Central air conditioning load allocation method based on computer vision crowd density analysis

Abstract

The application discloses a central air conditioner load deployment method based on computer vision people flow density analysis. The method comprises the following steps: collecting real-time video stream in a room provided with a central air conditioner, inputting a video image as an original background image into a single Gaussian background model for processing to obtain a denoising foreground image; using a target tracking KCF method to process to obtain a prediction image, and using a regional person number counting method to count the number of persons; establishing a person number-air conditioner load model; inputting the number of persons into the person number-air conditioner load model to output the air conditioner load of the central air conditioner as the current air conditioner load of the central air conditioner, and finally realizing real-time load deployment of the central air conditioner. The method effectively overcomes the problem that the central air conditioner of a building cannot carry out flexible load deployment, and effectively improves the utilization efficiency of the air conditioner load.

Description

Technical Field

[0001] This invention relates to a method for central air conditioning load allocation, specifically a method for central air conditioning load allocation based on computer vision-based human flow density analysis. Background Technology

[0002] Central air conditioning is widely used in commercial buildings, universities, and enterprises. However, due to the lack of unified allocation and proper management, the energy efficiency of central air conditioning systems has not been effectively improved. This not only wastes a lot of electricity but also causes unnecessary economic expenditures.

[0003] The need for energy-efficient systems like central air conditioning in smart buildings remains constant. However, effectively managing these loads to ensure optimal building temperatures while minimizing economic costs has been a persistent challenge. This is because targeted load allocation hasn't been implemented based on the specific application scenarios of the central air conditioning system. For example, how to allocate airflow to different areas based on the number of people in the building is crucial. Frequent occupancy leads to varying operational requirements for central air conditioning in different areas. The key to real-time central air conditioning allocation in building environments lies in accurately and promptly counting the number of people moving within the monitored areas. Summary of the Invention

[0004] To address the problems existing in the background technology, the present invention provides a central air conditioning load allocation method based on computer vision human flow density analysis.

[0005] The technical solution adopted in this invention is:

[0006] The central air conditioning load allocation method of the present invention includes the following steps:

[0007] S1. Real-time video streams of rooms with central air conditioning are captured using surveillance cameras. For each frame of the real-time video stream, the video image is used as the original background image and processed in a single Gaussian background model to obtain a denoised foreground image. Specifically, the output images of surveillance cameras with fixed positions are used to ensure that the background information of each frame of the obtained image does not change significantly.

[0008] S2. Based on each denoised foreground image, the target tracking KCF method is used to process and obtain the predicted image. The number of people in the room in the predicted image is calculated using the regional people counting method. That is, the target tracking KCF method is used to track people and targets in the detected location to realize the counting of the number of people in the location.

[0009] S3. Establish a personnel quantity-air conditioning load model.

[0010] S4. For the number of people in the room in each predicted image, input the number of people into the number of people-air conditioning load model. The number of people-air conditioning load model outputs the air conditioning load of the central air conditioning system, which is used as the current air conditioning load of the central air conditioning system, and finally realizes the real-time load allocation of the central air conditioning system.

[0011] In step S1, the video image is input into the single Gaussian background model as the original background image for processing. Specifically, the single Gaussian background model is first initialized to the original background image, that is, the original background image is processed under a preset image residual standard deviation value to obtain the foreground image. Then, the pixel value of each pixel in the foreground image and the original background image is detected. For the pixel values ​​of the same position of each original background image and the foreground image, the residual of the pixels at the same position of the original background image and the foreground image is determined to obtain the denoised foreground image.

[0012] The process involves determining the residual value of pixels at the same location in each original background and foreground image, specifically by calculating the pixel residual value based on the pixel values ​​at the same location in the original background and foreground images and comparing it with a preset pixel residual threshold T. h The comparison is performed when the calculated pixel residual value exceeds the pixel residual threshold T. h Then, the pixels of the current foreground image are retained, provided that the calculated pixel residual value does not exceed the pixel residual threshold T. h Then update the pixels of the current foreground image, and obtain the final denoised foreground image after the update is completed.

[0013] The first frame image is used as the original background image, and the initial value of the standard deviation of the image residual is set to S. td (init) = 0. The expected value for the initial background is set to: μ. t＝0 (x, y) = I(x, y, t = 0), with standard deviation σ t＝0 (x, y) = S td (init).

[0014] The single Gaussian background model calculates the probability of occurrence of each pixel in each frame of the image, and the probability of the background pixel at time t must satisfy: |I(x, y, t) - μ t-1 (x, t)|<λσ t-1 The probability of a foreground pixel appearing must satisfy: |I(x, y, t) - μ t-1 (x, y)|≥λσ t-1 Let I(x, y, t) be the pixel value at pixel (x, y) at time t, and μ be the pixel value at pixel (x, y). t-1 This represents the expected value of the foreground pixel at time t-1. σ represents the threshold coefficient. t-1 This represents the standard value of the foreground pixel at time t-1.

[0015] The pixel residual threshold T h Specifically as follows:

[0016] T h =λσ t-1

[0017] Where λ represents the threshold coefficient; σ t-1 This represents the standard value of the pixel in the previous frame of the denoised foreground image at time t-1.

[0018] The updating of the current foreground image pixels specifically refers to updating the pixel parameter values ​​of the current foreground image pixels, including the expected value and standard deviation, as detailed below:

[0019] μ t (x, y) = (1-α)*μ t-1 (x, y) + α*I(x, y, t)

[0020]

[0021] Where, μ t (x, y) and μ t-1 (x, y) represent the foreground image of the current frame at time t. - The expected value of the pixel in the foreground image at time t-1 and the previous frame, where x and y are the coordinates of the pixel in the foreground image, respectively; α represents the update factor; I(x, y, t) represents the expected value of the pixel in the foreground image at time t-1. - The pixel value of the pixel at time σ; t (x, y) and σ t-1 (x, y) represent the standard deviations of the pixels in the current frame foreground image at time t and the previous frame foreground image at time t-1, respectively.

[0022] In step S2, the predicted image is obtained after processing using the target tracking KCF method, and the number of people in the room in the predicted image is calculated using the area people counting method, as follows:

[0023]

[0024] Where, N t and N t-1 P represents the number of people in the room at time t in the current frame prediction image and at time t-1 in the previous frame prediction image, respectively; i This represents the target path vector of the i-th target detected in the predicted image by the KCF method for target tracking. This represents the detection region detected in the predicted image by the KCF method for object tracking. The normal vector of the boundary point (a, b, c, d).

[0025] The target tracking KCF method described above detects the target path vector P of the i-th target in the predicted image. i Specifically, the KCF method for target tracking detects the i-th target in the predicted image from the point where it enters the detection region. To leave the testing area The vector formed by the starting and ending points of the path.

[0026] In step S3, the personnel quantity-air conditioning load model is established as follows:

[0027] S = S norm -ΔS≤S total

[0028]

[0029] Among them, S and S norm ΔS represents the air conditioning load of the central air conditioning system in the room before and after adjustment at time t; ΔS represents the adjustment amount of the air conditioning load of the central air conditioning system in the room at time t; S total ρ represents the total air conditioning load of the central air conditioning system; N represents the output coefficient of the central air conditioning system's air conditioning load. max Indicates the maximum number of people that the room can accommodate; N t Indicates the predicted image in the current frame at time t - The number of people in the room at any given time.

[0030] The beneficial effects of this invention are:

[0031] This invention relates to a real-time load allocation method for central air conditioning based on human flow density analysis using computer vision technology. It utilizes image information output from fixed-position monitoring cameras and the KCF target tracking method to detect the movement of people in different areas of a building and to count the number of people. Finally, the central air conditioning controller allocates the air conditioning output in each monitored room to regulate the temperature of the building, thereby maximizing the efficiency of air conditioning energy consumption and demonstrating good energy-saving potential. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method of the present invention;

[0033] Figure 2 This is a schematic diagram of the initialization of the original background image;

[0034] Figure 3 This is a schematic diagram of a denoised foreground image;

[0035] Figure 4 This is a schematic diagram illustrating the effect of the KCF method on person detection based on denoised foreground images. Detailed Implementation

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] like Figure 1 As shown, the central air conditioning load allocation method of the present invention includes the following steps:

[0038] S1. Real-time video streams of rooms equipped with central air conditioning are captured using surveillance cameras. For each frame of the real-time video stream, the video image is used as the original background image and processed in a single Gaussian background model to obtain a denoised foreground image. Specifically, images from surveillance cameras with fixed positions are used to ensure that the background information of each frame does not change significantly. Figure 2 and Figure 3 The figures shown are schematic diagrams of the initialization of the original background image and the denoised foreground image, respectively.

[0039] In step S1, the video image is input into the single Gaussian background model as the original background image for processing. Specifically, the single Gaussian background model is first initialized to the original background image, that is, the original background image is processed to obtain the foreground image under a preset image residual standard deviation value, which is specifically set to 20. Then, the pixel value of each pixel in the foreground image and the original background image is detected. For the pixel values ​​of the same position in each original background image and the foreground image, the residual of the pixels at the same position in the original background image and the foreground image is determined to obtain the denoised foreground image.

[0040] For the pixel values ​​at the same positions in each original background and foreground image, a residual determination is performed on the pixels at the same positions in the original background and foreground images. Specifically, the pixel residual value is calculated based on the pixel values ​​at the same positions in the original background and foreground images and compared with a preset pixel residual threshold T. h The comparison is performed. If the calculated pixel residual value exceeds the pixel residual threshold Th, the pixels of the current foreground image are retained; otherwise, the calculated pixel residual value does not exceed the pixel residual threshold T. h Then update the pixels of the current foreground image, and obtain the final denoised foreground image after the update is completed.

[0041] The first frame image is used as the original background image, and the initial value of the standard deviation of the image residual is set to S. td (init) = 0. The expected value for the initial background is set to: μ. t＝0(x, y) = I(x, y, t = 0), with standard deviation σ t＝0 (x, y) = S td (init).

[0042] The single Gaussian background model calculates the probability of occurrence of each pixel in each frame of the image, and the probability of the background pixel at time t must satisfy: |I(x, y, t) - μ t-1 (x, t)|<λσ t-1 The probability of a foreground pixel appearing must satisfy: |I(x, y, t) - μ t-1 (x, t)|≥λσ t-1 Let I(x, y, t) be the pixel value at pixel (x, y) at time t, and μ be the pixel value at pixel (x, y). t-1 Let λ represent the expected value of the foreground pixel at time t-1, and let σ represent the threshold coefficient. t-1 This represents the standard value of the foreground pixel at time t-1.

[0043] Pixel residual threshold T h Specifically as follows:

[0044] T h =λσ t-1

[0045] Where λ represents the threshold coefficient, which is specifically set to λ = 3; σ t-1 This represents the standard value of the pixel in the previous frame of the denoised foreground image at time t-1.

[0046] Update the pixels of the current foreground image, specifically by updating the pixel parameter values ​​of the current foreground image pixels, including the expected value and standard deviation, as follows:

[0047] μ t (x, y) = (1-α)*μ t-1 (x, y) + α*I(x, y, t)

[0048]

[0049] Where, μ t (x, y) and μ t-1 (x, y) represent the expected values ​​of pixels in the current frame foreground image at time t and the previous frame foreground image at time t-1, respectively, where x and y are the coordinates of the pixels in the foreground image; α represents the update factor; I(x, y, t) represents the expected values ​​of pixels in the current frame foreground image at time t-1. - The pixel value of the pixel at time σ; t (x, y) and σ t-1 (x, y) represent the standard deviations of the pixels in the current frame foreground image at time t and the previous frame foreground image at time t-1, respectively.

[0050] S2. Based on each denoised foreground image, the predicted image is obtained after processing using the target tracking KCF method, such as... Figure 4 As shown, the number of people in the room in the predicted image is calculated using the regional people counting method; that is, the target tracking KCF method is used to track people and targets in the detected location to realize the counting of the number of people in the location.

[0051] In step S2, the predicted image is obtained after processing using the target tracking KCF method. The number of people in the room in the predicted image is then calculated using the area people counting method, as follows:

[0052]

[0053] Where, N t and N t-1 P represents the number of people in the room at time t in the current frame prediction image and at time t-1 in the previous frame prediction image, respectively; i This represents the target path vector of the i-th target detected in the predicted image by the KCF method for target tracking. This represents the detection region detected in the predicted image by the KCF method for object tracking. The normal vector of the boundary point (a, b, c, d).

[0054] The target path vector P of the i-th target detected in the predicted image by the KCF method for target tracking. i Specifically, the KCF method for target tracking detects the i-th target in the predicted image from the point where it enters the detection region. To leave the testing area The vector formed by the starting and ending points of the path.

[0055] S3. Establish a personnel quantity-air conditioning load model.

[0056] In step S3, the personnel quantity-air conditioning load model is established as follows:

[0057] S = S norm -ΔS≤S total

[0058]

[0059] Among them, S and S norm ΔS represents the air conditioning load of the central air conditioning system in the room before and after time t; ΔS represents the air conditioning load of the central air conditioning system in the room at time t. - The amount of air conditioning load controlled at any given time; S total ρ represents the total air conditioning load of the central air conditioning system; N represents the output coefficient of the central air conditioning system's air conditioning load.max Indicates the maximum number of people that the room can accommodate; N t Indicates the predicted image in the current frame at time t - The number of people in the room at any given time.

[0060] S4. For the number of people in the room in each predicted image, input the number of people into the number of people-air conditioning load model. The number of people-air conditioning load model outputs the air conditioning load of the central air conditioning system, which is used as the current air conditioning load of the central air conditioning system, and finally realizes the real-time load allocation of the central air conditioning system.

[0061] Based on the above steps, the system ultimately achieves real-time control of the switching relays of the air conditioning load through a controller, thereby enabling real-time allocation of the air conditioning load within the building. During air conditioning control, the temperature variation within the monitored area of ​​the building is guaranteed to not exceed 2°C. This invention effectively overcomes the problem of flexible load allocation in building central air conditioning systems, and effectively improves the utilization efficiency of the air conditioning load.

Claims

1. A method for central air conditioning load allocation based on computer vision-based pedestrian density analysis, characterized in that: The method includes the following steps: S1. Use a surveillance camera to collect real-time video streams of rooms with central air conditioning installed. For each frame of video image in the real-time video stream, input the video image as the original background image into a single Gaussian background model for processing to obtain a denoised foreground image. S2. Based on each denoised foreground image, the target tracking KCF method is used to process it to obtain the prediction image, and the number of people in the room in the prediction image is calculated using the regional people counting method. S3. Establish a personnel quantity-air conditioning load model; S4. For the number of people in the room in each predicted image, input the number of people into the number of people-air conditioning load model. The number of people-air conditioning load model outputs the air conditioning load of the central air conditioning as the current air conditioning load of the central air conditioning, and finally realizes the real-time load allocation of the central air conditioning. In step S3, the personnel quantity-air conditioning load model is established as follows: in, and This indicates that the central air conditioning in the room is... The air conditioning load before and after real-time adjustment; This indicates that the central air conditioning in the room is... The amount of air conditioning load controlled at any given time; This indicates the total air conditioning load of the central air conditioning system; This indicates the output coefficient of the central air conditioning system's air conditioning load. Indicates the maximum number of people that the room can accommodate; Indicates the image predicted in the current frame. The number of people in the room at any given time.

2. The central air conditioning load allocation method based on computer vision-based pedestrian density analysis according to claim 1, characterized in that: In step S1, the video image is input into the single Gaussian background model as the original background image for processing. Specifically, the single Gaussian background model is first initialized to the original background image, that is, the original background image is processed under a preset image residual standard deviation value to obtain the foreground image. Then, the pixel value of each pixel in the foreground image and the original background image is detected. For the pixel values ​​of the same position of each original background image and the foreground image, the residual of the pixels at the same position of the original background image and the foreground image is determined to obtain the denoised foreground image.

3. The central air conditioning load allocation method based on computer vision-based pedestrian density analysis according to claim 2, characterized in that: The process involves determining the residual value of pixels at the same location in each original background and foreground image, specifically by calculating the pixel residual value based on the pixel values ​​at the same location in the original background and foreground images and comparing it with a preset pixel residual threshold. The comparison is performed when the calculated pixel residual value exceeds the pixel residual threshold. Then, the pixels of the current foreground image are retained, provided that the calculated pixel residual value does not exceed the pixel residual threshold. Then update the pixels of the current foreground image, and obtain the final denoised foreground image after the update is completed.

4. The central air conditioning load allocation method based on computer vision-based pedestrian density analysis according to claim 3, characterized in that: The pixel residual threshold Specifically as follows: in, Indicates the threshold coefficient; This indicates that the previous frame's denoised foreground image is in The standard value of a pixel at a given time.

5. A central air conditioning load allocation method based on computer vision-based pedestrian density analysis according to claim 3, characterized in that: The updating of the current foreground image pixels specifically refers to updating the pixel parameter values ​​of the current foreground image pixels, including the expected value and standard deviation, as detailed below: in, and These represent the foreground image in the current frame. The time and the foreground image of the previous frame at The expected value of the pixel at time t, where x and y are the coordinates of the pixel in the foreground image, respectively; Indicates the update factor; Indicates the foreground image in the current frame. The pixel value of the pixel at any given time; and These represent the foreground image in the current frame. The time and the foreground image of the previous frame at The standard deviation of the pixels at time step.

6. The central air conditioning load allocation method based on computer vision-based pedestrian density analysis according to claim 1, characterized in that: In step S2, the predicted image is obtained after processing using the target tracking KCF method, and the number of people in the room in the predicted image is calculated using the area people counting method, as follows: in, and These represent the predicted image in the current frame, respectively. In the time and the previous frame of the predicted image The number of people in the room at any given time; This represents the target path vector of the i-th target detected in the predicted image by the KCF method for target tracking. This represents the detection region detected in the predicted image by the KCF method for object tracking. boundary points The normal vector, .

7. A method for central air conditioning load allocation based on computer vision-based pedestrian density analysis according to claim 6, characterized in that: The target tracking KCF method described above determines the target path vector of the i-th target detected in the predicted image. Specifically, the KCF method for target tracking detects the i-th target in the predicted image from the point where it enters the detection region. To leave the testing area The vector formed by the starting and ending points of the path.

Images