Automatic aiming point calibration method and system based on machine learning

Abstract

The invention relates to an aiming point automatic calibration method and system based on machine learning, and belongs to the technical field of automatic calibration. The method comprises the following steps: obtaining real scene enhanced image data through a picture enhancement model according to current real scene image data; obtaining a reference target image through a target detection model according to the real scene enhanced image data; obtaining a gray level smooth aiming point image and a gray level smooth reference target image through a preprocessing model according to the aiming point image and the reference target image; according to the grayscale smooth aiming point image and the grayscale smooth reference target image, performing feature extraction through a hybrid convolutional neural network to obtain grayscale smooth aiming point image features and grayscale smooth reference target image features; and automatically calibrating the aiming point through an aiming point calibration model according to the gray smooth aiming point image features and the gray smooth reference target image features. Automatic aiming point calibration based on machine learning is realized, and the precision of automatic calibration is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic calibration, and in particular relates to a method and system for automatic calibration of an aiming point based on machine learning. Background Art

[0002] Current augmented reality (AR) gaming technology has made significant progress on smartphones and tablets, which can capture images of the real world through cameras and overlay virtual objects on the screen. However, such applications are limited by the camera performance and processing power of the device, and it is difficult to achieve a high-quality AR experience in low-light or complex environments. The existing aiming system relies on visual input and is easily affected by light conditions, especially at night or in low-light environments, where its accuracy and stability will be greatly reduced. Infrared technology can achieve clear imaging in darker environments without relying on visible light by capturing the thermal infrared signals radiated by objects, which can well break through the limitations of AR technology in low-light or complex environments. However, due to the influence of complex environmental conditions, long-distance imaging, and high-speed changes between the target and the background, the overall image presented is still blurred, lacking details and insufficient contrast. At the same time, the infrared emission signal is cone-shaped, all of which affect the accuracy of aiming. Therefore, how to reduce the deviation between the infrared emission signal and the aiming point has become a research hotspot. Summary of the invention

[0003] In order to solve the above problems existing in the prior art, the present invention provides a method and system for automatic calibration of aiming point based on machine learning. The purpose of the present invention can be achieved through the following technical solutions: A method for automatic calibration of aiming point based on machine learning, comprising: Acquire current real scene image data through a camera, and obtain real scene enhanced image data through a picture enhancement model according to the current real scene image data; Obtaining a reference target image through a target detection model according to the real scene enhanced image data; Acquire an aiming point image by emitting infrared light from a muzzle, and obtain a grayscale smoothed aiming point image and a grayscale smoothed reference target image through a preprocessing model according to the aiming point image and the reference target image; Extracting features of the grayscale smoothed aiming point image and the grayscale smoothed reference target image by using a hybrid convolutional neural network to obtain grayscale smoothed aiming point image features and grayscale smoothed reference target image features; The aiming point is automatically calibrated according to the grayscale smoothed aiming point image features and the grayscale smoothed reference target image features through an aiming point calibration model.

[0004] Preferably, obtaining the real scene enhanced image data through the picture enhancement model according to the current real scene image data includes: Obtaining current real scene smoothed image data through a mean filtering template according to the current real scene image data; Removing the current real scene smoothed image data from the current real scene image data to obtain a high frequency component; The real scene enhanced image data is obtained by performing a Hadamard operation according to the high frequency component and the current real scene image data.

[0005] Preferably, obtaining a reference target image through a target detection model according to the real scene enhanced image data comprises: Extracting a candidate target point area based on the real scene enhanced image data by using an improved density peak search method; Obtaining a positioning target mapping diagram based on the candidate target point area through a local discriminant model; Performing background suppression by a curvature filtering method according to the positioning target mapping image to obtain a background suppression target area; Obtaining a background suppression enhancement target area through Hadamard operation according to the background suppression target area and the high frequency component; The reference target image is obtained by threshold segmentation according to the background suppression and enhancement target area.

[0006] Preferably, the extracting the candidate target point area by improving the density peak search method according to the real scene enhanced image data comprises: Extracting the real scene enhanced image data to obtain pixel points, and obtaining the pixel point grayscale value by grayscale value calculation according to the pixel points; The minimum distance is obtained by calculating the distance between the pixel point and the gray value of the pixel point through the adjacent high-density point; Obtaining a local density peak joint characteristic factor by multiplying the pixel gray value and the minimum distance; According to the local density peak combined with the characteristic factors, the local density maximum point is obtained by sorting in descending order; Constructing a target area to be measured with the local maximum density point as the center, and obtaining a grayscale probability value by calculating the grayscale probability of the target area to be measured; The candidate target point region is obtained by sorting the grayscale probability values ​​in descending order.

[0007] Preferably, obtaining a positioning target mapping diagram according to the candidate target point area through a local discriminant model includes: Obtain the grayscale mean value of the central sub-block, the gradient value of the central sub-block, the grayscale mean value of each layer of sub-blocks, and the gradient value of each layer of sub-blocks; A grayscale mean difference and a gradient difference are calculated by subtracting the grayscale mean of the central sub-block, the gradient value of the central sub-block, the grayscale mean of each layer sub-block, and the gradient value of each layer sub-block; Obtaining a diagonal sub-block gradient product by multiplication according to the gradient difference, wherein the diagonal sub-block gradient product carries a direction label; The mean value of the gradient of each sub-block of each layer is calculated by averaging the gradient values ​​of each sub-block of each layer; the gradient increment is calculated by subtracting the gradient of each sub-block of each layer; the mean value of the gradient increment is calculated by subtracting the mean value of the gradient of each sub-block of each layer; Obtaining the positioning target mapping image through regional positioning judgment according to the diagonal sub-block gradient product, the grayscale mean difference, the gradient difference, the gradient increment, and the gradient mean increment; The regional positioning judgment includes: Determine whether the diagonal sub-block gradient product of the candidate target point area layer one and the candidate target point area layer two has four different direction labels, if yes, the area is marked as 0; if no, determine whether the grayscale mean difference and the gradient difference are less than or equal to 0, if yes, the area is marked as 0; if no, determine whether the gradient increment is greater than 0, if yes, obtain the positioning target mapping; if no, mark the area as 0.

[0008] Preferably, obtaining a grayscale smoothed aiming point image and a grayscale smoothed reference target image through a preprocessing model according to the aiming point image and the reference target image comprises: Obtaining an aiming point grayscale image and a reference target grayscale image by grayscale image conversion according to the aiming point image and the reference target image; The grayscale smoothed aiming point image and the grayscale smoothed reference target image are obtained by filtering the aiming point grayscale image and the reference target grayscale image.

[0009] Preferably, the automatically calibrating the aiming point through an aiming point calibration model according to the grayscale smoothed aiming point image features and the grayscale smoothed reference target image features comprises: The gyroscope emits infrared to obtain the emission time, emission direction and emission acceleration; According to the launch time, the grayscale smooth aiming point image feature and the grayscale smooth reference target image feature are matched by a brute force matching algorithm to obtain a matching result; The emission direction is screened according to the matching result to obtain the best aiming position.

[0010] An aiming point automatic calibration system based on machine learning, comprising an image enhancement module, a target detection module, a preprocessing module, a feature extraction module, and an aiming calibration module; The image enhancement module is used to obtain current real scene image data through a camera, and obtain real scene enhanced image data through an image enhancement model according to the current real scene image data; The target detection module is used to obtain a reference target image through a target detection model according to the real scene enhanced image data; The preprocessing module is used to acquire an aiming point image by emitting infrared light from a muzzle, and to obtain a grayscale smoothed aiming point image and a grayscale smoothed reference target image through a preprocessing model according to the aiming point image and the reference target image; The feature extraction module is used to extract features from the grayscale smoothed aiming point image and the grayscale smoothed reference target image through a hybrid convolutional neural network to obtain grayscale smoothed aiming point image features and grayscale smoothed reference target image features; The aiming calibration module is used to automatically calibrate the aiming point through an aiming point calibration model according to the grayscale smoothed aiming point image features and the grayscale smoothed reference target image features.

[0011] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned automatic calibration method of aiming point based on machine learning when executing the program.

[0012] A storage medium containing computer executable instructions, which are used to perform the above-mentioned machine learning-based aiming point automatic calibration method when executed by a computer processor.

[0013] The beneficial effects of the present invention are: (1) Through the image enhancement model, target detection model and hybrid convolutional neural network, the target scene is detected and features are extracted to reduce the interference of background noise and light changes on the calibration results and improve the accuracy of automatic calibration of the aiming point.

[0014] (2) The image quality in complex scenes is optimized through mean filtering, high-frequency component extraction and Hadamard operation. The target area is accurately located by improving the density peak search method, which enhances the adaptability to the scene and improves the accuracy of automatic calibration of the aiming point.

[0015] (3) By integrating feature extraction and regional positioning judgment, a screening mechanism is introduced in target detection and aiming point calibration, which can reduce the errors caused by image quality differences and environmental changes and ensure the robustness and stability of the aiming point automatic calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0017] Figure 1 The figure is a flow chart of an aiming point automatic calibration method based on machine learning of the present invention. DETAILED DESCRIPTION

[0018] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0019] See also Figure 1 , an aiming point automatic calibration method based on machine learning, comprising: S1: Acquire current real scene image data through a camera, and obtain real scene enhanced image data through a picture enhancement model according to the current real scene image data; S2: Obtaining a reference target image through a target detection model according to the real scene enhanced image data; S3: acquiring an aiming point image by emitting infrared light from the muzzle, and obtaining a grayscale smoothed aiming point image and a grayscale smoothed reference target image through a preprocessing model according to the aiming point image and the reference target image; S4: extracting features of the grayscale smoothed aiming point image and the grayscale smoothed reference target image through a hybrid convolutional neural network to obtain grayscale smoothed aiming point image features and grayscale smoothed reference target image features; S5: Automatically calibrate the aiming point according to the grayscale smoothed aiming point image features and the grayscale smoothed reference target image features through an aiming point calibration model.

[0020] In step S1, the image enhancement model includes: S101: Obtaining current real scene smoothed image data through a mean filtering template according to the current real scene image data; The filtering operator of the mean filter template is: , in, is the filtering operator; S102: removing the current real scene smoothed image data from the current real scene image data to obtain a high frequency component; The high frequency component is expressed as: , in, H is the high frequency component, I represents the current real scene image data, represents the current real scene smoothed image data, max represents the maximum value function, Indicates that the current real scene smoothed image data is subtracted from the current real scene image data to obtain a removed result, and the high frequency component is obtained by taking the maximum value between the removed result and 0; S103: Obtain the real scene enhanced image data through Hadamard operation according to the high frequency component and the current real scene image data.

[0021] In this embodiment, a 13×13 mean filter template is used to smooth the image. The middle of the filter operator is a 5×5 zero matrix, the outer side is filled with 2 rows and 2 columns, the filling value is -2, and the outermost side is filled with 2 rows and 2 columns, the filling value is -1.

[0022] In step S2, the target detection model includes: S201: extracting a candidate target point area according to the real scene enhanced image data by using an improved density peak search method; S202: Obtaining a positioning target mapping map according to the candidate target point area through a local discriminant model; S203: performing background suppression by a curvature filtering method according to the positioning target mapping image to obtain a background suppression target area; S204: Obtaining a background suppression enhancement target region by performing a Hadamard operation according to the background suppression target region and the high frequency component; S205: Obtaining the reference target image by threshold segmentation according to the background suppression and enhancement target area; The threshold segmentation expression is: , in, Thursday is the reference target image, m is the background mean, i is a constant, s is the background standard deviation.

[0023] Specifically, the improved density peak search method includes: S201-1: extracting the real scene enhanced image data to obtain pixel points, and obtaining the grayscale value of the pixel points by grayscale value calculation according to the pixel points; The gray value calculation expression is: , in, r m is the gray value of the pixel, I(x,y) represents the pixel point; S201-2: Calculating the minimum distance by using the distance between the pixel point and the gray value of the pixel point and the adjacent high-density point; The calculation expression of the distance between adjacent high-density points is: , in, δ(i,j) represents the minimum distance, minutes represents the minimum function, r 1 , r 2 is the gray value of the pixel, d (i,j) Represents the distance to adjacent high-density points. x i represents the horizontal coordinate of the i-th pixel point, x j represents the horizontal coordinate of the jth pixel, y i represents the ordinate of the i-th pixel, y j Represents the ordinate of the jth pixel; S201-3: Obtaining a local density peak joint characteristic factor by multiplying the grayscale value of the pixel point and the minimum distance; The multiplication calculation expression is: , in, oh represents the local density peak joint characteristic factor, r represents the gray value of the pixel, d represents the minimum distance; S201-4: obtaining the local maximum density point by sorting the local density peaks and characteristic factors in descending order; S201-5: constructing a target area to be measured with the local maximum density point as the center, and obtaining a grayscale probability value by calculating the grayscale probability of the target area to be measured; The grayscale probability calculation expression is: , in, R k is the grayscale probability value, M represents the number of rows in the target area to be tested, N represents the number of columns of the target area to be measured, i represents the target area to be tested in the i-th row, j represents the target area to be tested in the jth column, s(i,j) is the gray value, s k is the target pixel value to be measured, logP s(i,j) Indicates the proportion of pixels with gray value s(i,j) in the area; S201-6: Obtain the candidate target point region by sorting in descending order according to the grayscale probability values.

[0024] In this embodiment, the local density peak joint characteristic factor is sorted in descending order, and the first 20 pixel points are taken as the local maximum density point; the target area to be measured is established with the local maximum density point as the center, and the local grayscale probability value of the target area to be measured is calculated, and the local grayscale probability value is sorted in descending order. The larger the local grayscale probability value, the more uniform the grayscale distribution in the area, and the smaller the local grayscale probability value, the more uneven the grayscale distribution in the area. The first 12 pixel points of the local grayscale probability value are selected as the candidate target points.

[0025] Specifically, the local discriminant model includes: S202-1: Obtaining a central sub-block grayscale mean value, a central sub-block gradient value, a sub-block grayscale mean value of each layer, and a sub-block gradient value of each layer according to the candidate target point region; S202-2: Obtain a grayscale mean difference and a gradient difference by subtracting the grayscale mean of the central sub-block, the gradient value of the central sub-block, the grayscale mean of each layer sub-block, and the gradient value of each layer sub-block; The grayscale mean difference expression is: , in, MD Bi represents the grayscale mean difference, M T represents the grayscale mean of the central sub-block, M Bi Represents the grayscale mean of the sub-blocks of each layer; The gradient difference expression is: , in, GD Bi represents the gradient difference, G T represents the central sub-block gradient value, G Bi Represents the gradient value of each layer sub-block; S202-3: Obtaining a diagonal sub-block gradient product by multiplication according to the gradient difference, wherein the diagonal sub-block gradient product carries a direction label; The diagonal sub-block gradient product expression is: , in, FG Bi represents the diagonal sub-block gradient product, GD Bi represents the gradient difference of the i-th sub-block in the B-th layer, GD B(b-i) represents the gradient difference of the bi-th sub-block in the B-th layer, b Indicates the number of sub-blocks in layer B; S202-4: Calculate the mean value of the gradient of each sub-block of each layer by averaging the gradient values ​​of each sub-block; calculate the gradient increment by subtracting the gradient of each sub-block of each layer; and calculate the mean gradient increment by subtracting the mean value of the gradient of each sub-block of each layer; S202-5: Obtain the positioning target mapping image through regional positioning judgment according to the diagonal sub-block gradient product, the grayscale mean difference, the gradient difference, the gradient increment, and the gradient mean increment; The regional positioning judgment includes: Determine whether the diagonal sub-block gradient product of the candidate target point area layer one and the candidate target point area layer two has four different direction labels, if yes, the area is marked as 0; if no, determine whether the grayscale mean difference and the gradient difference are less than or equal to 0, if yes, the area is marked as 0; if no, determine whether the gradient increment is greater than 0, if yes, obtain the positioning target mapping; if no, mark the area as 0.

[0026] In step S3, the preprocessing model includes: S301: Obtaining an aiming point grayscale image and a reference target grayscale image through grayscale image conversion according to the aiming point image and the reference target image; S302: Obtain the grayscale smoothed aiming point image and the grayscale smoothed reference target image through filtering according to the aiming point grayscale image and the reference target grayscale image.

[0027] In step S5, the aiming point calibration model includes: S501: emitting infrared light through a gyroscope to obtain emission time, emission direction and emission acceleration; S502: Matching the grayscale smoothed aiming point image feature and the grayscale smoothed reference target image feature by a brute force matching algorithm according to the launch time to obtain a matching result; S503: Filter the emission direction according to the matching result to obtain the best aiming position.

[0028] In this embodiment, the steps for implementing the augmented reality game of the aiming point automatic calibration method based on machine learning are as follows: The augmented reality game device includes a server, a smart phone, a game gun, and wearable equipment. The game gun is provided with a mobile phone holder, an infrared transmitting tube, an infrared receiving tube, and a laser transmitting tube; the wearable equipment includes a headband, a chest armor, and leg armor, and the headband, chest armor, and leg armor are all provided with an infrared receiving tube; the smart phone is wirelessly connected to the game gun and the wearable equipment; the wearable equipment senses the infrared signal of the outgoing transmitting tube through the infrared receiving tube, and transmits it to the smart phone through Bluetooth; the smart phone is wirelessly connected to the server; the infrared signal is transmitted to the server, and the server automatically calibrates the infrared signal through the aiming point automatic calibration method based on machine learning.

[0029] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0030] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0031] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0032] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A method for automatic calibration of aiming point based on machine learning, characterized in that: The following steps are involved: Acquire current real scene image data through a camera, and obtain real scene enhanced image data through a picture enhancement model according to the current real scene image data; Obtaining a reference target image through a target detection model according to the real scene enhanced image data; Acquire an aiming point image by emitting infrared light from a muzzle, and obtain a grayscale smoothed aiming point image and a grayscale smoothed reference target image through a preprocessing model according to the aiming point image and the reference target image; Extracting features of the grayscale smoothed aiming point image and the grayscale smoothed reference target image by using a hybrid convolutional neural network to obtain grayscale smoothed aiming point image features and grayscale smoothed reference target image features; The aiming point is automatically calibrated according to the grayscale smoothed aiming point image features and the grayscale smoothed reference target image features through an aiming point calibration model.

2. The aiming point automatic calibration method based on machine learning according to claim 1, characterized in that: The obtaining of real scene enhanced image data by using a picture enhancement model according to the current real scene image data comprises: Obtaining current real scene smoothed image data through a mean filtering template according to the current real scene image data; Removing the current real scene smoothed image data from the current real scene image data to obtain a high frequency component; The real scene enhanced image data is obtained by performing a Hadamard operation according to the high frequency component and the current real scene image data.

3. The aiming point automatic calibration method based on machine learning according to claim 1, characterized in that: The obtaining of a reference target image through a target detection model according to the real scene enhanced image data comprises: Extracting a candidate target point area based on the real scene enhanced image data by using an improved density peak search method; Obtaining a positioning target mapping diagram based on the candidate target point area through a local discriminant model; Performing background suppression by a curvature filtering method according to the positioning target mapping image to obtain a background suppression target area; Obtaining a background suppression enhancement target area through Hadamard operation according to the background suppression target area and the high frequency component; The reference target image is obtained by threshold segmentation according to the background suppression and enhancement target area.

4. The aiming point automatic calibration method based on machine learning according to claim 3, characterized in that: The extracting of the candidate target point area by improving the density peak search method according to the real scene enhanced image data comprises: Extracting the real scene enhanced image data to obtain pixel points, and obtaining the pixel point grayscale value by grayscale value calculation according to the pixel points; The minimum distance is obtained by calculating the distance between the pixel point and the gray value of the pixel point through the adjacent high-density point; Obtaining a local density peak joint characteristic factor by multiplying the pixel gray value and the minimum distance; According to the local density peak combined with the characteristic factors, the local density maximum point is obtained by sorting in descending order; Constructing a target area to be measured with the local maximum density point as the center, and obtaining a grayscale probability value by calculating the grayscale probability of the target area to be measured; The candidate target point region is obtained by sorting the grayscale probability values ​​in descending order.

5. The aiming point automatic calibration method based on machine learning according to claim 3, characterized in that: The obtaining of a positioning target mapping diagram according to the candidate target point area through a local discriminant model comprises: Obtain the grayscale mean value of the central sub-block, the gradient value of the central sub-block, the grayscale mean value of each layer of sub-blocks, and the gradient value of each layer of sub-blocks; A grayscale mean difference and a gradient difference are calculated by subtracting the grayscale mean of the central sub-block, the gradient value of the central sub-block, the grayscale mean of each layer sub-block, and the gradient value of each layer sub-block; Obtaining a diagonal sub-block gradient product by multiplication according to the gradient difference, wherein the diagonal sub-block gradient product carries a direction label; The mean value of the gradient of each sub-block of each layer is calculated by averaging the gradient values ​​of each sub-block of each layer; the gradient increment is calculated by subtracting the gradient of each sub-block of each layer; the mean value of the gradient increment is calculated by subtracting the mean value of the gradient of each sub-block of each layer; Obtaining the positioning target mapping image through regional positioning judgment according to the diagonal sub-block gradient product, the grayscale mean difference, the gradient difference, the gradient increment, and the gradient mean increment; The regional positioning judgment includes: Determine whether the diagonal sub-block gradient product of the candidate target point area layer one and the candidate target point area layer two has four different direction labels, if yes, the area is marked as 0; if no, determine whether the grayscale mean difference and the gradient difference are less than or equal to 0, if yes, the area is marked as 0; if no, determine whether the gradient increment is greater than 0, if yes, obtain the positioning target mapping; if no, mark the area as 0.

6. The aiming point automatic calibration method based on machine learning according to claim 1, characterized in that: The step of obtaining a grayscale smoothed aiming point image and a grayscale smoothed reference target image through a preprocessing model according to the aiming point image and the reference target image comprises: Obtaining an aiming point grayscale image and a reference target grayscale image by grayscale image conversion according to the aiming point image and the reference target image; The grayscale smoothed aiming point image and the grayscale smoothed reference target image are obtained by filtering the aiming point grayscale image and the reference target grayscale image.

7. The aiming point automatic calibration method based on machine learning according to claim 1, characterized in that: The automatic calibration of the aiming point by using the aiming point calibration model according to the grayscale smooth aiming point image feature and the grayscale smooth reference target image feature comprises: The gyroscope emits infrared to obtain the emission time, emission direction and emission acceleration; According to the launch time, the grayscale smooth aiming point image feature and the grayscale smooth reference target image feature are matched by a brute force matching algorithm to obtain a matching result; The emission direction is screened according to the matching result to obtain the best aiming position.

8. An aiming point automatic calibration system based on machine learning, comprising an image enhancement module, a target detection module, a preprocessing module, a feature extraction module, and an aiming calibration module, characterized in that: include: The image enhancement module is used to obtain current real scene image data through a camera, and obtain real scene enhanced image data through an image enhancement model according to the current real scene image data; The target detection module is used to obtain a reference target image through a target detection model according to the real scene enhanced image data; The preprocessing module is used to acquire an aiming point image by emitting infrared light from a muzzle, and to obtain a grayscale smoothed aiming point image and a grayscale smoothed reference target image through a preprocessing model according to the aiming point image and the reference target image; The feature extraction module is used to extract features from the grayscale smoothed aiming point image and the grayscale smoothed reference target image through a hybrid convolutional neural network to obtain grayscale smoothed aiming point image features and grayscale smoothed reference target image features; The aiming calibration module is used to automatically calibrate the aiming point through an aiming point calibration model according to the grayscale smoothed aiming point image features and the grayscale smoothed reference target image features.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the aiming point automatic calibration method based on machine learning is implemented as described in any one of claims 1-7.

10. A storage medium containing computer executable instructions, characterized in that: When the computer executable instructions are executed by a computer processor, they are used to perform the aiming point automatic calibration method based on machine learning as described in any one of claims 1 to 7.

Images