Visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam

Abstract

The present invention discloses a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, comprising the following steps: obtaining a monitored video image; successively calculating the phase information of the monitored video image and the optical flow information of the pixels in the monitored video image; performing equal-area region division on the monitored video image, and respectively calculating the optical flow information of the pixels within the regions; and respectively calculating the Gini impurity within different regions based on the optical flow information of the pixels within the regions to determine the degree of pixel motion disorder within the regions and perform intrusion recognition. The visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam is not affected by video image data samples at the practical application level, and has the advantages of low cost and being convenient for popularization and use; at the level of micro-target recognition, the present invention replaces the texture features in deep learning with pixel motion features, and has a high recognition accuracy when recognizing micro-targets.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydropower station safety monitoring, and relates to a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam. Background Art

[0002] As one of the main contents of the operation and maintenance work of hydropower stations, the water discharge of hydropower station dams is often sudden due to external factors such as power grid peak regulation, load changes, and water regime. Therefore, it is of great value to timely and accurately identify whether there are personnel or animals invading the downstream river channel. At present, the inspection of the downstream river channel during dam water discharge mainly adopts manual inspection, that is, staff take specific vehicles to patrol both banks of the river channel. This method is time-consuming and laborious, and the inspection range of the downstream river channel is large, and the cost of single inspection is high. In recent years, with the development of computer vision technology, especially with the improvement of image acquisition technology and image processing technology, visual surveillance has gradually become the main inspection method for river channels. However, due to the large monitoring range of the river channel and the small number of pixels occupied by personnel targets in the image, the existing monitoring methods based on image target recognition have poor effects; on the other hand, factors such as few image data samples and insufficient training of deep learning models also lead to low recognition accuracy; in addition, the current deep learning-based models are easily affected by external conditions such as light and illuminance, and the recognition results have a high degree of uncertainty, resulting in that the current monitoring system not only does not provide convenience for inspection personnel, but also increases the maintenance cost to a certain extent.

[0003] In summary, the existing technology has the problem of low intrusion recognition accuracy due to factors such as few pixels occupied by the recognition target and interference of the recognition process by light. Summary of the Invention

[0004] The purpose of the present invention is to provide a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, which solves the problem of low intrusion recognition accuracy in the existing technology due to factors such as few pixels occupied by the recognition target and interference of the recognition process by light.

[0005] The technical solution adopted by the present invention is a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, including the following steps:

[0006] Step 1: Obtain monitoring video images;

[0007] Step 2: Calculate the phase information of the monitoring video image and the optical flow information of the pixels in the monitoring video image in sequence;

[0008] Step 3: Divide the monitoring video image into equal-area regions, and calculate the optical flow information of the pixels in the regions respectively;

[0009] Step 4: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in the region to determine the degree of pixel motion chaos in the region and perform intrusion recognition.

[0010] The features of the present invention also lie in:

[0011] Step 2 includes:

[0012] Step 2.1: Perform convolution processing on the surveillance video image using the Gabor wavelet to obtain the phase information of the surveillance video image;

[0013] Step 2.2: Calculate the optical flow information of the pixels in the surveillance video image based on the phase information of the surveillance video image.

[0014] The convolution processing is as follows:

[0015] ,

[0016] where, represents the intensity information of the surveillance video image, represents the phase information of the surveillance video image, represents the horizontal pixel coordinate of the surveillance video image, represents the vertical pixel coordinate of the surveillance video image, represents the time of the surveillance video image, represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the convolution operation, represents the surveillance video image.

[0017] The Gabor wavelet is as follows:

[0018] ,

[0019] where, represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the Gabor wavelet function, represents the coordinate of the Gabor wavelet kernel in the horizontal direction, represents the coordinate of the Gabor wavelet kernel in the vertical direction, represents the wavelength of the Gabor wavelet, represents the direction of the Gabor wavelet, represents the phase shift of the Gabor wavelet, represents the standard deviation of the Gaussian factor of the Gabor wavelet Represents the spatial aspect ratio of the Gabor wavelet.

[0020] The optical flow information of the pixels in the monitored video image is as follows:

[0021] ,

[0022] Among them, Represents the phase information of the monitored video image, Represents the horizontal pixel coordinate of the monitored video image, Represents the vertical pixel coordinate of the monitored video image, Represents the time of the monitored video image, Represents the optical flow information in the horizontal direction of the monitored video image, Represents the optical flow information in the vertical direction of the monitored video image.

[0023] The calculation formula for the optical flow information of the pixels in the area is as follows:

[0024] ,

[0025] ,

[0026] Among them, Represents The optical flow information in the horizontal direction of the monitored video image at time Represents The optical flow information in the vertical direction of the monitored video image at time Represents The phase information of the monitored video image in the 0° direction at time Represents The phase information of the monitored video image in the 90° direction at time Represents the horizontal pixel coordinate of the monitored video image, Represents the vertical pixel coordinate in the monitored video image, Represents the time of the monitored video image.

[0027] Step four includes:

[0028] Step 4.1: Statistically calculate the distribution probability of the optical flow information of the pixels in different regions based on the optical flow information of the pixels in the region;

[0029] Step 4.2: Calculate the Gini impurity in different regions respectively based on the distribution probability of the optical flow of the pixels in different regions;

[0030] Step 4.3: Judge the degree of pixel motion disorder in the region based on the Gini impurity in the region and perform intrusion recognition;

[0031] When the Gini impurity within a region is 0, the optical flow values of all pixels within the region are consistent, indicating no intrusion; when the Gini impurity within the region is between 0 and 1, the optical flow values of all pixels within the region are inconsistent, indicating an intrusion.

[0032] The calculation formula for the distribution probability of the optical flow information of pixels within a region is as follows:

[0033] ,

[0034] where, represents the distribution probability of the optical flow information of pixels within region , represents a certain region within the surveillance video image, N represents region the number of pixels within, represents region the distribution of pixels within in the horizontal direction 、 represents region the distribution of pixels within in the vertical direction.

[0035] The calculation formula for the Gini impurity within a region is as follows:

[0036] ,

[0037] where, represents the Gini impurity of region , represents region the distribution probability of the optical flow information of pixels within.

[0038] The discrimination formula for the degree of pixel motion disorder within a region is as follows:

[0039] ,

[0040] where, represents the th region Gini impurity, represents the th region on the surveillance video image, represents the total number of regions after the division of the surveillance video image region.

[0041] The beneficial effects of the present invention are as follows: At the practical application level, the present invention is not affected by video image data samples and has the advantages of low cost and easy promotion; at the level of micro-target recognition, the present invention uses pixel motion features to replace texture features in deep learning, and has high recognition accuracy when recognizing micro-targets; in terms of anti-interference in recognition, the present invention uses image phase features to replace image gray features and has high recognition accuracy under conditions such as low light and poor lighting; in terms of convenience, compared with the current recognition method based on deep learning models, the present invention does not require training the model with data, does not require annotating video sample data, and has a high deployment efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] A visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, as Figure 1 shown, includes the following steps:

[0045] Step 1: Obtain surveillance video images;

[0046] Step 2: Calculate the phase information of the surveillance video images and the optical flow information of the pixels in the surveillance video images in sequence;

[0047] Step 2.1: Perform convolution processing on the surveillance video images using Gabor wavelets to obtain the phase information of the surveillance video images;

[0048] The convolution processing is as follows:

[0049] ,

[0050] where represents the intensity information of the surveillance video images, represents the phase information of the surveillance video images, represents the horizontal pixel coordinates of the surveillance video images, represents the vertical pixel coordinates of the surveillance video images, represents the time of the surveillance video images, represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the convolution operation, represents the surveillance video images;

[0051] The Gabor wavelet is as follows:

[0052] ,

[0053] where, represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the Gabor wavelet function, represents the coordinate of the Gabor wavelet kernel in the horizontal direction, represents the coordinate of the Gabor wavelet kernel in the vertical direction, represents the wavelength of the Gabor wavelet, represents the direction of the Gabor wavelet, represents the phase shift of the Gabor wavelet, represents the standard deviation of the Gaussian factor of the Gabor wavelet, represents the spatial aspect ratio of the Gabor wavelet;

[0054] Step 2.2: Calculate the optical flow information of the pixels in the monitored video image based on the phase information of the monitored video image;

[0055] The optical flow information of the pixels in the monitored video image is as follows:

[0056] ,

[0057] where, represents the phase information of the monitored video image, represents the horizontal pixel coordinate of the monitored video image, represents the vertical pixel coordinate of the monitored video image, represents the time of the monitored video image, represents the optical flow information in the horizontal direction of the monitored video image, represents the optical flow information in the vertical direction of the monitored video image;

[0058] Step Three: Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in each region respectively;

[0059] The calculation formula for the optical flow information of the pixels in the region is as follows:

[0060] ,

[0061] ,

[0062] where, represents The optical flow information in the horizontal direction of the surveillance video image at a moment, denote The optical flow information in the vertical direction of the surveillance video image at a moment, denote The phase information of the surveillance video image in the 0° direction at a moment, denote The phase information of the surveillance video image in the 90° direction at a moment, denote the horizontal pixel coordinates of the surveillance video image, denote the vertical pixel coordinates in the surveillance video image, denote the moment of the surveillance video image;

[0063] Step 4. Calculate the Gini impurity in different regions based on the optical flow information of the pixels in the region to judge the degree of pixel motion chaos in the region and perform intrusion recognition;

[0064] Step 4.1. Statistically calculate the distribution probability of the optical flow information of the pixels in different regions based on the optical flow information of the pixels in the region;

[0065] The calculation formula for the distribution probability of the optical flow information of the pixels in the region is as follows:

[0066] ,

[0067] where, denote the distribution probability of the optical flow information of the pixels in region in the region, denote a certain region in the surveillance video image, N denote region the number of pixels in the region, denote region the distribution of the pixels in the region in the horizontal direction 、 denote region the distribution of the pixels in the region in the vertical direction;

[0068] Step 4.2. Calculate the Gini impurity in different regions respectively based on the distribution probability of the optical flow of the pixels in different regions;

[0069] The calculation formula for the Gini impurity in the region is as follows:

[0070] ,

[0071] where, denote the Gini impurity of region , denote region the distribution probability of the optical flow information of the pixels in the region;

[0072] Step 4.3: Determine the degree of pixel motion chaos within the region based on the Gini impurity within the region and perform intrusion recognition;

[0073] When the Gini impurity within the region is 0, the optical flow values of all pixels within the region are the same, that is, there is no intrusion; when the Gini impurity within the region is between 0 and 1, the optical flow values of all pixels within the region are inconsistent, that is, there is an intrusion;

[0074] The discriminant formula for the degree of pixel motion chaos within the region is as follows:

[0075] ,

[0076] where, represents the Gini impurity of the th region , represents the th region on the surveillance video image, represents the total number of regions after the division of the surveillance video image region.

[0077] In the theory of computer vision, a grayscale image is composed of an image phase and an image intensity. Among them, the image phase describes the motion information of the structure within the image, and the image intensity represents the brightness information of the image. In this invention, image phase information is used for intrusion recognition, which is more robust than the structure motion measurement method based on grayscale information.

[0078] Embodiment 1

[0079] This embodiment proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, as Figure 1 shown, including the following steps:

[0080] Step 1: Obtain the surveillance video image;

[0081] Step 2: Calculate the image phase information of the surveillance video image and the optical flow information of the pixels in the surveillance video image in sequence;

[0082] Step 3: Divide the surveillance video image into equal-area regions, and calculate the optical flow information of the pixels within the regions respectively;

[0083] Step 4: Calculate the Gini impurity within different regions based on the optical flow information of the pixels within the regions to determine the degree of pixel motion chaos within the regions and perform intrusion recognition.

[0084] Embodiment 2

[0085] This embodiment proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, as Figure 1 shown, including the following steps:

[0086] Step 1: Obtain the monitored video image;

[0087] Step 2: Calculate the phase information of the monitored video image and the optical flow information of the pixels in the monitored video image in sequence;

[0088] Step 2.1: Perform convolution processing on the monitored video image using Gabor wavelets to obtain the phase information of the monitored video image;

[0089] Step 2.2: Calculate the optical flow information of the pixels in the monitored video image based on the phase information of the monitored video image;

[0090] Step 3: Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in each region respectively;

[0091] Step 4: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in each region to judge the degree of pixel motion disorder in the region and perform intrusion recognition.

[0092] Example 3

[0093] This example proposes a visual recognition method for personnel intrusion applicable to the scene of water discharge from a hydropower dam, as Figure 1 shown, including the following steps:

[0094] Step 1: Obtain the monitored video image;

[0095] Step 2: Calculate the phase information of the monitored video image and the optical flow information of the pixels in the monitored video image in sequence;

[0096] Step 2.1: Perform convolution processing on the monitored video image using Gabor wavelets to obtain the phase information of the monitored video image;

[0097] The convolution processing is as follows:

[0098] ,

[0099] where, represents the intensity information of the monitored video image, represents the phase information of the monitored video image, represents the horizontal pixel coordinate of the monitored video image, represents the vertical pixel coordinate of the monitored video image, represents the time of the monitored video image, represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the convolution operation, represents the monitored video image;

[0100] The Gabor wavelet is as follows:

[0101] ,

[0102] where represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the Gabor wavelet function, represents the horizontal coordinate of the Gabor wavelet kernel, represents the vertical coordinate of the Gabor wavelet kernel, represents the wavelength of the Gabor wavelet, represents the direction of the Gabor wavelet, represents the phase offset of the Gabor wavelet, represents the standard deviation of the Gaussian factor of the Gabor wavelet, represents the spatial aspect ratio of the Gabor wavelet;

[0103] Step 2.2: Calculate the optical flow information of the pixels in the monitored video image based on the phase information of the monitored video image;

[0104] Step Three: Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in each region respectively;

[0105] Step Four: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in each region to determine the degree of pixel motion chaos in the region and perform intrusion recognition.

[0106] Example 4

[0107] This example proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower dam, as Figure 1 shown, including the following steps:

[0108] Step One: Obtain the monitored video image;

[0109] Step Two: Calculate the phase information of the monitored video image and the optical flow information of the pixels in the monitored video image in sequence;

[0110] Step 2.1: Perform convolution processing on the monitored video image using a Gabor wavelet to obtain the phase information of the monitored video image;

[0111] Step 2.2: Calculate the optical flow information of the pixels in the monitored video image based on the phase information of the monitored video image;

[0112] The optical flow information of the pixels in the monitored video image is as follows:

[0113] ,

[0114] Among them, represents the phase information of the monitored video image, represents the horizontal pixel coordinate of the monitored video image, represents the vertical pixel coordinate of the monitored video image, represents the time of the monitored video image, represents the optical flow information in the horizontal direction of the monitored video image, represents the optical flow information in the vertical direction of the monitored video image;

[0115] Step 3: Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in each region respectively;

[0116] Step 4: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in each region to judge the degree of pixel motion chaos in the region and perform intrusion recognition.

[0117] Example 5

[0118] This example proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, as Figure 1 shown, including the following steps:

[0119] Step 1: Obtain the monitored video image;

[0120] Step 2: Calculate the phase information of the monitored video image and the optical flow information of the pixels in the monitored video image in sequence;

[0121] Step 3: Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in each region respectively;

[0122] The calculation formula for the optical flow information of the pixels in the region is as follows:

[0123] ,

[0124] ,

[0125] Among them, represents the optical flow information in the horizontal direction of the monitored video image at time represents the optical flow information in the vertical direction of the monitored video image at time represents the phase information of the monitored video image at time in the 0° direction, represents the phase information of the monitored video image at time in the 90° direction, represents the horizontal pixel coordinate of the surveillance video image, represents the vertical pixel coordinate in the surveillance video image, represents the time of the surveillance video image;

[0126] Step 4: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in the region to determine the degree of pixel motion chaos in the region and perform intrusion recognition.

[0127] Embodiment 6

[0128] This embodiment proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, as Figure 1 shown, including the following steps:

[0129] Step 1: Obtain the surveillance video image;

[0130] Step 2: Calculate the phase information of the surveillance video image and the optical flow information of the pixels in the surveillance video image in sequence;

[0131] Step 3: Divide the surveillance video image into equal-area regions, and calculate the optical flow information of the pixels in the regions respectively;

[0132] Step 4: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in the region to determine the degree of pixel motion chaos in the region and perform intrusion recognition;

[0133] Step 4.1: Statistically calculate the distribution probability of the optical flow information of the pixels in different regions based on the optical flow information of the pixels in the region;

[0134] Step 4.2: Calculate the Gini impurity in different regions respectively based on the distribution probability of the optical flow of the pixels in different regions;

[0135] Step 4.3: Determine the degree of pixel motion chaos in the region based on the Gini impurity in the region and perform intrusion recognition;

[0136] When the Gini impurity in the region is 0, the optical flow values of all pixels in the region are the same, that is, there is no intrusion; when the Gini impurity in the region is between 0 and 1, the optical flow values of all pixels in the region are inconsistent, that is, there is an intrusion.

[0137] Embodiment 7

[0138] This embodiment proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, as Figure 1 shown, including the following steps:

[0139] Step 1: Obtain the surveillance video image;

[0140] Step 2: Calculate the phase information of the surveillance video image and the optical flow information of the pixels in the surveillance video image in sequence;

[0141] Step 3. Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in each region respectively;

[0142] Step 4. Calculate the Gini impurity in different regions based on the optical flow information of the pixels in each region to judge the degree of pixel motion chaos in the region and perform intrusion recognition;

[0143] Step 4.1. Statistically calculate the distribution probability of the optical flow information of the pixels in different regions based on the optical flow information of the pixels in each region;

[0144] The calculation formula for the distribution probability of the optical flow information of the pixels in the region is as follows:

[0145] ,

[0146] where, represents the distribution probability of the optical flow information of the pixels in region ; represents a certain region in the monitored video image; N represents the number of pixels in region represents the distribution of the pixels in region in the horizontal direction 、 represents the distribution of the pixels in region in the vertical direction;

[0147] Step 4.2. Calculate the Gini impurity in different regions based on the distribution probability of the optical flow of the pixels in different regions respectively;

[0148] Step 4.3. Judge the degree of pixel motion chaos in the region based on the Gini impurity in the region and perform intrusion recognition;

[0149] When the Gini impurity in the region is 0, the optical flow values of all pixels in the region are the same, that is, there is no intrusion; when the Gini impurity in the region is between 0 and 1, the optical flow values of all pixels in the region are different, that is, there is an intrusion.

[0150] Example 8

[0151] This example proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, as Figure 1 shown, including the following steps:

[0152] Step 1. Obtain the monitored video image;

[0153] Step 2. Calculate the phase information of the monitored video image and the optical flow information of the pixels in the monitored video image in sequence;

[0154] Step 3. Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in each region respectively;

[0155] Step 4: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in the region to determine the degree of pixel motion chaos in the region and perform intrusion recognition;

[0156] Step 4.1: Statistically calculate the distribution probability of the optical flow information of the pixels in different regions based on the optical flow information of the pixels in the region;

[0157] Step 4.2: Calculate the Gini impurity in different regions respectively according to the distribution probability of the pixel optical flow in different regions;

[0158] The calculation formula for the Gini impurity in the region is as follows:

[0159] ,

[0160] where, represents the Gini impurity of region , represents the distribution probability of the optical flow information of the pixels in region ;

[0161] Step 4.3: Determine the degree of pixel motion chaos in the region based on the Gini impurity in the region and perform intrusion recognition;

[0162] When the Gini impurity in the region is 0, the optical flow values of all pixels in the region are the same, that is, there is no intrusion; when the Gini impurity in the region is between 0 and 1, the optical flow values of all pixels in the region are inconsistent, that is, there is an intrusion;

[0163] The discrimination formula for the degree of pixel motion chaos in the region is as follows:

[0164] ,

[0165] where, represents the Gini impurity of the th region , represents the th region on the surveillance video image, represents the total number of regions after the division of the surveillance video image region.

[0166] The present invention breaks through the problems currently plaguing object recognition algorithms based on deep learning, such as high training costs, low recognition accuracy for tiny objects in large scenes, and interference of recognition results by light, providing technical support for the intrusion of foreign objects during the water discharge process of hydropower dam. Considering the dynamic characteristics of the intrusion process of personnel and animals and relying on the river channel video monitoring points, the present invention proposes a visual recognition method for personnel intrusion applicable to the water discharge scenario of hydropower dams. Specifically, by obtaining the video images of the river channel video monitoring points, a method for calculating optical flow based on image phase is proposed to calculate the optical flow information of each pixel in the image. Further, the monitoring area is partitioned, and in combination with the optical flow information, the Gini impurity of the pixel movement in different regions is calculated to judge the activity degree of the local area, and the chaos degree is used to identify whether an intrusion is formed in the local area. The present invention can effectively improve the recognition accuracy of the intrusion of people and animals.

Claims

1. A visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam, characterized in that, It includes the following steps: Step 1: Obtain the monitored video image; Step 2: Calculate the phase information of the monitored video image and the optical flow information of the pixels in the monitored video image in sequence; Step 3: Divide the monitored video image into equal-area regions, and calculate the optical flow information of the pixels in the regions respectively; The calculation formula for the optical flow information of the pixels in the region is as follows: ， ， Among them, represents the optical flow information in the horizontal direction of the monitored video image at the moment represents the optical flow information in the vertical direction of the monitored video image at the moment represents the phase information of the monitored video image at the moment in the 0° direction represents the phase information of the monitored video image at the moment in the 90° direction represents the horizontal pixel coordinates of the monitored video image represents the vertical pixel coordinates in the monitored video image represents the moment of the monitored video image; Step 4: Calculate the Gini impurity in different regions based on the optical flow information of the pixels in the region to judge the degree of pixel motion chaos in the region and perform intrusion recognition; Step 4.1: Statistically calculate the distribution probability of the optical flow information of the pixels in different regions based on the optical flow information of the pixels in the region; The calculation formula for the distribution probability of the optical flow information of the pixels in the region is as follows: ， Among them, represents the probability distribution of pixel optical flow information within the region, represents a certain region within the monitored video image, N represents the region the number of pixels within, represents the region the distribution of pixels within in the horizontal direction 、 represents the region the distribution of pixels within in the vertical direction; Step 4.2: Calculate the Gini impurity in different regions respectively based on the distribution probability of the pixel optical flow in different regions; The calculation formula for the Gini impurity in the region is as follows: ， Among them, represents the Gini impurity of the region , and represents the probability distribution of the pixel optical flow information within the region . Step 4.3: Judge the degree of pixel motion chaos in the region based on the Gini impurity in the region and perform intrusion recognition; When the Gini impurity in the region is 0, the optical flow values of all pixels in the region are the same, that is, there is no intrusion; when the Gini impurity in the region is between 0 and 1, the optical flow values of all pixels in the region are inconsistent, that is, there is an intrusion; The discrimination formula for the degree of pixel motion chaos in the region is as follows: ， Among them, represents the th region of Gini impurity, represents the th region on the monitored video image, and represents the total number of regions after the division of the monitored video image regions.

2. The visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam according to claim 1, wherein The said Step 2 includes: Step 2.1: Perform convolution processing on the monitored video image using the Gabor wavelet to obtain the phase information of the monitored video image; Step 2.2: Calculate the optical flow information of the pixels in the monitored video image based on the phase information of the monitored video image.

3. The visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam according to claim 2, wherein The convolution processing is as follows: Among them, represents the intensity information of the surveillance video image, represents the phase information of the surveillance video image, represents the horizontal pixel coordinates of the surveillance video image, represents the vertical pixel coordinates of the surveillance video image, represents the time of the surveillance video image, represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the convolution operation, represents the surveillance video image.

4. The visual recognition method for personnel intrusion applicable to the water discharge scenario of a hydropower station dam according to claim 3, wherein The Gabor wavelet is as follows: ， Among them, represents the Gabor wavelet, represents the real part of the Gabor wavelet, represents the imaginary part of the Gabor wavelet, represents the Gabor wavelet function, represents the coordinate of the Gabor wavelet kernel in the horizontal direction, represents the coordinate of the Gabor wavelet kernel in the vertical direction, represents the wavelength of the Gabor wavelet, represents the direction of the Gabor wavelet, represents the phase shift of the Gabor wavelet, represents the standard deviation of the Gabor wavelet Gaussian factor, represents the spatial aspect ratio of the Gabor wavelet.

5. The method for visual recognition of personnel intrusion applicable to the water discharge scenario of a hydropower station dam according to claim 2, characterized in that The optical flow information of the pixels in the monitored video image is as follows: ， Among them, represents the phase information of the monitored video image, represents the horizontal pixel coordinate of the monitored video image, represents the vertical pixel coordinate of the monitored video image, represents the time of the monitored video image, represents the optical flow information in the horizontal direction of the monitored video image, represents the optical flow information in the vertical direction of the monitored video image.

Images