Intelligent air conditioner energy-saving control system and method based on visual recognition

Abstract

The invention relates to the technical field of air conditioner energy-saving control, in particular to an intelligent air conditioner energy-saving control system and method based on visual recognition, and the method comprises the steps of personnel image collection to obtain a thermal image and form an overlapping layer, dynamic action recognition to generate a behavior distribution pattern, and target area extraction to determine an air supply control number. The wind direction and wind speed are linked to form an adjusting item, and the energy-saving control execution unit outputs an air conditioner control state instruction. The method comprises the following steps: generating a personnel trajectory diagram through thermal inductance frequency tracking and spatial superposition projection; screening a static area by combining coordinate offset and attitude angles; establishing a control number according to a diagram block number and an air guide port image; forming an adjustment basis according to an angle difference and a contact boundary; the control state content is output through the image time frame, the air supply number is associated with the spatial position, the adaptability of an air supply path and personnel distribution is enhanced, and the air conditioner response capacity and resource use matching efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of air conditioning energy-saving control technology, and in particular to an intelligent air conditioning energy-saving control system and method based on visual recognition. Background Technology

[0002] The field of air conditioning energy-saving control technology encompasses the optimization of air conditioning system energy efficiency using various control methods and equipment. Its core focus is on energy-saving and environmentally friendly technologies for air conditioning equipment, improving energy utilization efficiency and reducing energy consumption through automated and intelligent control methods. Common technical measures within this field include temperature control, humidity regulation, airflow distribution, and optimization of operating modes. With the increasing demands for environmental protection and energy conservation and emission reduction, air conditioning energy-saving control technology is gradually developing towards intelligence and automation, combining advanced sensors, control systems, and data analysis technologies to achieve more precise energy-saving management.

[0003] Among them, the visual recognition-based intelligent air conditioning energy-saving control system refers to the use of image recognition technology to perceive and analyze information such as the human body, temperature, and humidity in the air-conditioned environment, thereby achieving intelligent adjustment of the operating status of air conditioning equipment. Research has been conducted on how to optimize the energy efficiency control of air conditioning through a visual recognition system. This is mainly achieved by real-time monitoring of environmental changes and user activity status, adjusting the air conditioning's operating mode and temperature control strategy to achieve energy savings. By introducing image recognition technology, the system automatically identifies the distribution of people, their activity status, and environmental parameters indoors, and dynamically adjusts these parameters in conjunction with the air conditioning control system, effectively solving the problem that traditional air conditioning systems cannot adapt to real-time changing environmental demands and energy-saving requirements.

[0004] Existing air conditioning control methods rely on the overall temperature and humidity of the environment as the trigger, ignoring the differences in the thermal distribution of people in the space. This makes it difficult to respond in a timely manner in areas where people stay or are densely distributed. The air supply path is limited by fixed preset and average parameters, lacking the ability to correlate with the state of specific areas. This leads to a disconnect between the air supply results and the space usage. Under conditions of multiple areas coexisting and dynamic changes in people, some areas may experience uneven heating and cooling, increased air conditioning load, and other phenomena. The control strategy lacks spatial zoning dimension and is difficult to support differentiated control needs. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a visual recognition-based intelligent air conditioning energy-saving control system and method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent air conditioning energy-saving control system based on visual recognition, the system comprising:

[0007] The personnel image acquisition module acquires images from an infrared vision camera, extracts the outline of the human head and shoulders, removes non-thermal areas, tracks high-response-frequency positions, performs spatial overlay of the outline center, determines the effective layer based on the overlap stability, and obtains the personnel image overlay layer.

[0008] The dynamic action recognition module, based on the overlapping layer of the personnel images, extracts the center coordinates of the same area in continuous images, calculates the inter-frame offset and boundary contact frequency, classifies and sets static and moving states, generates a label binding layer, and obtains a behavior distribution pattern in space.

[0009] The target area extraction module selects static concentrated blocks based on the behavior distribution pattern in the space, analyzes the angular relationship between the head and neck line and the waist line in the image, screens out blocks that present a horizontal posture, cross-matches the static label blocks and the sitting posture block numbers, establishes an air supply area number mapping, and obtains the target air supply control area number.

[0010] The wind direction and speed linkage module reads the air outlet image corresponding to the target air supply control area number, marks the angular deviation between the air guide and the center of the area, calculates the contact length between the airflow trajectory and the image block, and generates a linkage record by combining the offset direction and wind speed distribution to obtain the wind direction and speed linkage adjustment item.

[0011] As a further aspect of the present invention, the personnel image overlay layer includes a thermal response aggregation area, continuous trajectory overlay points, and layer screening index number; the spatial behavior distribution pattern includes a distribution of movement status identifiers, a static status aggregation block, and a status label mapping relationship; the target air supply control area number includes a static area index number, a sitting posture determination block number, and a corresponding air vent number; and the wind direction and wind speed linkage adjustment items include air duct angle deviation, airflow coverage contact length, and wind speed diffusion discrimination factor.

[0012] As a further aspect of the present invention, the personnel image acquisition module includes:

[0013] The infrared image receiving submodule acquires image data collected by infrared vision cameras deployed in the indoor area, scans and processes each frame of image, identifies and removes static blocks in the image whose infrared intensity remains unchanged for a long time, retains effective areas with infrared response fluctuations, and organizes continuous frame data in chronological order to form an infrared response image sequence set that can be used for subsequent processing.

[0014] The human body boundary extraction submodule extracts the head and shoulder contour regions with human thermal radiation characteristics from the infrared response image sequence set, calls the edge gray-scale gradient change parameters to establish a contour distribution model, identifies continuous edge segments according to the preset gradient threshold and completes the region connection operation, filters out the boundary contour regions with human morphological characteristics, and obtains the head and shoulder structure boundary distribution matrix.

[0015] The thermal overlap mapping submodule tracks the spatial coordinate changes of each contour center point in the image sequence based on the head and shoulder structure boundary distribution matrix, performs superposition calculation on the repeated occurrence positions of the center points in three-dimensional space, statistically analyzes the frequency of thermal response occurrence in space, sets a thermal response stability threshold, removes low-frequency response segments, retains the contour thermal area superposition results, and generates a personnel image superposition layer.

[0016] As a further aspect of the present invention, the dynamic action recognition module includes:

[0017] The center trajectory extraction submodule acquires the contour data of each thermal response region in the overlapping layer of the personnel image, extracts the center point position of the same region in the image sequence, calls the contour centroid coordinates in each frame image, establishes a continuous temporal mapping relationship based on the frame sequence index, calculates the coordinate difference between adjacent frames and performs spatial registration, and generates the target area offset trajectory sequence.

[0018] The state classification and judgment submodule, based on the target region offset trajectory sequence, counts the number of intersections of the centroid of each region's contour within the image boundary range, calls the spatial dispersion index of the intersection frequency and coordinate points in each region, and judges the state of each region according to the set image boundary intersection threshold and position fluctuation tolerance range to obtain the behavior state label matrix.

[0019] The behavior pattern generation submodule extracts the projection coordinates of each marked area in space based on the behavior state label matrix, calls the contour positioning information in the personnel image overlay layer, establishes a one-to-one mapping relationship between area coordinates and label states, draws layers according to the mapping structure and completes label binding, and obtains the behavior distribution pattern in space.

[0020] As a further aspect of the present invention, the target region extraction module includes:

[0021] The static block positioning submodule acquires the behavior distribution pattern within the space, locates the block areas marked as static, extracts the corresponding numbers and records their position coordinates in the planar layer, filters the set of block numbers with static characteristics through status labels, and establishes a static area number list.

[0022] The sitting posture recognition submodule extracts the image of the person in the corresponding block based on the static area number list, detects the spatial structural relationship between the head and neck line and the waist line of the human body in the image, calculates the angle value formed by the two, and filters it with reference to the horizontal posture judgment threshold to obtain the sitting posture block number set.

[0023] The control number binding submodule calls the static area number list and the sitting posture state block number set, extracts the intersection of the two numbers, associates the spatial mapping number of each block in the air conditioner air outlet image, completes coordinate binding based on the consistency of the numbers, and obtains the target air supply control area number.

[0024] As a further aspect of the present invention, the wind direction and wind speed linkage module includes:

[0025] The angle difference extraction submodule reads the air conditioning air outlet image corresponding to the target air supply control area number, extracts the coordinate position of the center point of the air duct and the center point of the control area block, calculates the angle value of the line connecting the two points according to the layer reference axis, records the angle results in the order of the numbers, and establishes a list of air duct direction deviations.

[0026] The airflow coverage judgment submodule, based on the air duct direction deviation list, overlays the current airflow layer coverage map, extracts the contact area between the corresponding airflow boundary of each air outlet and the target tile boundary, calculates the ratio of the intersection side length to the perimeter length of the tile, and determines whether the airflow effectively covers the target area according to the preset contact threshold, thereby obtaining the airflow contact state matrix of the target area.

[0027] The linkage parameter calculation submodule calls the airflow contact state matrix of the target area, identifies the area number with insufficient coverage, and calculates the wind direction correction value and wind speed compensation amount respectively by combining the wind direction deviation value corresponding to each number and the diffusion direction range in the current wind speed layer, and integrates the correction data to generate wind direction and wind speed linkage adjustment items.

[0028] As a further aspect of the present invention, the process of determining whether the airflow effectively covers the target area according to the preset contact threshold is as follows: when the ratio of the intersecting side length to the perimeter length of the map is greater than the preset contact threshold, the corresponding area is marked as effectively covered in the airflow contact state matrix of the target area; when the ratio of the intersecting side length to the perimeter length of the map is not greater than the preset contact threshold, it is marked as insufficiently covered.

[0029] The process of calculating the windward angle correction value and the wind speed compensation amount separately is as follows: the windward angle correction value is set to an adjustment amount that is equal in magnitude and opposite in direction to the windward direction deviation value. When the absolute value of the windward direction deviation value exceeds the preset angle deviation threshold, the wind speed compensation amount is also set to increase the current wind speed level based on the calculation of the windward angle correction value.

[0030] As a further aspect of the present invention, the system further includes:

[0031] The energy-saving control execution module, based on the wind direction and wind speed linkage adjustment item, performs angle correction and wind speed selection, sets it to directional, variable speed or low frequency air supply, writes the control number and image frame into the record table and marks the control status, and obtains the air conditioning control output status command.

[0032] The air conditioning control output status commands include air supply angle control parameters, wind speed adjustment mode number, and control status time synchronization record.

[0033] As a further aspect of the present invention, the energy-saving control execution module includes:

[0034] The air guide angle adjustment submodule extracts the air guide angle setting value corresponding to each air supply number based on the wind direction and wind speed linkage adjustment item. The control device adjusts the angle of the air guide mechanism according to the setting to complete the actual angle deflection operation. It records the air outlet number, the angle before and after correction and the response time information to form an air guide angle adjustment record table.

[0035] The wind speed mode setting submodule calls the air supply number in the wind guide angle adjustment record table, associates the wind speed trajectory distribution status corresponding to the number, and sets the air supply strategy to be adopted for the current number according to the judgment rules based on the wind direction deflection amplitude and trajectory diffusion range parameters, and obtains the air supply mode status table.

[0036] The control instruction writing submodule extracts the control number and its corresponding status identifier according to the air supply mode status table, combines the timestamp information in the image frame sequence, writes it into the synchronization record data table, and appends the number index and operation identifier to the instruction field to obtain the air conditioning control output status instruction.

[0037] A vision-recognition-based intelligent air conditioner energy-saving control method, wherein the vision-recognition-based intelligent air conditioner energy-saving control method is executed based on the aforementioned vision-recognition-based intelligent air conditioner energy-saving control system, includes the following steps:

[0038] S1: Acquire images from infrared vision cameras installed in indoor spaces, extract the boundaries of human head and shoulder contours in the images, exclude static areas with no infrared response, track the contours of areas with high thermal response frequency, calculate the center point of thermal areas and continuously project and overlay them in space, and filter effective layers by analyzing the stability of thermal area positions to obtain personnel image overlay layers.

[0039] S2: Based on the overlapping layer of the human images, extract the center coordinates of the same human body region in the continuous images, compare the intersection frequency of the inter-frame offset trajectory and the image boundary, mark the active offset region as the moving state, and mark the region with small fluctuation in the center position as the stationary state. Draw a behavior distribution layer in space and bind state labels to obtain a behavior distribution pattern in space.

[0040] S3: Based on the behavior distribution pattern in the space, locate the static state label set of blocks, detect the angle between the head and neck line and the waist line of the human body image in the block, filter out the area blocks with the angle tending to be horizontal, cross-filter the static block number and the sitting block number, associate the mapping relationship of the air conditioning air outlet area image, and obtain the target air supply control area number.

[0041] S4: Read the air conditioner air outlet image corresponding to the target air supply control area number, mark the angle difference between the center of the air duct image and the center point of the control area, calculate the contact side length between the current airflow image coverage area and the target block, determine whether the airflow covers the area, and combine the air duct deflection and wind speed diffusion state to obtain the wind direction and wind speed linkage adjustment item.

[0042] S5: Based on the wind direction and wind speed linkage adjustment item, execute the control command to adjust the air outlet angle to the specified direction, and set the low frequency, variable speed or directional air supply mode in combination with the wind speed trajectory status. Write the control status number and image time frame synchronously into the record table and add the corresponding identifier to obtain the air conditioning control output status command.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0044] In this invention, a thermal response trajectory map of personnel is constructed by thermal frequency tracking and continuous spatial overlay projection. The state attributes within the area are determined based on the coordinate offset range and trajectory shape. The static posture area is screened by combining the angle between the head and neck line and the waist line. The target number is established by the correspondence between the map block number and the air vent image. The wind direction and wind speed adjustment basis is formed based on the angle difference between the air vent center and the control area and the contact boundary length. The air supply mode is designed by combining the wind speed diffusion state. The control state output content is associated with the image time frame, and the air supply number is associated with the spatial position, making the air supply path distribution more spatially adaptable and improving the air conditioning response capability and resource utilization matching degree. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the overall system flow of the present invention.

[0046] Figure 2 This is a flowchart of the system modules of the present invention;

[0047] Figure 3 This is a flowchart of the method of the present invention. Detailed Implementation

[0048] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0049] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0050] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0051] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0052] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0053] Please see Figure 1 This invention provides a technical solution: an intelligent air conditioning energy-saving control system based on visual recognition, the system comprising:

[0054] The personnel image acquisition module acquires images from infrared vision cameras deployed in the indoor space, extracts the outline boundaries of human head and shoulders in the image, excludes static blocks with no infrared response, tracks the outline of areas with high thermal response frequency, continuously projects and superimposes the center point of the outline in space, and filters the effective layers based on the stability of the thermal zone position to obtain the personnel image overlay layer.

[0055] The dynamic action recognition module extracts the center coordinates of the same human body region in continuous images based on the overlapping layer of human images. It compares the frequency of inter-frame offset trajectory and image boundary intersection, marks the active offset region as the moving state, and marks the region with smaller fluctuations in the center position as the stationary state. It draws a spatial distribution layer and binds state labels to obtain a spatial behavior distribution pattern.

[0056] The target area extraction module, based on the spatial behavior distribution pattern, locates the static state label cluster blocks, detects the angle formed by the head and neck line and waist line of the human image in the block, selects the area blocks with the angle tending to be horizontal, filters the static block number and the sitting block number, and completes the number binding by associating the image mapping relationship of the air conditioning air outlet area to obtain the target air supply control area number.

[0057] The wind direction and speed linkage module reads the air conditioning outlet image corresponding to the target air supply control area number, marks the angle difference between the center of the air guide image and the center point of the control area, calculates the contact side length between the current airflow image coverage area and the target block, determines whether the airflow covers the area, and combines the air guide deflection and wind speed diffusion state to make a combined judgment to obtain the wind direction and speed linkage adjustment item.

[0058] The energy-saving control execution module, based on the wind direction and wind speed linkage adjustment item, executes control commands to adjust the air outlet angle to the marked direction, sets low-frequency, variable speed or directional air supply mode in combination with wind speed trajectory status, writes the control status number and image time frame synchronously into the record table, and adds corresponding identifiers to obtain the air conditioning control output status command.

[0059] The personnel image overlay layer includes thermal response clustering areas, continuous trajectory overlay points, and layer screening index numbers. The spatial behavior distribution pattern includes the distribution of movement status indicators, stationary status clustering blocks, and status label mapping relationships. The target air supply control area number includes the stationary area index number, sitting posture judgment block number, and corresponding air vent number. The wind direction and wind speed linkage adjustment items include the air guide angle deviation, airflow coverage contact length, and wind speed diffusion discrimination factor. The air conditioning control output status commands include air supply angle control parameters, wind speed adjustment mode number, and control status time synchronization recording.

[0060] Please see Figure 2 The personnel image acquisition module includes:

[0061] The infrared image receiving submodule acquires image data collected by infrared vision cameras deployed in the indoor area, scans and processes each frame of image, identifies and removes static blocks in the image whose infrared intensity remains unchanged for a long time, retains effective areas with infrared response fluctuations, and organizes continuous frame data in chronological order to form an infrared response image sequence set that can be used for subsequent processing.

[0062] Image data with a resolution of 640×480 pixels and a frame rate of 10fps, collected by infrared vision cameras deployed in the indoor area, is acquired. Specifically, each frame is divided into 1200 16×16 pixel blocks, and the average infrared intensity value of each block is calculated. To identify and remove static blocks in the image where the infrared intensity remains unchanged for a long time, the "long time" is first set to 100 consecutive frames of image data, i.e., 10 seconds. Secondly, the fluctuation threshold for "unchanged" infrared intensity is set to 1.5 intensity units. This threshold is based on the following: in an unmanned environment, 1000 frames of image data are continuously collected from a fixed heat source indoors, such as a running server, and the maximum fluctuation value of the average infrared intensity of each block is statistically analyzed. If, after statistical analysis, 99.5% of the static blocks have an intensity fluctuation value below 1.2, then 1.5 is selected as the threshold to filter out the influence of sensor noise. For example, in... In 100 consecutive frames, the average infrared intensity value sequence of patch T125 is [85.2, 85.3, 85.1, 85.2, 85.4, 85.3, 85.1, 85.5, 85.2, 85.3]. The difference between its maximum and minimum values ​​is 0.4, which is less than 1.5. Therefore, this patch is determined to be a static patch and removed from the current processing sequence. Conversely, patch T256, due to personnel movement, has an intensity value sequence of [125.7, 128.9, 135.2, 138.5, 139.1, 137.8, 136.5, 139.8, 140.1, 139.5]. Its fluctuation value is 14.4, which is greater than 1.5. Therefore, it is determined to be a valid area and retained. The valid area patch data retained in all frames are concatenated according to their respective frame numbers to form an infrared response image sequence set that can be used for subsequent processing.

[0063] The human body boundary extraction submodule extracts the head and shoulder contour regions with human thermal radiation characteristics from the image based on the infrared response image sequence set. It calls the edge gray-scale gradient change parameters to establish a contour distribution model, identifies continuous edge segments according to the preset gradient threshold and completes the region connection operation, and filters out the boundary contour regions with human morphological characteristics to obtain the head and shoulder structure boundary distribution matrix.

[0064] By filtering the infrared intensity values ​​of pixels within the effective area of ​​the acquired infrared response image sequence, pixels with intensity values ​​between 140 and 180 are extracted. This range corresponds to the thermal radiation characteristics of a standard human body in an indoor environment, thus initially identifying the head and shoulder contour region with human thermal radiation characteristics. Subsequently, edge grayscale gradient change parameters are applied to each pixel within the identified region. The horizontal gradient is calculated by using the difference in infrared intensity between the pixel and its eight neighboring pixels. gradient in the vertical direction For example, pixels The infrared intensity value is 155, and the pixel on its right is... Intensity is 160, left pixel If the strength is 150, then The pixels below it Intensity is 158, top pixel If the strength is 152, then and based on Obtain the gradient magnitude at that point, substitute it into the numerical value, Then, edge recognition is performed according to a preset gradient threshold of 25.0. The threshold is set with reference to the following experiment: 10 sets of sample images containing clear head and shoulder contours are collected. If the average gradient magnitude of the contour edge pixels is calculated to be 35.4 and the average gradient magnitude of the background area is 8.2, then the average of the two is taken. After adding a margin, the value was set to 25.0. Therefore, pixels with a gradient magnitude greater than 25.0 were marked as edge points. For example, pixel [missing information]. If the gradient magnitude is 30.2, which is greater than 25.0, it is marked as an edge point. The 8-neighborhood connectivity algorithm is performed on all spatially continuous edge points to complete the region connection operation, forming several closed or semi-closed continuous edge segments. The aspect ratio of the minimum bounding rectangle formed by each edge segment is calculated, and non-typical human body contours with aspect ratios less than 0.6 or greater than 1.5 are filtered out, as well as fragmented edge segments with a total length of less than 60 pixels. For example, if the bounding rectangle formed by an edge segment is 50 pixels long and 20 pixels wide, with an aspect ratio of 2.5, which is greater than 1.5, it is filtered out. If another edge segment has a total length of 35 pixels, which is less than 60 pixels, it is also filtered out. The head and shoulder structure boundary distribution matrix is ​​obtained.

[0065] The thermal overlap mapping submodule tracks the spatial coordinate changes of each contour center point in the image sequence based on the head and shoulder structure boundary distribution matrix. It performs superposition calculations on the repeated occurrence positions of the center points in three-dimensional space, counts the frequency of thermal response occurrences in space, sets a thermal response stability threshold, removes low-frequency response segments, retains the contour thermal area superposition results, and generates a personnel image superposition layer.

[0066] Based on the generated head and shoulder structure boundary distribution matrix, for each frame in the image sequence, the geometric center coordinates of each independent head and shoulder contour region are calculated. For example, in the t-th frame image, the identified contour regions... The set of boundary pixels included is Then the coordinates of its center point for Assuming the calculation yields The image coordinates are mapped to a preset 50×40 two-dimensional grid coordinate system for indoor space. This grid corresponds to a 5m×4m physical space indoors. Each grid cell represents 0.1m×0.1m. The coordinates (325,118) are mapped to the grid cell (32,12), and the count value of the grid cell is incremented by 1. After processing 1200 frames (i.e. 2 minutes) of image sequence continuously, the cumulative count value of each cell in the grid is counted, i.e. the frequency of thermal response, as shown in Table 1.

[0067] Table 1: Statistics of Indoor Grid Thermal Response Frequency

[0068]

[0069] As shown in Table 1, this table records the frequency of thermal response occurrences of different indoor grid cells within the observation period. The thermal response stability threshold is set to 60. This threshold is based on the criterion that a person staying in a certain area for more than 6 seconds (60 frames) is considered a valid stay. This criterion is derived from the observation and statistics of the daily behavior of 5 test subjects in the office. Grid cells with a cumulative count value of less than 60, such as cell (15,28) with a count value of 45, are considered to be transient or interference signals, and their count values ​​are reset to 0. Cell (32,12) with a count value of 857 is retained. Cell (22,10) with a count value of 68, which is greater than 60, is retained. Cell (18,25) with a count value of 58, which is not greater than 60, is reset to 0. Finally, all grid cells with non-zero count values ​​and their count values ​​are retained to generate a personnel image overlay layer.

[0070] The dynamic action recognition module includes:

[0071] The center trajectory extraction submodule obtains the contour data of each thermal response area in the overlapping layer of personnel images, extracts the center point position of the same area in the image sequence, calls the contour centroid coordinates in each frame image, establishes a continuous temporal mapping relationship based on the frame sequence index, calculates the coordinate difference between adjacent frames and performs spatial registration, and generates the target area offset trajectory sequence.

[0072] The contour data of each thermal response region in the personnel image overlay layer is obtained. Specifically, for each grid cell with a count value greater than 0, its original contour pixel data within the corresponding time frame of the original infrared response image sequence set is extracted. Then, the center point position of the same region in the image sequence is continuously extracted. For example, a thermal response region A with grid cell (32,12) as its core is identified in the personnel image overlay layer. This region persists in the image sequence from frame 101 to frame 250. Then, the contour centroid coordinates of this region in these 150 frames are retrieved sequentially. For example, the centroid coordinates of frame 101 are (325,118), frame 102 are (326,119), ..., frame 250 are (330,145). Based on the frame indexes 101 to 250, a continuous temporal mapping relationship of these coordinate points is established. Then, the coordinate difference between adjacent frames is calculated. For example, the coordinate difference between frame 101 and frame 102 is (...). , The coordinate difference between frame 102 and frame 103 is (). , The image displacement is spatially registered based on the camera calibration parameters. For example, if the camera focal length is 50mm and the pixel size is 0.005mm / pixel, then the image displacement of 1 pixel corresponds to a physical space movement of 0.005mm. Thus, the image displacement is converted into the movement distance in physical space, generating a target area offset trajectory sequence that reflects the target's movement in real space.

[0073] The state classification and judgment submodule, based on the target region offset trajectory sequence, counts the number of intersections of the centroid of each region's contour within the image boundary range, calls the spatial dispersion index of the intersection frequency and coordinate points in each region, and judges the state of each region according to the set image boundary intersection threshold and position fluctuation tolerance range to obtain the behavior state label matrix.

[0074] Based on the generated target region offset trajectory sequence, four boundary lines are defined in the image coordinate system, namely... , , , The total number of times the trajectory of the centroid of each region crosses these four boundary lines is counted; this is the number of intersections. For example, if the trajectory sequence of region A enters from the left side of the image and leaves from the right side within the observation period, its number of intersections is 2. Simultaneously, the spatial dispersion index of all coordinate points within each region is used. This index is specifically calculated as the average Euclidean distance from all coordinate points in the region's trajectory to its average position (centroid). This value reflects the degree of positional fluctuation. For example, if the set of trajectory points for region A is... The average position is Then the spatial discreteness is Assuming the calculated spatial dispersion of region A is 0.3 meters and that of region B is 0.7 meters, and setting the image boundary crossing threshold to 1 and the positional fluctuation tolerance range to 0.5 meters, the crossing threshold of 1 is based on the logic that a single complete entry and exit from a room will inevitably result in at least one boundary crossing. The 0.5-meter positional fluctuation tolerance range was determined experimentally; in a seated office position, the unconscious swaying range of the human head and shoulders typically does not exceed 0.4 meters, hence 0.5 meters is chosen as the boundary distinguishing between stillness and movement. Based on this limitation, the state of each region is determined. If the number of intersections of a region is greater than or equal to 1, or its spatial dispersion is greater than 0.5 meters, it is determined to be in a "moving" state. For example, the number of intersections of region A is 2, which is greater than or equal to 1, so it is determined to be in a "moving" state. The number of intersections of region B is 0, but its spatial dispersion is 0.7 meters, which is greater than 0.5 meters, so it is also determined to be in a "moving" state. The number of intersections of region C is 0, and its spatial dispersion is 0.2 meters, which is not greater than 0.5 meters, so it is determined to be in a "stationary" state. The behavior state label matrix is ​​then obtained.

[0075] The behavior pattern generation submodule extracts the projection coordinates of each marked area in space based on the behavior status label matrix, calls the contour positioning information in the personnel image overlay layer, establishes a one-to-one mapping relationship between area coordinates and label status, draws layers according to the mapping structure and completes label binding, and obtains the behavior distribution pattern in space.

[0076] Extract the state labels and their corresponding identifiers for each region in the behavior state label matrix. For example, the state label for region A is "moving," the state label for region B is "moving," and the state label for region C is "stationary." Then, call the personnel image overlay layer to obtain the projection coordinates of these regions in the spatial grid. Specifically, region A corresponds to a set of cells with grid cell (32,12) as the core, region B corresponds to another set of cells with (15,28) as the core, and region C corresponds to a set of cells with (40,35) as the core. Next, establish a one-to-one mapping relationship between region coordinates and label states, i.e., [(( 31, 11), (32, 11), (31, 12), (32, 12)), “Moving”], [((14, 27), (15, 27), (14, 28), (15, 28)), “Moving”], [((39, 34), (40, 34), (39, 35), (40, 35)), “Still”], Based on this mapping structure, on the two-dimensional layer representing the interior space, the grid cells belonging to regions A and B are drawn in red and bound with the “Moving” label, and the grid cells belonging to region C are drawn in blue and bound with the “Still” label on the layer, thus obtaining the behavior distribution pattern within the space.

[0077] The target region extraction module includes:

[0078] The static block positioning submodule acquires the behavior distribution pattern in the space, locates the block areas marked as static, extracts the corresponding numbers and records their position coordinates in the planar layer, filters the set of block numbers with static characteristics through status labels, and establishes a list of static area numbers.

[0079] Traverse all marked areas in the behavior distribution pattern within the space, locate the tile areas marked as "stationary". For example, if a region C with grid cell (40,35) as its core is found in the pattern and its state label is "stationary", then extract the unique number C of this region and record the set of all grid cell coordinates in the 50×40 planar layer, such as {(39,34),(40,34),(39,35),(40,35)}. At the same time, if the state labels of regions A and B are found to be "moving", then ignore these regions. By performing this state label screening on all regions, a set containing only tile numbers with the static feature is finally obtained. For example, if there are three regions A, B, and C in the scene, and their states are "moving", "moving", and "stationary" respectively, then the final set is {C}. Based on this, a list of static region numbers with the content [C] is established.

[0080] The sitting posture recognition submodule extracts the images of people in the corresponding blocks based on the static area number list, detects the spatial structural relationship between the head and neck line and the waist line of the human body in the image, calculates the angle value formed by the two, and filters according to the horizontal posture judgment threshold to obtain the sitting posture block number set.

[0081] Based on the established list of static region numbers, i.e., [C], the original infrared image data of the region within the corresponding time frame is extracted sequentially. For example, for region C, image slices are extracted from the period marked as static from frame 301 to frame 450. Then, the spatial structural relationship between the human head-neck line and waist line is detected in the image slices. The specific detection process is as follows: human skeletal key point recognition technology is used to locate the key points of the top of the head, neck, shoulders, and waist. The head-neck line is defined as the line connecting the key points of the top of the head and the neck, and the waist line is defined as the line connecting the key points of the left and right waists. The angle between these two lines on the two-dimensional imaging plane of the camera is calculated. Assuming that for the image of region C... Analysis revealed that the angle between the head and neck line and the horizontal direction was 82 degrees, and the angle between the waist line and the horizontal direction was 7 degrees, with a total angle of 75 degrees. A horizontal posture judgment threshold of 60 degrees was used for screening. This threshold was based on the following: measurements of 200 standard sitting posture images and 200 standing posture images showed that the average angle between the head and neck line and the waist line was approximately 75 degrees in the sitting posture, while in the standing posture, they were nearly parallel, with an average angle of less than 20 degrees. Therefore, 60 degrees was chosen as the threshold for distinguishing between sitting and standing postures. Since 75 degrees is greater than 60 degrees, region C was determined to be a sitting posture, and its number C was added to the sitting posture state block number set. The final obtained sitting posture state block number set is [C].

[0082] The control number binding submodule calls the static area number list and the sitting posture state block number set, extracts the intersection of the two numbers, associates the spatial mapping number of each block in the air conditioning outlet image, completes coordinate binding based on the number consistency, and obtains the target air supply control area number.

[0083] The system retrieves the list of static area numbers [C] and the set of seated state tile numbers [C], extracts the intersection of these two lists (area numbers that simultaneously meet the conditions of "static" and "seated"), and obtains the intersection result [C] through set operations. This intersection is then associated with an image of an indoor air conditioning vent. This image has been pre-divided into tiles corresponding to an indoor spatial grid, and each tile has been assigned a unique spatial mapping number. For example, the area directly below the air conditioning vent corresponds to numbers Z001 to Z0 in the spatial grid. The 16 blocks of 16, and the calculated area C where the seated person is located, has spatial grid coordinates {(39,34),(40,34),(39,35),(40,35)} that coincide with the numbers Z014 and Z015 in the air outlet block, as shown in Table 2. Based on the principle of number consistency, the coordinate binding is completed, and the numbers Z014 and Z015 of the air outlet block are associated with the number C of the person area. The final target air supply control area number is [Z014,Z015].

[0084] Table 2: Correlation Table between Air Conditioning Vents and Personnel Area Numbers

[0085]

[0086] As shown in Table 2, this table illustrates the binding relationship between personnel areas and air conditioning vent blocks, ensuring the accuracy of air supply control.

[0087] The wind direction and speed linkage module includes:

[0088] The angle difference extraction submodule reads the air conditioning outlet image corresponding to the target air supply control area number, extracts the coordinate position of the center point of the air duct and the center point of the control area block, calculates the angle value of the line connecting the two points according to the layer reference axis, records the angle results in the order of the numbers, and establishes a list of air duct direction deviations.

[0089] Read the air supply images corresponding to the target air supply control area numbers [Z014, Z015]. For number Z014, extract the preset coordinates of the center point of the corresponding air vent in the air conditioning system model, and denot them as... Simultaneously, extract the coordinates of the center point of the control area Z014, and denot them as follows: Based on the unified layer settings, with the top-left corner of the image as the origin and the horizontal X-axis pointing to the right as the positive direction, calculate the line connecting the two points. Angle value with the horizontal reference axis The calculation process is as follows Substitute the values, The included angle value is -135.0 degrees, which is the deviation of the air guide direction. Similarly, the included angle value of Z015 is calculated to be -136.5 degrees. These angle results are recorded in numerical order to create a list of air guide direction deviations with the content [(-135.0,Z014),(-136.5,Z015)].

[0090] The airflow coverage judgment submodule, based on the air duct direction deviation list, overlays the current airflow layer coverage map, extracts the contact area between the corresponding airflow boundary of each air outlet and the target tile boundary, calculates the ratio of the intersection side length to the perimeter length of the tile, and determines whether the airflow effectively covers the target area according to the preset contact threshold, thus obtaining the airflow contact state matrix of the target area.

[0091] Based on the list of air vent direction deviations [(-135.0,Z014),(-136.5,Z015)], retrieve the airflow coverage simulation layer under the current air conditioning set fan speed. This layer depicts the airflow diffusion boundary in the room at the current air vent angle. For number Z014, extract its corresponding airflow boundary line segment. The four boundary line segments of the target block Z014 ,calculate and The contact area, i.e. the length of the intersecting line segment, is calculated as follows: assuming the perimeter of block Z014 is 0.4 meters (0.1 meters × 4), the calculated length of the intersecting side is 0.08 meters. Therefore, the ratio of the two is... The system determines the coverage based on a preset contact threshold of 0.25. The logic behind this threshold is that effective coverage requires the airflow to contact at least one-quarter of the perimeter of the target area to form an enveloping airflow. When the ratio of the intersecting side length to the perimeter length of the tile is greater than 0.25, the corresponding area is marked as "effective coverage" in the airflow contact state matrix of the target area. When the ratio is not greater than 0.25, it is marked as "insufficient coverage". Since the calculated ratio of 0.20 is not greater than 0.25, the corresponding position of Z014 in the state matrix is ​​marked as "insufficient coverage". The same calculation is performed on Z015. Assuming its ratio is 0.18, it is also marked as "insufficient coverage", thus obtaining the airflow contact state matrix of the target area.

[0092] The linkage parameter calculation submodule calls the airflow contact state matrix of the target area, identifies the area number with insufficient coverage, and calculates the wind direction correction value and wind speed compensation amount respectively by combining the wind direction deviation value corresponding to each number and the diffusion direction range in the current wind speed layer. It integrates the correction data and generates wind direction and wind speed linkage adjustment items.

[0093] The airflow contact state matrix of the target area is invoked to identify the area number marked as "insufficiently covered," such as Z014. Combining this with the wind direction deviation value of -135.0 degrees corresponding to this number, and the fact that the main diffusion angle range of the airflow at this wind speed level is ±15 degrees as found in the current wind speed layer, the wind direction correction value and wind speed compensation amount are calculated respectively. Specifically, the wind direction correction value is set to an adjustment amount equal in magnitude but opposite in direction to the wind direction deviation value; that is, the wind direction correction value is... Degree, the absolute value of the deviation in the air guiding direction. Whether the angle deviation exceeds the preset threshold of 45 degrees is checked. This threshold was determined experimentally. When the wind guide angle deviation exceeds 45 degrees, even if the angle is corrected, the current wind speed alone is insufficient to effectively deliver the airflow to the target location, requiring an increase in airflow. Since 135.0 degrees exceeds 45 degrees, in addition to calculating the wind guide angle correction value, the wind speed compensation is also set to increase the current wind speed level by one level. For example, if the current wind speed is "low," it will be increased to "medium," and the wind guide angle correction value will be adjusted accordingly. The adjustment amount for increasing the wind speed level by one grade is integrated to generate a wind direction and wind speed linkage adjustment item for region Z014. The same calculation is performed on Z015 to generate a wind direction and wind speed linkage adjustment item for region Z015. Assume its correction value is... The wind speed level has been increased by one level.

[0094] The energy-saving control execution module includes:

[0095] The air guide angle adjustment submodule extracts the air guide angle setting value corresponding to each air supply number based on the wind direction and wind speed linkage adjustment item. The control device adjusts the angle of the air guide mechanism according to the setting to complete the actual angle deflection operation. It records the air outlet number, the angle before and after correction and the response time information to form an air guide angle adjustment record table.

[0096] Based on the generated wind direction and speed linkage adjustment items, the wind direction angle setting value corresponding to air supply number Z014 is extracted (the current wind direction angle plus a correction value; for example, assuming the initial wind direction angle is 0 degrees, the setting value is...). (degrees), and the guide angle setting value corresponding to number Z015 (assuming it is). The control device adjusts the servo motor of the air guide mechanism associated with the Z014 and Z015 areas according to the set value to complete the actual air guide plate angle deflection operation. At the same time as the operation is executed, the air outlet number, the angle before correction, the angle after correction, and the precise time information of the operation are recorded, as shown in Table 3.

[0097] Table 3: Guide Angle Adjustment Record Sheet

[0098]

[0099] As shown in Table 3, this table records the specific parameters for each adjustment of the air guide angle, forming an air guide angle adjustment record table.

[0100] The wind speed mode setting submodule calls the air supply number in the wind guide angle adjustment record table, associates the wind speed trajectory distribution status corresponding to the number, and sets the air supply strategy to be adopted for the current number according to the judgment rules based on the wind direction deflection amplitude and trajectory diffusion range parameters, and obtains the air supply mode status table.

[0101] The air supply number in the wind direction angle adjustment record table is called, for example, Z014, and the corresponding wind speed trajectory distribution state is associated with this number. This state is obtained from the airflow simulation model and describes the diffusion pattern of the airflow at a specific wind speed and angle, based on the recorded wind direction deflection amplitude, i.e. The parameters of the airflow trajectory diffusion range under the current wind speed level, such as an effective coverage width of 0.8 meters at a "medium" wind speed, are classified and set according to preset judgment rules. These rules are: when the absolute value of the wind direction deflection is greater than 45 degrees and the target area width is less than the current wind speed diffusion range, a "precise air delivery" strategy is adopted; when the absolute value of the deflection is less than 45 degrees, a "soft scanning" strategy is adopted; and when the target area width is greater than the diffusion range, a "wide-area coverage" strategy is adopted. Due to the deflection amplitude of Z014... The degree is greater than 45 degrees, and the width of the target personnel area (about 0.4 meters) is less than the diffusion range of 0.8 meters. Therefore, the air supply strategy to be adopted for the current number Z014 is set to "precise air supply". This result is stored in the air supply mode status table, such as [Z014, precise air supply], [Z015, precise air supply]. The air supply mode status table is then retrieved.

[0102] The control instruction writing submodule extracts the control number and its corresponding status identifier according to the air supply mode status table, combines the timestamp information in the image frame sequence, writes it into the synchronization record data table, and appends the number index and operation identifier to the instruction field to obtain the air conditioning control output status instruction.

[0103] Based on the contents of the air supply mode status table [Z014, Precise Air Supply], the control number Z014 and its corresponding status identifier "Precise Air Supply" are extracted. Combined with the timestamp information in the current image frame sequence, such as 09:30:16.000, this data is written into a synchronous record data table. At the same time, the target number index Z014 and the specific operation identifier are appended to the instruction field. This operation identifier converts "Precise Air Supply" into a control code that the device can recognize, such as "MODE_FOCUS", and converts the previously calculated wind speed level "Medium" into the code "FAN_MED". The final generated instruction field content is "TARGET:Z014;ANGLE:135.0;FAN:FAN_MED;MODE:MODE_FOCUS", which is the air conditioning control output status instruction sent to the air conditioning main control board.

[0104] Please see Figure 3 A visual recognition-based intelligent air conditioner energy-saving control method includes the following steps:

[0105] S1: Acquire images from infrared vision cameras installed in indoor spaces, extract the boundaries of human head and shoulder contours in the images, exclude static areas with no infrared response, track the contours of areas with high thermal response frequency, calculate the center point of thermal areas and continuously project and overlay them in space, and filter effective layers by analyzing the stability of thermal area positions to obtain personnel image overlay layers.

[0106] S2: Based on the overlapping layer of human images, extract the center coordinates of the same human body region in continuous images, compare the intersection frequency of the inter-frame offset trajectory and the image boundary, mark the active offset region as the moving state, and mark the region with small fluctuation in the center position as the stationary state. Draw the behavior distribution layer in space and bind the state label to obtain the behavior distribution pattern in space.

[0107] S3: Based on the spatial behavior distribution pattern, locate the static state label set of blocks, detect the angle between the head and neck line and the waist line of the human body image in the block, filter out the area blocks with the angle tending to be horizontal, cross-filter the static block number and the sitting block number, associate the mapping relationship of the air conditioning air outlet area image, and obtain the target air supply control area number.

[0108] S4: Read the air conditioning outlet image corresponding to the target air supply control area number, mark the angle difference between the center of the air duct image and the center point of the control area, calculate the contact side length between the current airflow image coverage area and the target block, determine whether the airflow covers the area, and combine the air duct deflection and wind speed diffusion status to obtain the wind direction and wind speed linkage adjustment item.

[0109] S5: Based on the wind direction and wind speed linkage adjustment item, execute the control command to adjust the air outlet angle to the specified direction, and set the low frequency, variable speed or directional air supply mode in combination with the wind speed trajectory status. Write the control status number and image time frame synchronously into the record table and add the corresponding identifier to obtain the air conditioning control output status command.

[0110] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A visual recognition-based intelligent air conditioning energy-saving control system, characterized in that, The system includes: The personnel image acquisition module acquires images from an infrared vision camera, extracts the outline of the human head and shoulders, removes non-thermal areas, tracks high-response-frequency positions, performs spatial overlay of the outline center, determines the effective layer based on the overlap stability, and obtains the personnel image overlay layer. The dynamic action recognition module, based on the overlapping layer of the personnel images, extracts the center coordinates of the same area in continuous images, calculates the inter-frame offset and boundary contact frequency, classifies and sets static and moving states, generates a label binding layer, and obtains a behavior distribution pattern in space. The target area extraction module selects static concentrated blocks based on the behavior distribution pattern in the space, analyzes the angular relationship between the head and neck line and the waist line in the image, screens out blocks that present a horizontal posture, cross-matches the static label blocks and the sitting posture block numbers, establishes an air supply area number mapping, and obtains the target air supply control area number. The wind direction and speed linkage module reads the air outlet image corresponding to the target air supply control area number, marks the angular deviation between the air guide and the center of the area, calculates the contact length between the airflow trajectory and the image block, and generates a linkage record by combining the offset direction and wind speed distribution to obtain the wind direction and speed linkage adjustment item.

2. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 1, characterized in that: The personnel image overlay layer includes thermal response aggregation areas, continuous trajectory overlay points, and layer screening index numbers. The spatial behavior distribution pattern includes the distribution of movement status identifiers, static status aggregation blocks, and status label mapping relationships. The target air supply control area number includes the static area index number, the sitting posture determination block number, and the corresponding air vent number. The wind direction and wind speed linkage adjustment items include the air duct angle deviation, airflow coverage contact length, and wind speed diffusion discrimination factor.

3. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 1, characterized in that, The personnel image acquisition module includes: The infrared image receiving submodule acquires image data collected by infrared vision cameras deployed in the indoor area, scans and processes each frame of image, identifies and removes static blocks in the image whose infrared intensity remains unchanged for a long time, retains effective areas with infrared response fluctuations, and organizes continuous frame data in chronological order to form an infrared response image sequence set that can be used for subsequent processing. The human body boundary extraction submodule extracts the head and shoulder contour regions with human thermal radiation characteristics from the infrared response image sequence set, calls the edge gray-scale gradient change parameters to establish a contour distribution model, identifies continuous edge segments according to the preset gradient threshold and completes the region connection operation, filters out the boundary contour regions with human morphological characteristics, and obtains the head and shoulder structure boundary distribution matrix. The thermal overlap mapping submodule tracks the spatial coordinate changes of each contour center point in the image sequence based on the head and shoulder structure boundary distribution matrix, performs superposition calculation on the repeated occurrence positions of the center points in three-dimensional space, statistically analyzes the frequency of thermal response occurrence in space, sets a thermal response stability threshold, removes low-frequency response segments, retains the contour thermal area superposition results, and generates a personnel image superposition layer.

4. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 1, characterized in that, The dynamic action recognition module includes: The center trajectory extraction submodule acquires the contour data of each thermal response region in the overlapping layer of the personnel image, extracts the center point position of the same region in the image sequence, calls the contour centroid coordinates in each frame image, establishes a continuous temporal mapping relationship based on the frame sequence index, calculates the coordinate difference between adjacent frames and performs spatial registration, and generates the target area offset trajectory sequence. The state classification and judgment submodule, based on the target region offset trajectory sequence, counts the number of intersections of the centroid of each region's contour within the image boundary range, calls the spatial dispersion index of the intersection frequency and coordinate points in each region, and judges the state of each region according to the set image boundary intersection threshold and position fluctuation tolerance range to obtain the behavior state label matrix. The behavior pattern generation submodule extracts the projection coordinates of each marked area in space based on the behavior state label matrix, calls the contour positioning information in the personnel image overlay layer, establishes a one-to-one mapping relationship between area coordinates and label states, draws layers according to the mapping structure and completes label binding, and obtains the behavior distribution pattern in space.

5. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 1, characterized in that, The target region extraction module includes: The static block positioning submodule acquires the behavior distribution pattern within the space, locates the block areas marked as static, extracts the corresponding numbers and records their position coordinates in the planar layer, filters the set of block numbers with static characteristics through status labels, and establishes a static area number list. The sitting posture recognition submodule extracts the image of the person in the corresponding block based on the static area number list, detects the spatial structural relationship between the head and neck line and the waist line of the human body in the image, calculates the angle value formed by the two, and filters it with reference to the horizontal posture judgment threshold to obtain the sitting posture block number set. The control number binding submodule calls the static area number list and the sitting posture state block number set, extracts the intersection of the two numbers, associates the spatial mapping number of each block in the air conditioner air outlet image, completes coordinate binding based on the consistency of the numbers, and obtains the target air supply control area number.

6. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 1, characterized in that, The wind direction and wind speed linkage module includes: The angle difference extraction submodule reads the air conditioning air outlet image corresponding to the target air supply control area number, extracts the coordinate position of the center point of the air duct and the center point of the control area block, calculates the angle value of the line connecting the two points according to the layer reference axis, records the angle results in the order of the numbers, and establishes a list of air duct direction deviations. The airflow coverage judgment submodule, based on the air duct direction deviation list, overlays the current airflow layer coverage map, extracts the contact area between the corresponding airflow boundary of each air outlet and the target tile boundary, calculates the ratio of the intersection side length to the perimeter length of the tile, and determines whether the airflow effectively covers the target area according to the preset contact threshold, thereby obtaining the airflow contact state matrix of the target area. The linkage parameter calculation submodule calls the airflow contact state matrix of the target area, identifies the area number with insufficient coverage, and calculates the wind direction correction value and wind speed compensation amount respectively by combining the wind direction deviation value corresponding to each number and the diffusion direction range in the current wind speed layer, and integrates the correction data to generate wind direction and wind speed linkage adjustment items.

7. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 6, characterized in that: The process of determining whether the airflow effectively covers the target area according to the preset contact threshold is as follows: when the ratio of the intersecting side length to the perimeter length of the map is greater than the preset contact threshold, the corresponding area is marked as effectively covered in the airflow contact state matrix of the target area; when the ratio of the intersecting side length to the perimeter length of the map is not greater than the preset contact threshold, it is marked as insufficiently covered. The process of calculating the windward angle correction value and the wind speed compensation amount separately is as follows: the windward angle correction value is set to an adjustment amount that is equal in magnitude and opposite in direction to the windward direction deviation value. When the absolute value of the windward direction deviation value exceeds the preset angle deviation threshold, the wind speed compensation amount is also set to increase the current wind speed level based on the calculation of the windward angle correction value.

8. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 1, characterized in that, The system also includes: The energy-saving control execution module, based on the wind direction and wind speed linkage adjustment item, performs angle correction and wind speed selection, sets it to directional, variable speed or low frequency air supply, writes the control number and image frame into the record table and marks the control status, and obtains the air conditioning control output status command. The air conditioning control output status commands include air supply angle control parameters, wind speed adjustment mode number, and control status time synchronization record.

9. The intelligent air conditioning energy-saving control system based on visual recognition according to claim 8, characterized in that, The energy-saving control execution module includes: The air guide angle adjustment submodule extracts the air guide angle setting value corresponding to each air supply number based on the wind direction and wind speed linkage adjustment item. The control device adjusts the angle of the air guide mechanism according to the setting to complete the actual angle deflection operation. It records the air outlet number, the angle before and after correction and the response time information to form an air guide angle adjustment record table. The wind speed mode setting submodule calls the air supply number in the wind guide angle adjustment record table, associates the wind speed trajectory distribution status corresponding to the number, and sets the air supply strategy to be adopted for the current number according to the judgment rules based on the wind direction deflection amplitude and trajectory diffusion range parameters, and obtains the air supply mode status table. The control instruction writing submodule extracts the control number and its corresponding status identifier according to the air supply mode status table, combines the timestamp information in the image frame sequence, writes it into the synchronization record data table, and appends the number index and operation identifier to the instruction field to obtain the air conditioning control output status instruction.

10. A method for intelligent air conditioning energy-saving control based on visual recognition, characterized in that, The method, used in the vision-based intelligent air conditioning energy-saving control system according to any one of claims 1-9, includes the following steps: S1: Acquire images from infrared vision cameras installed in indoor spaces, extract the boundaries of human head and shoulder contours in the images, exclude static areas with no infrared response, track the contours of areas with high thermal response frequency, calculate the center point of thermal areas and continuously project and overlay them in space, and filter effective layers by analyzing the stability of thermal area positions to obtain personnel image overlay layers. S2: Based on the overlapping layer of the human images, extract the center coordinates of the same human body region in the continuous images, compare the intersection frequency of the inter-frame offset trajectory and the image boundary, mark the active offset region as the moving state, and mark the region with small fluctuation in the center position as the stationary state. Draw a behavior distribution layer in space and bind state labels to obtain a behavior distribution pattern in space. S3: Based on the behavior distribution pattern in the space, locate the static state label set of blocks, detect the angle between the head and neck line and the waist line of the human body image in the block, filter out the area blocks with the angle tending to be horizontal, cross-filter the static block number and the sitting block number, associate the mapping relationship of the air conditioning air outlet area image, and obtain the target air supply control area number. S4: Read the air conditioner air outlet image corresponding to the target air supply control area number, mark the angle difference between the center of the air duct image and the center point of the control area, calculate the contact side length between the current airflow image coverage area and the target block, determine whether the airflow covers the area, and combine the air duct deflection and wind speed diffusion state to obtain the wind direction and wind speed linkage adjustment item. S5: Based on the wind direction and wind speed linkage adjustment item, execute the control command to adjust the air outlet angle to the specified direction, and set the low frequency, variable speed or directional air supply mode in combination with the wind speed trajectory status. Write the control status number and image time frame synchronously into the record table and add the corresponding identifier to obtain the air conditioning control output status command.

Images