time period, the office space is divided into a plurality of sub-areas according to the number of workstations in the office space, and image data of each sub-area is continuously collected through a camera connected to a 5G gateway at a fixed time interval t;

[0065] Step 102, semantic segmentation is performed on the collected image data to obtain training samples, and a portrait workstation feature extraction model is trained based on the training samples;

[0066] The collected image data is used as the input of the portrait workstation feature extraction model, and the training samples are used as sample labels of the collected image data; the portrait workstation feature extraction model is used to identify and semantically segment the portrait workstation features in the image data;

[0067] Step 103, within a second preset time period, extracting portrait workstation features of employees in the target office space through a portrait workstation feature extraction model; determining the employee's posture according to the employee's portrait workstation features, and judging the employee's comfort level according to the employee's posture;

[0068] The target space is divided into sub-areas according to the number of workstations in the target space, image data of each sub-area of ​​the target space within a second preset time period is collected, and the heat dissipation behavior of the employees is determined, specifically including:

[0069] Gray-scale the image data collected at fixed time intervals t in the sub-region of the target space within the second preset time period to obtain gray-scale image data;

[0070] pass Determine the change value of the xth row and yth column in the ath sub-region picture data at adjacent time points; wherein, Indicates the change value of the xth row and yth column in the ath sub-region image data at adjacent time points; Indicates the ath sub-region In the grayscale image at the moment Line The pixel value of the column; Indicates the ath sub-region In the grayscale image at the moment Line The pixel value of the column; Indicates the preset motion threshold;

[0071] pass Compare the changed areas in the image data at several adjacent time points to determine whether the changed areas are the same area; wherein, Indicates at time and time Changing areas and times in image data and time The intersection of the changed area in the image data and the pixels above the basic boundary; It represents the operation of taking pixels above the basic boundary in the recognition area; Represents the intersection operation;

[0072] pass Determine the periodicity of motion occurring in the same spatial region; Indicates that the sub-area is collected at each time point in the second preset time period. The number of reciprocating motions that occur in the same area of ​​the frame image; express Frame picture The index of the number of intervals; Represents the indicator function, which is used to The judgment result is converted to 1 or 0; represents a preset threshold area;

[0073] The image data collected in the second preset time period is scaled, cut and stacked to obtain an enhanced image of the sub-region of the target space at each moment in the second preset time period;

[0074] Extracting portrait position features from the enhanced image of the target space through the portrait position feature extraction model;

[0075] Determine the basic boundary of each portrait workstation feature. The basic boundary is the part where the portrait outline and the workstation outline are repeatedly marked.

[0076] Set the upper and lower bounds based on the basic boundary. The upper and lower bounds are parallel boundaries on both sides of the basic boundary. The distances between the upper and lower bounds and the basic boundary are custom parameters.

[0077] The portrait contour is intercepted by the upper and lower bounds, and the fuzzy curvature of the intercepted portion of the portrait contour away from the workstation and the convergence degree of the portrait and the workstation within the second preset time period are calculated;

[0078] The comfort level of the employee is determined based on the heat dissipation behavior of the employee in the second preset time period, the fuzzy curvature of the intercepted portion of the portrait contour away from the workstation, and the degree of convergence between the portrait and the workstation. The comfort level is calculated as follows:

[0079] The fuzzy curvature of the portion of the employee's portrait contour away from one end of the workstation in the second preset time period is classified and determined by a preset classification threshold; if the fuzzy curvature of the portion of the employee's portrait contour away from one end of the workstation in the second preset time period is greater than or equal to the posture threshold, the employee is determined to be in a crouching posture, otherwise the employee is determined to be in a stretching posture;

[0080] If the employee is in a curled-up posture, the degree of closeness between the employee's portrait and the workstation in the second preset time period is classified and determined by using a preset closeness threshold; if the degree of closeness between the employee's portrait and the workstation in the second preset time period is ≥ the closeness threshold, the employee's comfort level is determined to be the first comfort level, otherwise the employee's comfort level is determined to be the second comfort level;

[0081] If the employee is in a stretching posture, the employee's heat dissipation behavior is determined by a preset behavior threshold; if , it is determined that the employee has heat dissipation behavior, and the comfort level of the employee is the third comfort level; otherwise, the comfort level of the employee is the second comfort level;

[0082] Step 104, dynamically adjusting the wind speed and temperature of the air outlet of the office space based on the employee comfort level, as follows:

[0083] Determine the coverage and wind direction trajectory of the air outlet;

[0084] Setting the wind speed and temperature of the air outlet during the second preset time period as the initial wind speed and initial temperature of the air outlet;

[0085] When the air outlet sweeps over the non-workstation area, the initial wind speed and initial temperature are output;

[0086] When the air outlet sweeps over the Rth workstation area, if the comfort level of the employee at the workstation during the second preset time period is the second comfort level, the air outlet outputs the initial wind speed and initial temperature;

[0087] If the air outlet sweeps over the Uth workstation area, and if the comfort level of the employee at the workstation during the second preset time period is the first comfort level or the third comfort level, the initial wind speed and initial temperature are weighted and adjusted using a preset linear adjustment factor;

[0088] Collect image data of the target office space within the next preset time period, determine the comfort level of each employee, and repeatedly weight the initial wind speed and initial temperature according to the employee's comfort level and the preset linear adjustment factor; repeat this process until the comfort level of all employees is the second comfort level.

[0089] In one embodiment of the present invention, the image data of each sub-area is continuously collected at a fixed time interval t by a camera connected to a 5G gateway, including:

[0090] Divide the sample office space into an equal number of sub-areas according to the number of workstations in the sample office space;

[0091] Continuously collect image data of each sub-area at fixed time intervals t through a camera connected to the 5G gateway within a first preset time period;

[0092] The image data collected by the camera within the first preset time period is scaled, cut and stacked to obtain enhanced images of several sub-areas at each moment within the first preset time period.

[0093] In one embodiment of the present invention, semantic segmentation is performed on the collected image data to obtain training samples, and a portrait workstation feature extraction model is trained according to the training samples, including:

[0094] The part of the enhanced image where the employee is located at the workstation is defined as a recognition area, and a first annotation is performed on the outline of the recognition area;

[0095] The part of the enhanced image where the employee is not at the workstation is regarded as an invalid background, and the background is filled with a second annotation;

[0096] Classify and label the human figure contour and workstation contour in the recognition area;

[0097] If part of the contour of the recognition area, part of the portrait contour and part of the workstation contour overlap, then the part of the contour is labeled multiple times;

[0098] Annotate the enhanced image to obtain a semantic matrix;

[0099] Rotate, scale and mirror the semantic matrix and the enhanced image synchronously to obtain an extended matrix and an extended image;

[0100] The extended image and the enhanced image are jointly constructed into a sample set, and the semantic matrix and the extended matrix are jointly constructed into a label set. The deep learning model is trained based on the sample set and the label set to obtain the portrait workstation feature extraction model.

[0101] In one embodiment of the present invention, calculating the fuzzy curvature of the portion of the employee portrait contour that is away from one end of the workstation within the second preset time period includes:

[0102] The calculation formula for the fuzzy curvature of the intercepted portion of the portrait contour of the enhanced image at a moment in the second preset time period away from one end of the workstation is as follows:

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] in, Indicates the fuzzy curvature of the intercepted portion of the portrait contour away from one end of the workstation at a moment in the second preset time period; Represents the local curvature of the i-th pixel point of the intercepted part of the portrait contour away from the workstation; and They respectively represent the blurred horizontal and vertical coordinates of the pixel points intercepted by the portrait outline after smoothing; Represents the local curvature weight parameter of the i-th pixel; Indicates the number of pixels of the intercepted part of the portrait outline away from the workstation; It represents the area of ​​the triangle constructed by three adjacent pixels of the part of the portrait outline away from the workstation; , and Respectively represent the Euclidean distance between the corresponding three pixels; and are the original horizontal and vertical coordinates of the i-th pixel point of the intercepted part of the portrait contour away from the workstation; and Represents the two ends of the smoothing window, and smoothes the i-th pixel within the window range; Indicates the first pixels; Indicates the first Pixels and pixels The Euclidean distance of

[0108] According to the fuzzy curvature of the intercepted portion of the portrait contour away from one end of the workstation at each moment in the second preset time period, the discrete data is removed to obtain the effective fuzzy curvature;

[0109] The average value of the effective fuzzy curvature is calculated to obtain the fuzzy curvature of the intercepted portion of the employee portrait contour away from one end of the workstation within the second preset time period.

[0110] In one embodiment of the present invention, the degree of convergence of the portraits within the second preset time period is calculated using the following calculation formula:

[0111] pass Determine the degree of convergence of the portraits within a second preset time period; wherein, Indicates the degree of convergence of the portrait; Indicates the number of pixels of the intercepted part of the portrait outline; The number of pixels that represent the base boundary.

[0112] The data acquisition system based on 5G gateway is applied to the method described, including:

[0113] Data collection module: used to divide the office space into several sub-areas according to the number of workstations in the office space within a first preset time period, and continuously collect image data of each sub-area through a camera connected to the 5G gateway at a fixed time interval t;

[0114] Data processing module: used to process the image data separately to obtain grayscale image data and enhanced images;

[0115] Portrait station extraction module: used to perform semantic segmentation on the collected image data, obtain training samples, and train the portrait station feature extraction model based on the training samples;

[0116] Posture comfort determination module: used to extract the portrait workstation features of employees in the target office space through the portrait workstation feature extraction model within the second preset time period; determine the employee's posture according to the employee's portrait workstation features, and determine the employee's comfort according to the employee's posture;

[0117] Air conditioning outlet adjustment module: used to dynamically adjust the wind speed and temperature of the office space outlet based on employee comfort.

[0118] The camera monitors the sample space at a fixed angle, and the workstations in the sample space are manually framed, so that image data of the workstations in the sample space are collected at fixed time intervals t within a first preset time period. The collected image data are multiple frames of sample space images taken by the camera, and the images of the sample space are segmented after manual frame selection; since the distance between each workstation and the camera is different, the size of the image of each workstation is different; a standard image size is set, and the image of each workstation is scaled with the frame center of the workstation. If the proportions are inconsistent, the edges of the scaled image data are cut off to fit the standard image size, so that the image of each workstation is converted to the standard image size; weights are assigned to multiple frames of scaled and cut images of the same sub-area at the same time, and the weights are superimposed, and the average value is calculated to generate an enhanced image.

[0119] Use manual annotation tools to annotate the relevant parts of the image frame by frame, and store the annotation results in the form of a matrix; generate a two-dimensional semantic matrix based on the annotation results, each element in the matrix corresponds to a pixel in the image and has a specific category label (such as 0 for background, 1 for employee, 2 for workstation, and 3 for identification area); use image data enhancement technology to synchronously process the enhanced image and semantic matrix through rotation (such as 90 degrees, 180 degrees, etc.), scaling (zooming in or out), and mirroring (flipping left and right); obtain the semantic matrix of the enhanced image and the extended image;

[0120] The images in the sample set are used as the input of the deep learning model. After being processed by multiple hidden layers, the output semantic segmentation matrix is ​​obtained, which represents the extracted portrait workstation features. The semantic segmentation matrix output by the deep learning model is compared with the label set, and the error is calculated through the loss function. The gradient of the loss relative to each weight parameter and bias parameter is calculated through the back propagation algorithm. According to the gradient, the weight parameters and bias parameters of the hidden layer are updated. The convergence of the deep learning model is determined by cross-validation. If the deep learning model converges, the portrait workstation feature extraction model is obtained.

[0121] In one embodiment of the present invention, the image data collected from each sub-area of ​​the target space within the second preset time period is grayed. Graying is to convert a color image into a gray image, in which the value of each pixel represents the brightness of the pixel, and the brightness value ranges from 0 to 255. By comparing the gray images of the same portrait workstation at adjacent time points, the brightness change of each pixel in the gray image is determined; when the brightness change of the pixel exceeds the preset motion threshold, the pixel is determined to have changed and marked as 1, otherwise marked as 0. Subsequently, the pixels that have changed in the image data at all time points within the second preset time period are determined to identify the change area. If the change area is in a dispersed state, it indicates that the employee is in a normal office behavior state and no heat dissipation behavior is detected; and if the change area is concentrated in a fixed position, it indicates that there is motion behavior in the same area. Based on the setting of the basic boundary, the change area is further judged; when the change area is above the basic boundary, it indicates that the employee may have a heat dissipation action, such as swinging his arms or fanning; if the change area is below the basic boundary, it indicates that the employee may perform repetitive movements, such as shaking his legs. If a possible heat dissipation action is detected, the periodicity of the action is further analyzed by statistics. If the action is periodic, it is determined that the employee has heat dissipation behavior; if it is not periodic, it is determined to be normal office behavior.

[0122] In one embodiment of the present invention, an enhanced image at a certain moment within the second preset time period is first selected, and the enhanced image is subjected to feature extraction through a portrait workstation feature extraction model to obtain the portrait workstation features of the employee. Subsequently, the extracted portrait workstation features are cut according to the set upper and lower bounds to obtain the portion of the employee portrait contour away from the workstation, i.e., the back contour. Next, each pixel point in the back contour is smoothed. If the smoothing parameter n=2 is set, for the i-th pixel point of the back contour, the adjacent i-2, i-1, i+1 and i+2 pixel points are selected, and the i-th pixel point is smoothed to obtain the smoothed back contour. For the i-1, i and i+1 pixel points in the smoothed back contour, the local curvature is calculated using their geometric relationship. By summarizing each local curvature in the back contour, the fuzzy curvature of the overall back contour is obtained. In the second preset time period, after calculating the fuzzy curvature of the back contour at each moment through the above process, the maximum and minimum values ​​in the fuzzy curvature are removed, and the remaining fuzzy curvatures are averaged to finally obtain the fuzzy curvature of the back contour in the second preset time period.

[0123] In one embodiment of the present invention, the system counts the number of pixels of the intercepted portion of the portrait contour (the number of back contours and chest contours), and then counts the number of pixels of the basic boundary (the number of overlaps between the portrait contour and the workstation contour); based on the ratio of the number of pixels of the intercepted portion of the portrait contour to the number of pixels of the basic boundary, the degree of convergence of the portrait is obtained to reflect the degree of proximity of the employee's body posture to the workstation.

[0124] It should be noted that the first comfort level indicates that employees perceive the ambient temperature as "colder"; the second comfort level indicates that employees perceive the ambient temperature as "moderate"; and the first comfort level indicates that employees perceive the ambient temperature as "hotter".

[0125] It should be noted that according to the fuzzy curvature that reflects the employee's body posture, when the human body feels the environment is "colder", it will reduce the heat dissipation area (curl up and get closer to the workstation) to reduce heat loss; when the human body feels the environment is "hotter", it will maintain a stretched posture and increase air flow (heat dissipation behavior) to dissipate heat; when the human body feels the environment is "moderate", it will maintain a curled or stretched posture according to its own habits.

[0126] In one embodiment of the present invention, during the second preset time period, if the comfort level of most employees is determined to be "moderate", and the comfort level of the employee at the Uth workstation is determined to be "slightly cold", the wind speed and temperature of the workstation are linearly decreased based on the initial wind speed and initial temperature through the linear adjustment factor.

[0127] In the next preset time period, the comfort level of the employee at the Uth workstation is re-evaluated based on the collected image data; if the employee's comfort level is still "cold", the linear adjustment factor will further reduce the wind speed and temperature based on the initial wind speed and initial temperature. This cycle of adjustment will continue until the employee's comfort level reaches "moderate". This process continues until the comfort level of employees at all workstations is adjusted to a "moderate" state.

[0128] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A data collection method based on a 5G gateway, characterized in that: include: Step 101, within a first preset time period, the office space is divided into a plurality of sub-areas according to the number of workstations in the office space, and image data of each sub-area is continuously collected through a camera connected to a 5G gateway at a fixed time interval t; Step 102, semantic segmentation is performed on the collected image data to obtain training samples, and a portrait workstation feature extraction model is trained based on the training samples; The collected image data is used as the input of the portrait workstation feature extraction model, and the training samples are used as sample labels of the collected image data; the portrait workstation feature extraction model is used to identify and semantically segment the portrait workstation features in the image data; Step 103, within a second preset time period, extracting portrait workstation features of employees in the target office space through a portrait workstation feature extraction model; determining the employee's posture according to the employee's portrait workstation features, and judging the employee's comfort level according to the employee's posture; The target space is divided into sub-areas according to the number of workstations in the target space, image data of each sub-area of ​​the target space within a second preset time period is collected, and the heat dissipation behavior of the employees is determined, specifically including: Gray-scale the image data collected at fixed time intervals t in the sub-region of the target space within the second preset time period to obtain gray-scale image data; pass Determine the change value of the xth row and yth column in the ath sub-region picture data at adjacent time points; wherein, Indicates the change value of the xth row and yth column in the ath sub-region image data at adjacent time points; Indicates the ath sub-region In the grayscale image at the moment Line The pixel value of the column; Indicates the ath sub-region In the grayscale image at the moment Line The pixel value of the column; Indicates the preset motion threshold; pass Compare the changed areas in the image data at several adjacent time points to determine whether the changed areas are the same area; wherein, Indicates at time and time Changing areas and times in image data and time The intersection of the changed area in the image data and the pixels above the basic boundary; It represents the operation of taking pixels above the basic boundary in the recognition area; Represents the intersection operation; pass Determine the periodicity of motion occurring in the same spatial region; Indicates that the sub-area is collected at each time point in the second preset time period. The number of reciprocating motions that occur in the same area of ​​the frame image; express Frame picture The index of the number of intervals; Represents the indicator function, which is used to The judgment result is converted to 1 or 0; represents a preset threshold area; The image data collected in the second preset time period is scaled, cut and stacked to obtain an enhanced image of the sub-region of the target space at each moment in the second preset time period; Extracting portrait position features from the enhanced image of the target space through the portrait position feature extraction model; Determine the basic boundary of each portrait workstation feature. The basic boundary is the part where the portrait outline and the workstation outline are repeatedly marked. Set the upper and lower bounds based on the basic boundary. The upper and lower bounds are parallel boundaries on both sides of the basic boundary. The distances between the upper and lower bounds and the basic boundary are custom parameters. The portrait contour is intercepted by the upper and lower bounds, and the fuzzy curvature of the intercepted portion of the portrait contour away from the workstation and the convergence degree of the portrait and the workstation within the second preset time period are calculated; The comfort level of the employee is determined based on the heat dissipation behavior of the employee in the second preset time period, the fuzzy curvature of the intercepted portion of the portrait contour away from the workstation, and the degree of convergence between the portrait and the workstation. The comfort level is determined as follows: The fuzzy curvature of the portion of the employee's portrait contour away from one end of the workstation in the second preset time period is classified and determined by a preset classification threshold; if the fuzzy curvature of the portion of the employee's portrait contour away from one end of the workstation in the second preset time period is greater than or equal to the posture threshold, the employee is determined to be in a crouching posture, otherwise the employee is determined to be in a stretching posture; If the employee is in a curled-up posture, the degree of closeness between the employee's portrait and the workstation in the second preset time period is classified and determined by using a preset closeness threshold; if the degree of closeness between the employee's portrait and the workstation in the second preset time period is ≥ the closeness threshold, the employee's comfort level is determined to be the first comfort level, otherwise the employee's comfort level is determined to be the second comfort level; If the employee is in a stretching posture, the employee's heat dissipation behavior is determined by a preset behavior threshold; if , it is determined that the employee has heat dissipation behavior, and the comfort level of the employee is the third comfort level; otherwise, the comfort level of the employee is the second comfort level; Step 104, dynamically adjusting the wind speed and temperature of the air outlet of the office space based on the employee comfort level, as follows: Determine the coverage and wind direction trajectory of the air outlet; Setting the wind speed and temperature of the air outlet during the second preset time period as the initial wind speed and initial temperature of the air outlet; When the air outlet sweeps over the non-workstation area, the initial wind speed and initial temperature are output; When the air outlet sweeps over the Rth workstation area, if the comfort level of the employee at the workstation during the second preset time period is the second comfort level, the air outlet outputs the initial wind speed and initial temperature; If the air outlet sweeps over the Uth workstation area, and if the comfort level of the employee at the workstation during the second preset time period is the first comfort level or the third comfort level, the initial wind speed and initial temperature are weighted and adjusted using a preset linear adjustment factor; Collect image data of the target office space within the next preset time period, determine the comfort level of each employee, and repeatedly weight the initial wind speed and initial temperature according to the employee's comfort level and the preset linear adjustment factor; repeat this process until the comfort level of all employees is the second comfort level.

2. The data collection method based on 5G gateway according to claim 1 is characterized in that: At a fixed time interval t, the camera connected to the 5G gateway continuously collects image data of each sub-area, including: Divide the sample office space into an equal number of sub-areas according to the number of workstations in the sample office space; Continuously collect image data of each sub-area at fixed time intervals t through a camera connected to the 5G gateway within a first preset time period; The image data collected by the camera within the first preset time period is scaled, cut and stacked to obtain enhanced images of several sub-areas at each moment within the first preset time period.

3. The data collection method based on 5G gateway according to claim 2 is characterized in that: Perform semantic segmentation on the collected image data to obtain training samples, and train the portrait workstation feature extraction model based on the training samples, including: The part of the enhanced image where the employee is located at the workstation is defined as a recognition area, and a first annotation is performed on the outline of the recognition area; The part of the enhanced image where the employee is not at the workstation is regarded as an invalid background, and the background is filled with a second annotation; Classify and label the human figure contour and workstation contour in the recognition area; If part of the contour of the recognition area, part of the portrait contour and part of the workstation contour overlap, then the part of the contour is labeled multiple times; Annotate the enhanced image to obtain a semantic matrix; Rotate, scale and mirror the semantic matrix and the enhanced image synchronously to obtain an extended matrix and an extended image; The extended image and the enhanced image are jointly constructed into a sample set, and the semantic matrix and the extended matrix are jointly constructed into a label set. The deep learning model is trained based on the sample set and the label set to obtain the portrait workstation feature extraction model.

4. The data collection method based on 5G gateway according to claim 3 is characterized in that: Calculating the fuzzy curvature of the intercepted portion of the employee portrait contour away from one end of the workstation within the second preset time period includes: The calculation formula for the fuzzy curvature of the intercepted portion of the portrait contour of the enhanced image at a moment in the second preset time period away from one end of the workstation is as follows: ； ； ； ； in, Indicates the fuzzy curvature of the intercepted portion of the portrait contour away from one end of the workstation at a moment in the second preset time period; Represents the local curvature of the i-th pixel point of the intercepted part of the portrait contour away from the workstation end; and They respectively represent the blurred horizontal and vertical coordinates of the pixel points intercepted by the portrait outline after smoothing; Represents the local curvature weight parameter of the i-th pixel; Indicates the number of pixels of the intercepted part of the portrait outline away from the workstation; It represents the area of ​​the triangle constructed by three adjacent pixels of the part of the portrait outline away from the workstation; , and Respectively represent the Euclidean distance between the corresponding three pixels; and are the original horizontal and vertical coordinates of the i-th pixel point of the intercepted part of the portrait contour away from the workstation; and Represents the two ends of the smoothing window, and smoothes the i-th pixel within the window range; Indicates the first pixels; Indicates the first Pixels and pixels The Euclidean distance of According to the fuzzy curvature of the intercepted portion of the portrait contour away from one end of the workstation at each moment in the second preset time period, the discrete data is removed to obtain the effective fuzzy curvature; The average value of the effective fuzzy curvature is calculated to obtain the fuzzy curvature of the intercepted portion of the employee portrait contour away from one end of the workstation within the second preset time period.

5. The 5G gateway-based data collection method according to claim 4, characterized in that: The degree of convergence of the portraits in the second preset time period is calculated using the following formula: pass Determine the degree of convergence of the portraits within a second preset time period; wherein, Indicates the degree of convergence of the portrait; Indicates the number of pixels of the intercepted part of the portrait outline; The number of pixels that represent the base boundary.

6. A 5G gateway-based data acquisition system, applied to the method according to any one of claims 1 to 5, characterized in that: include: Data collection module: used to divide the office space into several sub-areas according to the number of workstations in the office space within a first preset time period, and continuously collect image data of each sub-area through a camera connected to the 5G gateway at a fixed time interval t; Data processing module: used to process the image data separately to obtain grayscale image data and enhanced images; Portrait station extraction module: used to perform semantic segmentation on the collected image data, obtain training samples, and train the portrait station feature extraction model based on the training samples; Posture comfort determination module: used to extract the portrait workstation features of employees in the target office space through the portrait workstation feature extraction model within the second preset time period; determine the employee's posture according to the employee's portrait workstation features, and determine the employee's comfort according to the employee's posture; Air conditioning outlet adjustment module: used to dynamically adjust the wind speed and temperature of the office space outlet based on employee comfort.

Images