A method for recognizing scope image data

Abstract

The invention discloses a method for recognizing image data of a sight, and belongs to the technical field of image data processing. The method first collects image data, including RGB values, grayscale values ​​and position information of pixel points, and establishes a background environment model through the collected image data. Then, pixel point comparison is performed through the background environment model, and the newly collected image data is decomposed into two parts, including a background environment part and a non-background environment part, and an updated background environment model is obtained by updating weight calculation. Then, the image data of the non-background environment part is screened, and the probability that the pixel points of the non-background environment part belong to a mobile target is judged by an exponential distribution model, and a pixel point set of the mobile target image is obtained. Then, a target movement model and an image grayscale quantization model are established to describe the mobile target, and the image grayscale quantization model is compared with a reference sample for similarity through the Bhattacharyya coefficient principle to complete the establishment of an updated image recognition model.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to a method for recognizing sight image data. Background Art

[0002] The scope is an optical aiming device widely used in light weapon systems. The basic principle is to magnify the image of a distant target to improve the shooter's aiming accuracy. Scopes are usually divided into three categories according to their structural characteristics, including: telescopic, collimating and reflective. The history of scopes is long. The earliest aiming behavior can be traced back to the crossbow in Chinese history. The sight named "Wangshan" mounted on the crossbow can be regarded as the earliest mechanical sight. After the first industrial revolution, European mechanical workshops began to try to manufacture equipment with built-in optical aiming. By 1904, Zeiss in Germany officially developed mature optical scope manufacturing technology. In recent years, scopes have been widely used in the military field, and have gradually entered the civilian market such as commercial shooting ranges and sports shooting. With the continuous integration of scopes and modern electronic information technology, new photoelectric scopes such as holographic scopes, night vision scopes and thermal imaging scopes have been born. At present, the mainstream development trend of sights is intelligent integrated sights. By combining various sensor components such as CCD detectors or CMOS detectors, the sight converts the acquired optical image into an electrical signal and sends it to the processor for one-step processing, and displays the enhanced image to the user in real time. Therefore, high-quality image data processing methods are particularly important for sight image recognition.

[0003] At present, the Chinese invention patent with application number CN202310915583.X discloses a meta-learning optimized image recognition method and device, which discloses a meta-learning optimized image recognition method and device, including the following steps: randomly initialize parameters in the meta-training stage , randomly select a task set of size batch_size from the training set, and randomly divide each task into a support set and a query set, corresponding to the training sample and the test sample. Use the initialization parameters to perform several trainings on the support set, first save the initialization parameters of the current network, and then perform several trainings on the support set to obtain a set of updated parameters. Then use the new parameters to perform probability prediction on the query set, and calculate the loss of the task. The losses obtained by all tasks on the query set are summed to obtain the total loss, and a task set of size batch_size is randomly selected from the test set. For each task, two subsets of the support set and the query set are randomly divided to obtain the total test loss and recognition accuracy, and the image recognition is completed. Although the invention can effectively recognize static image data, the error is large when processing moving target image data, and it cannot effectively process moving target images in complex environments.

[0004] Traditional image recognition methods have many limitations when processing image data from riflescopes. Rifles often need to be used in environments with high complexity and high uncertainty, such as mountains, forests, and wilderness. Therefore, riflescope users often encounter complex situations where targets move quickly in low-light and dusty environments, making it difficult to clearly identify the acquired images. Summary of the invention

[0005] The technical problem solved by the present invention is that it is difficult for a user to clearly identify a moving target image obtained through a traditional optical sight under complex environmental conditions.

[0006] In view of the deficiencies in the prior art, the present invention provides the following technical solutions to solve the above technical problems: a method for recognizing image data of a sight is provided, firstly, image data is collected, including RGB values, grayscale values ​​and position information of pixels, and a background environment model is established through the collected image data. Then, pixel comparison is performed through the background environment model, and the newly collected image data is decomposed into two parts, including a background environment part and a non-background environment part, and an updated background environment model is obtained by updating the weight calculation. Then, the image data of the non-background environment part is screened, and the probability that the pixels of the non-background environment part belong to a mobile target is judged by an exponential distribution model, and a pixel set of the mobile target image is obtained. Then, a target movement model and an image grayscale quantization model are established to describe the mobile target, and the image grayscale quantization model is compared with the reference sample for similarity through the Bhattacharyya coefficient principle, and finally the establishment of the updated image recognition model is completed, which specifically includes the following steps:

[0007] As a further improvement of the present invention, K frames of image data are collected to initialize the background environment model. The collected image data includes the RGB value, gray value Gr and position information of the image pixel point. The gray value Gr is obtained by weighted average calculation of the RGB value, and its calculation expression is:

[0008]

[0009] Among them, R represents the red component of the image pixel, G represents the green component of the image pixel, and B represents the blue component value of the image pixel.

[0010] As a further improvement of the present invention, a normal distribution model is established by calculating the normal distribution probability density function, and the observed values ​​of the image pixels are obtained by using N independent normal distribution models. Describe it by the mean of the normal distribution of the first K frames at time t The probability density function corresponding to the N independent normal distributions is calculated, and its calculation expression is:

[0011]

[0012] =

[0013] Where K represents the number of frames. represents the pixel observation value at time t, represents the mean corresponding to the i-th normal distribution at time t, represents the probability density function corresponding to the i-th normal distribution function, represents the covariance corresponding to the i-th normal distribution at time t, represents pi, and e represents a natural constant.

[0014] As a further improvement of the present invention, the normal distribution corresponding to the pixel point observation value at time t is calculated by the probability density function corresponding to the N independent normal distributions to establish the background environment model. , its calculation expression is:

[0015]

[0016] Wherein, N represents the number of independent normal distributions, Represents the pixel observation value at time t The corresponding background environment model, Represents the weight corresponding to the i-th normal distribution at time t.

[0017] As a further improvement of the present invention, the new image data after collecting the K frames of images is obtained, and the pixel point observation values ​​of the collected new image data are calculated by the background environment model. Perform a one-by-one comparison:

[0018] If the observed value With the normal distribution mean If the absolute value of the difference is greater than or equal to the comparison threshold, the comparison is passed and the pixel point belongs to the non-background environment part. With the normal distribution mean If the absolute value of the difference is less than the comparison threshold, the comparison fails and the pixel belongs to the background environment. The calculation expression is:

[0019]

[0020] =

[0021] in, represents the mean corresponding to the i-th normal distribution at time t-1, represents the comparison threshold corresponding to the i-th normal distribution at time t, and K represents the number of frames.

[0022] As a further improvement of the present invention, after the pixel points are compared one by one, the weights are compared based on the comparison results. Update and calculate the updated weight , its calculation expression is:

[0023]

[0024] in, represents the weight corresponding to the i-th normal distribution at time t, represents the updated weight corresponding to the i-th normal distribution at time t, represents the learning rate;

[0025] By updating the weights The mean value corresponding to the normal distribution and the comparison threshold are corrected to calculate and obtain an updated environmental background model. , its calculation expression is:

[0026]

[0027]

[0028]

[0029] in, represents the updated mean corresponding to the i-th normal distribution at time t, represents the mean corresponding to the i-th normal distribution at time t-1, represents the updated comparison threshold corresponding to the i-th normal distribution at time t, represents the comparison threshold corresponding to the i-th normal distribution at time t-1, Represents the observed value The background environment model is updated accordingly.

[0030] As a further improvement of the present invention, the non-background environment pixel points are screened, and the probability that the non-background environment pixel points belong to the moving target is determined by an exponential distribution model:

[0031] If the probability that the non-background environment pixel belongs to the moving target is greater than or equal to the probability threshold , it is determined that the non-background part of the pixels belongs to the mobile target. If the probability that the non-background part of the pixels belongs to the mobile target is less than the probability threshold , then it is determined that the non-background environment pixels do not belong to the moving target, and the calculation expression is:

[0032] ( )=

[0033] ( )

[0034] in, represents the normalized distance between the pixel position of the non-background environment part and the centroid of the moving target, represents the paradigm distance threshold, represents the pixel position of the non-background environment part, ( ) indicates the The probability that the corresponding pixel belongs to a moving target, represents the centroid position of the moving target, represents the exponential weight parameter, Represents the probability threshold.

[0035] As a further improvement of the present invention, the moving target is described by a second-order autoregressive equation, and the target movement model is established to represent the movement of the target over time, and its calculation expression is:

[0036] - + = -

[0037] in, represents the moving target observation value at time t, represents the moving target observation value at time t-1, represents the error term.

[0038] As a further improvement of the present invention, grayscale values ​​are used instead of RGB values ​​to describe pixel information to improve the operation speed, a grayscale mapping function is established to describe the moving target image, and a grayscale quantization distribution model of the image is established. , its calculation expression is:

[0039]

[0040] =

[0041]

[0042] =

[0043] in, represents the pixel position of the non-background environment part, express The grayscale mapping function at represents the linear correlation coefficient, G represents the maximum grayscale quantization level, and m represents the grayscale quantization level value. represents the image grayscale quantization distribution model of the moving target, represents the corresponding value of the grayscale quantization level of the target at time t, C represents the normalization constant, represents the normalized distance between the pixel position of the non-background environment part and the centroid of the moving target, represents the size of the moving target area, represents the Kronecker function, Represents the kernel function.

[0044] As a further improvement of the present invention, similarity comparison is performed by using an image grayscale quantization model to convert the target observation value A collection of reference samples , the reference sample is compared by the Bhattacharyya coefficient principle With the sample to be identified The grayscale distribution of is measured for similarity, and the updated image recognition model is established through the calculated conditional probability distribution model. , its calculation expression is:

[0045] ,

[0046]

[0047] = +

[0048] in, Indicates the sample to be identified The corresponding grayscale quantization distribution at time t, Indicates the sample to be identified The corresponding grayscale quantization distribution at time t-1, Represents the reference sample and samples to be identified The corresponding Bhattacharyya coefficient is Indicates reference sample The average grayscale distribution state, Indicated in the reference sample The sample to be identified under the condition of average gray distribution The observation probability model, represents the control parameter, Represents the contribution rate parameter.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention establishes a background environment model to adapt to the use requirements of the sight under different background environment conditions, making the present invention applicable. At the same time, a normal distribution probability model is used to describe the background environment pixels, which effectively improves the utilization rate of the collected limited image data. By establishing a target movement model and an image grayscale quantization model, the computational complexity of image data processing is simplified, the operation speed is effectively improved, and the subsequent feature analysis of the moving target is convenient. By establishing an exponential distribution model with a memoryless characteristic, the accuracy of the probability judgment that the pixel point belongs to the moving target is enhanced. Then, the reference sample and the recognition sample are compared for similarity based on the Bhattacharyya coefficient principle to establish an updated image recognition model, which can efficiently complete the accurate recognition of the image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of the overall framework of the present invention;

[0051] Figure 2 A schematic diagram of a specific process provided for an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of image data recognition experimental results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0054] Example 1: Please refer to Figure 1The present invention provides a technical solution: a method for recognizing telescope image data, so as to solve the problem that the image of a mobile target obtained by a shooter through a telescope under complex environmental conditions is blurred and difficult to clearly identify the mobile target. The overall framework of an embodiment of the present invention is: first, image data is collected, including the RGB value, gray value and position information of the pixel point, and a background environment model is established through the collected image data. Then, pixel point comparison is performed through the background environment model, and the newly collected image data is decomposed into two parts, including a background environment part and a non-background environment part, and an updated background environment model is obtained by updating the weight calculation. Then, the image data of the non-background environment part is screened, and the probability that the pixel points of the non-background environment part belong to the mobile target is judged by the exponential distribution model, and the pixel point set of the mobile target image is obtained. Then, a target movement model and an image gray quantization model are established to describe the mobile target, and the image gray quantization model is compared with the reference sample for similarity through the Bhattacharyya coefficient principle to complete the establishment of the updated image recognition model.

[0055] Please refer to Figure 2 The specific implementation process of an embodiment of the present invention includes the following steps:

[0056] In one embodiment of the present invention, K frames of image data are collected to initialize the background environment model. The collected image data includes RGB values, grayscale values ​​Gr and position information of image pixels. The grayscale values ​​Gr are obtained by weighted average calculation of the RGB values. The calculation expression is:

[0057]

[0058] Among them, Gr represents the grayscale value of the image pixel, R represents the red component of the image pixel, G represents the green component of the image pixel, and B represents the blue component value of the image pixel.

[0059] In one embodiment of the present invention, a normal distribution model is established by calculating a normal distribution probability density function, and the observed values ​​of the image pixels are analyzed by N independent normal distribution models. Describe it by the mean of the normal distribution of the first K frames at time t The probability density function corresponding to the N independent normal distributions is calculated, and its calculation expression is:

[0060]

[0061] =

[0062] Where K represents the number of frames. represents the pixel observation value at time t, represents the mean corresponding to the i-th normal distribution at time t, represents the probability density function corresponding to the i-th normal distribution function, represents the covariance corresponding to the i-th normal distribution at time t, represents pi, and e represents a natural constant.

[0063] In one embodiment of the present invention, the normal distribution corresponding to the pixel point observation value at time t is calculated by the probability density function corresponding to the N independent normal distributions to establish the background environment model. , its calculation expression is:

[0064]

[0065] Wherein, N represents the number of independent normal distributions, Represents the pixel observation value at time t The corresponding background environment model, Represents the weight corresponding to the i-th normal distribution at time t.

[0066] In one embodiment of the present invention, new image data after collecting the K frames of images is obtained, and pixel observation values ​​of the collected new image data are calculated by the background environment model. Perform a one-by-one comparison:

[0067] If the observed value With the normal distribution mean If the absolute value of the difference is greater than or equal to the comparison threshold, the comparison is passed and the pixel point belongs to the non-background environment part. With the normal distribution mean If the absolute value of the difference is less than the comparison threshold, the comparison fails and the pixel belongs to the background environment. The calculation expression is:

[0068]

[0069] =

[0070] in, represents the mean corresponding to the i-th normal distribution at time t-1, represents the comparison threshold corresponding to the i-th normal distribution at time t, and K represents the number of frames.

[0071] In one embodiment of the present invention, after the pixel points are compared one by one, the weights are adjusted according to the comparison results. Update and calculate the updated weight , its calculation expression is:

[0072]

[0073] in, represents the weight corresponding to the i-th normal distribution at time t, represents the updated weight corresponding to the i-th normal distribution at time t, represents the learning rate;

[0074] By updating the weights The mean value corresponding to the normal distribution and the comparison threshold are corrected to calculate and obtain an updated environmental background model. , its calculation expression is:

[0075]

[0076]

[0077]

[0078] in, represents the updated mean corresponding to the i-th normal distribution at time t, represents the mean corresponding to the i-th normal distribution at time t-1, represents the updated comparison threshold corresponding to the i-th normal distribution at time t, represents the comparison threshold corresponding to the i-th normal distribution at time t-1, Represents the observed value The background environment model is updated accordingly.

[0079] Then, the non-background environment pixel points are screened, and the probability that the non-background environment pixel points belong to the moving target is determined by the exponential distribution model:

[0080] If the probability that the non-background environment pixel belongs to the moving target is greater than or equal to the probability threshold , it is determined that the non-background part of the pixels belongs to the mobile target. If the probability that the non-background part of the pixels belongs to the mobile target is less than the probability threshold , then it is determined that the non-background environment pixels do not belong to the moving target, and the calculation expression is:

[0081] ( )=

[0082] ( )

[0083] in, represents the normalized distance between the pixel position of the non-background environment part and the centroid of the moving target, represents the paradigm distance threshold, represents the pixel position of the non-background environment part, ( ) indicates the The probability that the corresponding pixel belongs to a moving target, represents the centroid position of the moving target, represents the exponential weight parameter, Represents the probability threshold.

[0084] In one embodiment of the present invention, a moving target is described by a second-order autoregressive equation, and the target movement model is established to represent the movement of the target over time, and its calculation expression is:

[0085] - + = -

[0086] in Represents the moving target observation value at time t represents the error term.

[0087] In one embodiment of the present invention, grayscale values ​​are used instead of RGB values ​​to describe pixel information to improve the operation speed, and a grayscale mapping function is established to describe the moving target image, and a grayscale quantization distribution model of the image is established. , its calculation expression is:

[0088]

[0089] =

[0090]

[0091] =

[0092] in, represents the pixel position of the non-background environment part, express The grayscale mapping function at represents the linear correlation coefficient, G represents the maximum grayscale quantization level, and m represents the grayscale quantization level value. represents the image grayscale quantization distribution model of the moving target, represents the corresponding value of the grayscale quantization level of the target at time t, C represents the normalization constant, represents the normalized distance between the pixel position of the non-background environment part and the centroid of the moving target, represents the size of the moving target area, represents the Kronecker function, Represents the kernel function.

[0093] In one embodiment of the present invention, the target observation value is compared by using an image grayscale quantization model. A collection of reference samples , the reference sample is compared by the Bhattacharyya coefficient principle With the sample to be identified The grayscale distribution of is measured for similarity, and the updated image recognition model is established through the calculated conditional probability distribution model. , its calculation expression is:

[0094] ,

[0095]

[0096] = +

[0097] in, Indicates the sample to be identified The corresponding grayscale quantization distribution at time t, Indicates the sample to be identified The corresponding grayscale quantization distribution at time t-1, Represents the reference sample and samples to be identified The corresponding Bhattacharyya coefficient is Indicates reference sample The average grayscale distribution state, Indicated in the reference sample The sample to be identified under the condition of average gray distribution The observation probability model, represents the control parameter, Represents the contribution rate parameter.

[0098] Example 2: Please refer to Figure 3 This is another embodiment of the present invention. This embodiment is different from the first embodiment in that it provides an experimental verification for identifying a moving target image. In order to verify and illustrate the technical effect adopted in this method, this embodiment adopts two traditional image recognition technical solutions to conduct comparative tests with the method of the present invention, and compares the test results by means of scientific argumentation to verify the effect of this method.

[0099] Please refer to Figure 3 The updated image data recognition method of the present invention, the image data recognition method based on the shadow blocking algorithm, and the image data recognition method based on the improved Bayesian algorithm are respectively used as method 1, method 2, and method 3, and the pixel offset size is used as the error value. The recognition results of the three methods on the same image data under the same experimental conditions are compared. The error of method 1 of the present invention is 38.4% lower than that of method 2, and 47.1% lower than that of method 3. The image recognition method of the present invention has the highest accuracy.

[0100] The present invention first collects image data, including the RGB value, grayscale value and position information of the pixel points, and establishes a background environment model through the collected image data. Then, the pixel points are compared through the background environment model, and the newly collected image data is decomposed into two parts, including the background environment part and the non-background environment part, and the updated background environment model is obtained by updating the weight calculation. Then, the image data of the non-background environment part is screened, and the probability that the pixel points of the non-background part belong to the mobile target is judged by the exponential distribution model, and the pixel point set of the mobile target image is obtained. Then, a target movement model and an image grayscale quantization model are established to describe the mobile target, and the image grayscale quantization model is compared with the reference sample for similarity through the Bhattacharyya coefficient principle to complete the establishment of the updated image recognition model.

[0101] The present invention effectively solves the problem that in application conditions with complex background environment characteristics, the target image obtained by the user through the traditional optical sight is often blurred, making it difficult for the user to clearly identify the moving target.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for recognizing sight image data, characterized in that: include: Collect image data and build background environment model; Collect new image data, perform pixel comparison on the new image data through the background environment model, decompose the image data into a background environment part and a non-background environment part, and update the background environment model; The probability that the pixel points of the non-background environment belong to the mobile target is determined by an exponential distribution probability model, and a set of image pixel points of the mobile target is obtained; Establishing a target movement model and an image grayscale quantization model to describe the moving target; Performing similarity comparison on the image grayscale quantization model by using the Bhattacharyya coefficient principle, and establishing an updated image recognition model; The non-background environment pixel points are screened, and the probability that the non-background environment pixel points belong to the moving target is determined by an exponential distribution model: If the probability that the non-background environment pixel belongs to the moving target is greater than or equal to the probability threshold , it is determined that the non-background environment part of the pixels belongs to the moving target; If the probability that the non-background environment pixel points belong to the moving target is less than the probability threshold , then it is determined that the non-background environment pixels do not belong to the moving target, and the calculation expression is: ( )= ； ( ) ； = ； in, represents the normalized distance between the pixel position of the non-background environment part and the centroid of the moving target, represents the paradigm distance threshold, represents the pixel position of the non-background environment part, ( ) indicates the The probability that the corresponding pixel belongs to a moving target, represents the centroid position of the moving target, represents the exponential weight parameter, represents the probability threshold; The moving target is described by a second-order autoregressive equation, and the target movement model is established, and its calculation expression is: - + = - ； in, represents the moving target observation value at time t, represents the moving target observation value at time t-1, represents the error term; The moving target image is described by establishing a grayscale mapping function, and a grayscale quantization distribution model of the image is established. , its calculation expression is: ； = ； ； in, represents the pixel position of the non-background environment part, express The grayscale mapping function at represents the linear correlation coefficient, G represents the maximum grayscale quantization level, and m represents the grayscale quantization level value. represents the image grayscale quantization distribution model of the moving target, represents the corresponding value of the grayscale quantization level of the target at time t, C represents the normalization constant, represents the normalized distance between the pixel position of the non-background environment part and the centroid of the moving target, represents the size of the moving target area, represents the Kronecker function, represents the kernel function; The moving target observation value is compared by similarity comparison through the image grayscale quantization model. A collection of reference samples , the reference sample is compared by the Bhattacharyya coefficient principle With the sample to be identified The grayscale distribution of is measured for similarity, and the updated image recognition model is established through the calculated conditional probability distribution model. , its calculation expression is: ， ； ； = + ； in, Indicates the sample to be identified The corresponding grayscale quantization distribution at time t, Indicates the sample to be identified The corresponding grayscale quantization distribution at time t-1, Represents the reference sample and samples to be identified The corresponding Bhattacharyya coefficient is Indicates reference sample The average grayscale distribution state, Indicated in the reference sample The sample to be identified under the condition of average gray distribution The observation probability model, represents the control parameter, Represents the contribution rate parameter.

2. The method for recognizing sight image data according to claim 1, wherein: The collected image data includes RGB values, gray values ​​and position information of image pixels. The gray value is obtained by weighted average calculation of the RGB value, and its calculation expression is: ； Among them, Gr represents the grayscale value of the image pixel, R represents the red component of the image pixel, G represents the green component of the image pixel, and B represents the blue component value of the image pixel.

3. A method for recognizing sight image data according to claim 2, characterized in that: The collected image data is described by establishing N independent normal distributions, and the means corresponding to the N independent normal distributions are calculated to obtain the probability density function corresponding to the N independent normal distributions, and the calculation expression thereof is: ； = ； Where K represents the number of frames. represents the pixel observation value at time t, represents the mean corresponding to the i-th normal distribution at time t, represents the probability density function corresponding to the i-th normal distribution function, represents the variance corresponding to the i-th normal distribution at time t, represents pi, and e represents a natural constant.

4. A method for recognizing sight image data according to claim 3, characterized in that: The normal distribution model corresponding to the pixel point observation value at time t is calculated through the probability density function corresponding to the N independent normal distributions to complete the establishment of the background environment model. The calculation expression is: ； Wherein, N represents the number of independent normal distributions, Represents the pixel observation value at time t The corresponding background environment model, Represents the weight corresponding to the i-th normal distribution at time t.

5. A method for recognizing sight image data according to claim 4, characterized in that: Collect new image data, and observe pixel points of the collected new image data through the background environment model Perform a one-by-one comparison: If the observed value With the normal distribution mean If the absolute value of the difference is greater than or equal to the comparison threshold, the comparison passes and the pixel point belongs to the non-background environment part; If the observed value With the normal distribution mean If the absolute value of the difference is less than the comparison threshold, the comparison fails and the pixel belongs to the background environment. The calculation expression is: ； = ； in, represents the mean corresponding to the i-th normal distribution at time t, represents the mean corresponding to the i-th normal distribution at time t-1, represents the comparison threshold corresponding to the i-th normal distribution at time t, and K represents the number of frames.

6. A method for recognizing sight image data according to claim 5, characterized in that: After the comparison is completed, the weight is compared with the result of the comparison. Update and calculate the updated weight , its calculation expression is: ； in, represents the weight corresponding to the i-th normal distribution at time t, represents the updated weight corresponding to the i-th normal distribution at time t, represents the learning rate; By updating the weights The mean value corresponding to the normal distribution and the comparison threshold are corrected to calculate and obtain an updated environmental background model. , its calculation expression is: ； ； ； in, represents the updated mean corresponding to the i-th normal distribution at time t, represents the mean corresponding to the i-th normal distribution at time t-1, represents the updated comparison threshold corresponding to the i-th normal distribution at time t, represents the comparison threshold corresponding to the i-th normal distribution at time t-1, Represents the observed value The background environment model is updated accordingly.

Images