An antiaircraft gun zero flight error detection method, device, equipment and storage medium

By acquiring reference images under ideal conditions and actual environmental data, and utilizing a multi-source error prediction model and error correction algorithm, the problem of insufficient accuracy in detecting zero-flight errors of anti-aircraft guns under environmental interference was solved, achieving high-precision detection in non-ideal environments.

CN121498634BActive Publication Date: 2026-03-27HEBEI JUNTAO TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610036748.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-01-13
Publication Date
2026-03-27
Estimated Expiration
2046-01-13

AI Technical Summary

Technical Problem

Existing anti-aircraft gun zero-flight error detection technology is not accurate enough under the interference of environmental factors, and it is difficult to balance environmental adaptability and detection accuracy.

Method used

By acquiring reference images under ideal conditions and actual environmental data, error prediction and correction are performed using muzzle shock wave intensity, optical visibility, illumination change rate, and camera attitude data. Combined with a multi-source error prediction model and error correction algorithm, accurate correction of the actual image is achieved.

Benefits of technology

Accurate detection of zero-flight error of anti-aircraft guns is achieved under non-ideal environments, improving environmental adaptability and detection accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121498634B_ABST
    Figure CN121498634B_ABST
Patent Text Reader

Abstract

The application provides a high-velocity gun zero-flying error detection method, device, equipment and storage medium, and relates to the technical field of weapon equipment test error correction. The method comprises the following steps: acquiring a target image collected by a camera fixed on a high-velocity gun barrel during a zero-flying test in an ideal environment as a reference image; acquiring a muzzle blast intensity, optical visibility, illumination mutation rate, camera posture data and an actual image collected by the camera during an actual zero-flying test of the high-velocity gun; performing error prediction on the reference image according to the muzzle blast intensity, optical visibility, illumination mutation rate and camera posture data, and obtaining a prediction error of the actual image relative to the reference image; performing error correction on the actual image according to the prediction error, and performing zero-flying error detection according to the corrected actual image. The application can realize accurate detection of zero-flying errors in different zero-flying test environments and improve environmental adaptability.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of weapon equipment test error correction, and in particular to a high-altitude gun zero flight error detection method, device, equipment and storage medium. BACKGROUND

[0002] High-altitude gun zero flight error detection is a key link to ensure the shooting accuracy of high-altitude guns. The core is to collect the gun muzzle and projectile images through a camera, analyze the deviation of the projectile trajectory from the reference trajectory, and realize accurate measurement of zero flight error.

[0003] However, in actual detection scenarios, environmental factors (such as sudden changes in light, smoke, low / high temperature extreme weather), mechanical jitter generated by the camera follow-up system and other interference factors will cause the collected images to have pixel offset, gray distortion, blur and other problems, which seriously affect the zero flight error detection accuracy.

[0004] Currently, zero flight testing is usually performed in an ideal test environment to reduce the impact of the environment on zero flight error detection, but this approach limits the implementation of zero flight testing and does not improve environmental adaptability. SUMMARY

[0005] Embodiments of the present application provide a high-altitude gun zero flight error detection method, device, equipment and storage medium to solve the problem that traditional zero flight error detection processes cannot balance environmental adaptability and detection accuracy.

[0006] In a first aspect, embodiments of the present application provide a high-altitude gun zero flight error detection method, comprising:

[0007] Obtain a reference image, which is a target image collected by a camera fixed on a high-altitude gun barrel during zero flight testing in an ideal environment;

[0008] Obtain the muzzle blast intensity, optical visibility, light mutation rate, camera attitude data and actual images collected by the camera during actual zero flight testing of the high-altitude gun;

[0009] Predict errors in the reference image according to the muzzle blast intensity, optical visibility, light mutation rate and camera attitude data to obtain the predicted error of the actual image relative to the reference image;

[0010] Correct errors in the actual image according to the predicted error, and perform zero flight error detection based on the corrected actual image.

[0011] In a possible implementation, the error prediction of the reference image according to the muzzle blast intensity, the optical visibility, the illumination abrupt change rate and the camera pose data comprises:

[0012] According to the muzzle blast intensity, a first error prediction result and a first error prediction confidence of the reference image are obtained.

[0013] According to the optical visibility, a second error prediction result and a second error prediction confidence of the reference image are obtained.

[0014] According to the illumination abrupt change rate, a third error prediction result and a third error prediction confidence of the reference image are obtained.

[0015] According to the camera pose data, a fourth error prediction result and a fourth error prediction confidence of the reference image are obtained.

[0016] The first error prediction confidence, the second error prediction confidence, the third error prediction confidence and the fourth error prediction confidence are normalized to obtain a normalized first confidence, a normalized second confidence, a normalized third confidence and a normalized fourth confidence.

[0017] The same type of error prediction result in the first error prediction result, the second error prediction result, the third error prediction result and the fourth error prediction result is fused according to the normalized first confidence, the normalized second confidence, the normalized third confidence and the normalized fourth confidence, and the prediction error of the actual image relative to the reference image is obtained according to the fusion result.

[0018] In a possible implementation, the first error prediction result and the first error prediction confidence of the reference image according to the muzzle blast intensity comprise:

[0019] According to the muzzle blast intensity and a preset "muzzle blast intensity-pixel offset" mapping model, a first target pixel offset corresponding to the muzzle blast intensity is obtained, and the first target pixel offset is taken as the first error prediction result.

[0020] According to the muzzle blast intensity and a first preset error prediction confidence function, a first error prediction confidence corresponding to the muzzle blast intensity is obtained.

[0021] In a possible implementation, the second error prediction result and the second error prediction confidence of the reference image according to the optical visibility comprise:

[0022] determine a degree of occlusion of the current visibility environment of the antiaircraft gun to the reference image according to the optical visibility, and take the degree of occlusion as a second error prediction result;

[0023] obtain a second error prediction confidence corresponding to the optical visibility according to the optical visibility and a second preset error prediction confidence function.

[0024] In a possible implementation, the third error prediction result and the third error prediction confidence of the reference image are obtained according to the illumination mutation rate, including:

[0025] obtain a target image gray offset corresponding to the illumination mutation rate according to the illumination mutation rate and a preset "illumination mutation-image gray offset" mapping model, and take the target image gray offset as the third error prediction result;

[0026] obtain a third error prediction confidence corresponding to the illumination mutation rate according to the illumination mutation rate and a third preset error prediction confidence function.

[0027] In a possible implementation, the fourth error prediction result and the fourth error prediction confidence of the reference image are obtained according to the camera pose data, including:

[0028] obtain a second target pixel offset corresponding to the camera pose data according to the camera pose data and a preset "pose angle change-pixel offset" mapping model, and take the second target pixel offset as the fourth error prediction result;

[0029] obtain a fourth error prediction confidence corresponding to the camera pose data according to the camera pose data and a fourth preset error prediction confidence function.

[0030] In a possible implementation, the prediction error of the actual image relative to the reference image is obtained by performing error prediction on the reference image according to the muzzle blast intensity, the optical visibility, the illumination mutation rate and the camera pose data, including:

[0031] input the muzzle blast intensity, the optical visibility, the illumination mutation rate and the camera pose data into a multi-source error prediction model to obtain the prediction error of the current actual image of the antiaircraft gun relative to the reference image;

[0032] The multi-source error prediction model is trained by image error data under different environment combinations, and the different environment combinations are combinations of different muzzle blast intensities, optical visibilities, illumination mutation rates and camera pose data.

[0033] In a second aspect, an embodiment of the present application provides a high-velocity gun zero-flying error detection device, comprising:

[0034] A first obtaining module is configured to obtain a reference image, which is a target image in a zero-flying test process collected by a camera fixed on a barrel of a high-velocity gun in an ideal environment;

[0035] A second obtaining module is configured to obtain a muzzle blast intensity, an optical visibility, a light change rate, camera posture data and an actual image collected by the camera in an actual zero-flying test of the high-velocity gun;

[0036] An error prediction module is configured to predict errors of the reference image according to the muzzle blast intensity, the optical visibility, the light change rate and the camera posture data, and obtain a predicted error of the actual image relative to the reference image;

[0037] An error correction module is configured to correct errors of the actual image according to the predicted error, and perform zero-flying error detection according to the corrected actual image.

[0038] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the method in the first aspect or any possible implementation manner of the first aspect when executing the computer program.

[0039] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method in the first aspect or any possible implementation manner of the first aspect.

[0040] In the embodiment of the present application, the target image in the zero-flying test process collected by the camera fixed on the barrel of the high-velocity gun in the ideal environment is obtained as the reference image, and the muzzle blast intensity, the optical visibility, the light change rate, the camera posture data and the actual image collected by the camera in the actual zero-flying test of the high-velocity gun are obtained, so that the reference image can be predicted in a targeted manner according to the muzzle blast intensity, the optical visibility, the light change rate and the camera posture data from the current environment of the zero-flying error detection, thereby obtaining an accurate predicted error of the actual image of the high-velocity gun relative to the reference image, and then correcting errors of the actual image according to the predicted error and performing zero-flying error detection according to the corrected actual image, so as to realize accurate detection of the zero-flying error in different zero-flying test environments and improve the environmental adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is an implementation flowchart of the high-velocity gun zero-flying error detection method provided by the embodiment of the present application;

[0042] Figure 2 Fig. 1 is a structural schematic diagram of the high-velocity gun zero-flying error detection device provided by the embodiment of the present application;

[0043] Figure 3 Fig. 2 is a schematic diagram of the electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0044] The embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0045] Reference is made to Figure 1 Fig. 3 shows an implementation flowchart of the high-velocity gun zero-flying error detection method provided by the embodiment of the present application, which is described in detail as follows:

[0046] Step 101, a reference image is acquired, which is a target image in a zero-flying test process collected by a camera fixed on a high-velocity gun barrel in an ideal environment.

[0047] For example, the ideal environment can be an environment without smoke, vibration and unstable light, such as a weather with clear sky, breeze, excellent visibility and appropriate temperature and humidity. In the embodiment, it is considered that the ideal environment is not always available when the high-velocity gun zero-flying error detection is needed each time. Therefore, when the actual environment is the ideal environment, the target image in the zero-flying test process collected by the camera fixed on the high-velocity gun barrel at this time is acquired as the reference image. When the high-velocity gun zero-flying error detection is needed later, if the environment is not the ideal environment, the reference image acquired before is corrected, so as to meet the requirement of zero-flying error detection in a non-ideal environment and ensure the accuracy of the zero-flying error detection.

[0048] In an embodiment, it is considered that the deviation of the reference image from the actual scene gradually increases due to factors such as device aging (e.g., camera lens wear and tear, sensor drift), environmental benchmark change (e.g., terrain change of the high-velocity gun location), etc. Therefore, the reference image can be periodically calibrated and updated or updated online.

[0049] The periodic calibration and update can set a “calibration period” (e.g., every 100 shots or every week), and when the calibration period is reached, the target image in the zero-flying test process is collected again in the ideal environment as the reference image. The online adaptive update can analyze the feature stability of continuous multiple images by using an “unsupervised learning algorithm” (e.g., K-means clustering), and when it is found that the error of the corrected image continuously exceeds a threshold value (e.g., the zero-flying error detection variance increases), the reference image is automatically fine-tuned without manual intervention.

[0050] Step 102, the muzzle blast intensity, optical visibility, light mutation rate, camera posture data and actual image collected by the camera in the actual zero-flying test of the high-velocity gun are acquired.

[0051] In this embodiment, considering that the high-velocity gun zero flight error detection faces a unique strong interference environment, such as muzzle blast, smoke, sudden change of light, extreme weather of low temperature or high temperature, sand, etc., therefore, the current environment data such as the intensity of the muzzle blast, the optical visibility, the sudden change rate of light, the camera attitude data during the actual zero flight test of the high-velocity gun are considered to predict the error of the reference image in the ideal environment, so as to correct the actual image collected by the camera during the actual zero flight test of the high-velocity gun by using the predicted error.

[0052] For example, the intensity of the muzzle blast can be collected by deploying a pressure sensor on the barrel of the high-velocity gun, the optical visibility can be collected by deploying an optical visibility sensor on the barrel of the high-velocity gun, the sudden change rate of light can be collected by deploying a light sensor on the barrel of the high-velocity gun, and the camera attitude data can be collected by deploying an inertial measurement unit on the barrel of the high-velocity gun.

[0053] In addition, in order to more accurately predict the error in the future, a temperature and humidity sensor can also be deployed in the environment of the high-velocity gun zero flight error detection to collect the current temperature and humidity, and an air pressure sensor can also be deployed to collect the current air pressure, so that the temperature and humidity and the air pressure are combined with the intensity of the muzzle blast, the optical visibility, the sudden change rate of light and the camera attitude data for subsequent error prediction.

[0054] For example, the camera for collecting the actual image during the actual zero flight test of the high-velocity gun can be the camera fixed on the barrel of the high-velocity gun in step 101, and the fixed position, collection time and collection frequency of the camera are the same as those for collecting the reference image, so as to avoid introducing additional errors.

[0055] Step 103, according to the intensity of the muzzle blast, the optical visibility, the sudden change rate of light and the camera attitude data, the error of the reference image is predicted to obtain the predicted error of the actual image relative to the reference image.

[0056] In this embodiment, considering that the muzzle blast will cause the camera to shake and the image to blur momentarily, the smoke in the test environment will block the target, the sudden change of light such as daytime / nighttime switching and explosion light will cause the image gray scale to shift, and the high-frequency shaking of the camera follow-up system will also cause the camera to shake and the image to blur, therefore, the reference image is comprehensively predicted in combination with the intensity of the muzzle blast, the optical visibility, the sudden change rate of light and the camera attitude data.

[0057] Step 104, according to the predicted error, the error of the actual image is corrected, and the zero flight error detection is performed according to the corrected actual image.

[0058] For example, the actual image can be obtained by subtracting the prediction error from the actual image, and the dynamic tracking error data of the antiaircraft gun zero flight test can be calculated according to the modified actual image and the image collected by the fire control tracker axis position sensor, including the maximum error, the minimum error, the mean square error, the average error, etc., and displayed in the form of error curve and table.

[0059] Specifically, considering that the muzzle shock wave, camera posture change caused jitter and image blur usually introduce pixel offset, optical visibility change usually causes different degrees of occlusion, day / night switching, explosion light and other light mutations usually cause image gray offset, therefore, when subtracting the prediction error from the actual image, the actual image can be subtracted by different types of prediction error. For example, first, calculate the image translation matrix according to the pixel offset in the prediction error, and perform geometric correction on the actual image to subtract the pixel offset prediction error in the actual image. Then, the occlusion degree in the prediction error is used as the fog density coefficient in the defogging algorithm to adjust the image transmittance, so as to remove the blur caused by occlusion in the actual image. Finally, according to the image gray offset in the prediction error, the real gray of the actual image is restored by using the gray stretching algorithm, so as to obtain the modified actual image after subtracting the prediction error.

[0060] The above process will be described in detail below.

[0061] For "calculate the image translation matrix according to the pixel offset in the prediction error, and perform geometric correction on the actual image to subtract the pixel offset prediction error in the actual image", suppose the pixel offset in the prediction error is , wherein is the pixel offset in the horizontal direction of the image, is the pixel offset in the vertical direction of the image, then the image translation matrix , if the actual image is , then the geometric corrected coordinates can be calculated according to the following formula :

[0062] ;

[0063] For "use the occlusion degree in the prediction error as the fog density coefficient in the defogging algorithm to adjust the image transmittance, so as to remove the blur caused by occlusion in the actual image", the dark channel prior defogging algorithm can be used to remove the blur caused by occlusion in the actual image. Wherein, the occlusion degree in the prediction error is as an example to illustrate the specific process:

[0064] First, according to the mapping relationship between the occlusion degree and the fog density coefficient, the occlusion degree corresponding to the fog density coefficient The mapping relationship between the shielding degree and the fog concentration coefficient can be determined according to the optical visibility corresponding to the shielding degree and the maximum value and the minimum value of the possible optical visibility in the actual zero flight test. The mapping relationship between the shielding degree and the fog concentration coefficient will be specifically introduced as follows:

[0065] If the shielding degree is The corresponding optical visibility is , then the shielding degree is The corresponding fog concentration coefficient is , The maximum value of the possible optical visibility in the actual zero flight test can be determined according to the optical visibility corresponding to the minimum shielding degree, for example, the maximum value of the optical visibility is determined as 1.2 times of the optical visibility corresponding to the minimum shielding degree, The minimum value of the possible optical visibility in the actual zero flight test can be determined according to the optical visibility corresponding to the maximum shielding degree, for example, the minimum value of the optical visibility is determined as 0.8 times of the optical visibility corresponding to the maximum shielding degree.

[0066] The determination process of the minimum shielding degree and the maximum shielding degree can be as follows:

[0067] Before determining the shielding degree in the prediction error, the blur degree of the target image in the zero flight test process collected by the camera under different optical visibilities is calibrated and clustered, and the blur degree corresponding to each cluster center is taken as a shielding degree level. Then, the shielding degree representing no shielding or approximate no shielding in the shielding degree level is determined as the minimum shielding degree, and the shielding degree representing complete shielding or approximate complete shielding in the shielding degree level is determined as the maximum shielding degree.

[0068] Then, the transmittance adjustment formula is used to realize dynamic adjustment of the shielding degree to the transmittance, wherein, is the pixel point adjusted transmittance, is the initial transmittance of the pixel point determined by the dark channel prior dehazing algorithm. The calculation formula of the initial transmittance is as follows: , wherein, is a dehazing coefficient, generally taken as 0.95, is the intensity value of the actual image in the color channel , is a local region centered on the pixel point , is an atmospheric light value.

[0069] Finally, the adjusted transmittance obtained based on the transmittance adjustment formula is used to determine the shielding degree in the prediction error according to the mapping relationship between the shielding degree and the fog concentration coefficient. obtain the actual image without the blur caused by the occlusion in the reference image Here, a smaller threshold is introduced Generally Take 0.1 to avoid the denominator being zero or unstable due to the adjusted transmittance being too small, causing noise amplification.

[0070] For “restoring the true gray scale of the actual image according to the gray scale shift in the prediction error of the image using a gray scale stretching algorithm”, assuming that the gray scale shift in the prediction error of the image is The gray scale value of the actual image without the gray scale shift can be obtained by , where is the original gray scale value of the actual image, is the gray scale value of the actual image without the gray scale shift. Then, according to restore the true gray scale value of the actual image , is the lower limit of gray scale stretching, which can be the minimum gray scale value of the reference image, is the upper limit of gray scale stretching, which can be the maximum gray scale value of the reference image.

[0071] In an embodiment, step 103 comprises:

[0072] According to the muzzle blast intensity, obtain the first error prediction result and the first error prediction confidence of the reference image.

[0073] According to the optical visibility, obtain the second error prediction result and the second error prediction confidence of the reference image.

[0074] According to the illumination mutation rate, obtain the third error prediction result and the third error prediction confidence of the reference image.

[0075] According to the camera pose data, obtain the fourth error prediction result and the fourth error prediction confidence of the reference image.

[0076] Normalize the first error prediction confidence, the second error prediction confidence, the third error prediction confidence, and the fourth error prediction confidence to obtain the normalized first confidence, the normalized second confidence, the normalized third confidence, and the normalized fourth confidence.

[0077] Fuse the same type of error prediction results in the first error prediction result, the second error prediction result, the third error prediction result, and the fourth error prediction result according to the normalized first confidence, the normalized second confidence, the normalized third confidence, and the normalized fourth confidence, and obtain the prediction error of the actual image relative to the reference image according to the fusion result.

[0078] ​In this embodiment, by using the muzzle blast intensity, optical visibility, light abrupt change rate, camera pose data to obtain the first error prediction result, the second error prediction result, the third error prediction result, the fourth error prediction result and the confidence thereof, the errors possibly caused by different factors in the current environment can be quantified more accurately.

[0079] For example, according to the muzzle blast intensity, the first error prediction result and the first error prediction confidence of the reference image are obtained, which includes:

[0080] According to the muzzle blast intensity and the preset “blast intensity-pixel offset” mapping model, the first target pixel offset corresponding to the muzzle blast intensity is obtained, and the first target pixel offset is taken as the first error prediction result.

[0081] According to the muzzle blast intensity and the first preset error prediction confidence function, the first error prediction confidence corresponding to the muzzle blast intensity is obtained.

[0082] The preset “blast intensity-pixel offset” mapping model can be calibrated by experiment, that is, the average increment of the horizontal / vertical pixel offset of the camera is determined when the blast intensity increases by 1 MPa, and then a polynomial fitting is performed. The first preset error prediction confidence function can be determined in combination with the mapping result of the preset “blast intensity-pixel offset” mapping model and the error of the actual pixel offset. Alternatively, the first preset error prediction confidence function can also be a segmented function. When the muzzle blast intensity, optical visibility, light abrupt change rate, and camera pose data are all normal, the confidence is evenly distributed among all factors (for example, the confidence corresponding to the muzzle blast intensity, optical visibility, light abrupt change rate, and camera pose data is 0.25), and when the muzzle blast intensity is abnormal (for example, strong blast scene, P≥5 MPa), the first error prediction confidence is enhanced, for example, from 0.25 to 0.4.

[0083] For example, according to the optical visibility, the second error prediction result and the second error prediction confidence of the reference image are obtained, which includes:

[0084] According to the optical visibility, the shielding degree of the current visibility environment of the high-velocity gun to the reference image is determined, and the shielding degree is taken as the second error prediction result.

[0085] According to the optical visibility and the second preset error prediction confidence function, the second error prediction confidence corresponding to the optical visibility is obtained.

[0086] In this embodiment, the blur degree of the target image in the zero flight test process collected by the camera under different optical visibility can be calibrated and clustered, and the blur degree corresponding to each cluster center is taken as a shielding degree level. After the current optical visibility of the antiaircraft gun is collected, the shielding degree corresponding to the current optical visibility of the antiaircraft gun can be selected by calculating the distance between the current optical visibility of the antiaircraft gun and the optical visibility corresponding to each cluster center. The shielding degree level corresponding to the cluster center with the smallest distance is taken as the shielding degree corresponding to the current optical visibility of the antiaircraft gun. Similarly, the second preset error prediction confidence function can also be a segmented function. When the muzzle blast intensity, optical visibility, illumination mutation rate and camera attitude data are all normal, the confidence is evenly distributed among all factors (for example, the confidence corresponding to the muzzle blast intensity, optical visibility, illumination mutation rate and camera attitude data is all taken as 0.25), and in a high dust scene (for example, optical visibility < 2km), the second error prediction confidence is enhanced, for example, from 0.25 to 0.35.

[0087] For example, according to the illumination mutation rate, the third error prediction result and the third error prediction confidence of the reference image are obtained, including:

[0088] According to the illumination mutation rate and the preset "illumination mutation-image gray offset" mapping model, the target image gray offset corresponding to the illumination mutation rate is obtained, and the target image gray offset is taken as the third error prediction result.

[0089] According to the illumination mutation rate and the third preset error prediction confidence function, the third error prediction confidence corresponding to the illumination mutation rate is obtained.

[0090] For example, the illumination intensity can be collected in real time by the illumination sensor, and the illumination mutation rate can be obtained by calculating the illumination intensity change rate of adjacent frames. Similarly, the preset "illumination mutation-image gray offset" mapping model can also be calibrated by experiment, for example, by calibrating the image gray offset under different explosion lights. Different explosion lights can be obtained by different speed projectiles. Similarly, the third preset error prediction confidence function can also be a segmented function. When the muzzle blast intensity, optical visibility, illumination mutation rate and camera attitude data are all normal, the confidence is evenly distributed among all factors (for example, the confidence corresponding to the muzzle blast intensity, optical visibility, illumination mutation rate and camera attitude data is all taken as 0.25), and in a high dust scene (for example, optical visibility < 2km), the second error prediction confidence is enhanced, for example, from 0.25 to 0.35.

[0091] For example, according to the camera attitude data, the fourth error prediction result and the fourth error prediction confidence of the reference image are obtained, including:

[0092] According to the camera pose data and the preset "pose angle change-pixel offset amount" mapping model, a second target pixel offset amount corresponding to the camera pose data is obtained, and the second target pixel offset amount is taken as the fourth error prediction result.

[0093] According to the camera pose data and the fourth preset error prediction confidence function, a fourth error prediction confidence corresponding to the camera pose data is obtained.

[0094] Similarly, the preset "pose angle change-pixel offset amount" mapping model can be obtained through experimental calibration, and the fourth preset error prediction confidence function can also be a segmented function. When the muzzle blast intensity, optical visibility, light mutation rate, and camera pose data are all normal, the confidence among all factors is evenly distributed (for example, the confidence corresponding to the muzzle blast intensity, optical visibility, light mutation rate, and camera pose data is all taken as 0.25), and when the shaking is severe (for example, the pose angle change is greater than 0.5° / s), the fourth error prediction confidence is enhanced, for example, from 0.25 to 0.3.

[0095] For example, after the confidence of a certain factor is enhanced, the error prediction confidences corresponding to various factors can be normalized so that the sum of the error prediction confidences corresponding to various factors is equal to 1.

[0096] For example, assuming that the confidences corresponding to the first error prediction result, the second error prediction result, the third error prediction result, and the fourth error prediction result determined by the segmented function are 0.4, 0.25, 0.25, and 0.25 respectively, after normalization, the normalized confidences corresponding to the first error prediction result, the second error prediction result, the third error prediction result, and the fourth error prediction result are 0.34, 0.22, 0.22, and 0.22 respectively.

[0097] For example, the muzzle blast intensity, optical visibility, light mutation rate, and camera pose data can also be subjected to abnormal data detection respectively, for example, the muzzle blast intensity, optical visibility, light mutation rate, and camera pose data are compared with the corresponding upper and lower threshold values respectively, when a certain data exceeds the upper and lower threshold values, the data is determined as abnormal data, and when the data is determined as abnormal data, the corresponding confidence is set to 0. Taking the muzzle blast intensity as an example, when the muzzle blast intensity exceeds the upper and lower threshold values, the muzzle blast intensity is determined as abnormal data. The upper and lower threshold values of the muzzle blast intensity, the upper and lower threshold values of the optical visibility, the upper and lower threshold values of the light mutation rate, and the upper and lower threshold values of the camera pose data can be determined by the maximum and minimum values of the muzzle blast intensity, optical visibility, light mutation rate, and camera pose data collected historically.

[0098] On this basis, considering that the first error prediction result, i.e., the first target pixel offset, and the fourth error prediction result, i.e., the second target pixel offset, belong to the same type of error prediction result, and the second error prediction result is the occlusion degree, and the third error prediction result is the target image gray offset, which are different types of error prediction results, the first target pixel offset and the second target pixel offset can be fused based on the normalized first confidence and the normalized fourth confidence. For example, the normalized first confidence and the normalized fourth confidence can be normalized again (if the normalized first confidence is , the normalized fourth confidence is , the first confidence after normalization again is , the fourth confidence after normalization again is , assuming that the normalized first confidence is 0.4, the normalized fourth confidence is 0.1, the first confidence after normalization again is 0.8, and the fourth confidence after normalization again is 0.2), the first target pixel offset and the second target pixel offset are weighted and summed according to the first confidence after normalization again and the fourth confidence after normalization again, to obtain a fused target pixel offset. After obtaining the fused target pixel offset, the fused target pixel offset, the occlusion degree, and the target image gray offset can be collectively used as the prediction error of the actual image relative to the reference image.

[0099] In addition, after obtaining the fused target pixel offset, the normalized first confidence and the normalized fourth confidence can be weighted and summed according to the first confidence after normalization again and the fourth confidence after normalization again, to obtain a fused confidence corresponding to the fused target pixel offset according to the contribution of the muzzle blast intensity and the camera pose data. Based on the fused confidence, the normalized second confidence, and the normalized third confidence, an error prediction result with higher reliability can be selected as the final prediction error of the actual image relative to the reference image (for example, if the normalized second confidence is less than a reliability threshold, only the fused target pixel offset and the target image gray offset are used as the prediction error of the actual image relative to the reference image. The reliability threshold is set here to remove prediction errors with extremely low reliability, and therefore, the reliability threshold can be set to a small value, for example, a value in the range of [0.1, 0.2]. Taking the reliability threshold as 0.1 as an example, if the normalized second confidence is less than 0.1, the occlusion degree is removed from the prediction error, and only the fused target pixel offset and the target image gray offset are used as the prediction error of the actual image relative to the reference image).

[0100] In another embodiment, step 103 includes:

[0101] The muzzle blast intensity, optical visibility, light abrupt change rate, and camera posture data are input into the multi-source error prediction model to obtain the prediction error of the actual image of the antiaircraft gun relative to the reference image.

[0102] The multi-source error prediction model is trained by image error data under different environment combinations, and the different environment combinations are combinations of different muzzle blast intensities, optical visibilities, light abrupt change rates, and camera posture data.

[0103] In this embodiment, another error prediction mode based on the muzzle blast intensity, optical visibility, light abrupt change rate, and camera posture data is provided, that is, a machine learning model is trained by collecting image error data under different environment combinations.

[0104] For example, image error data (which can include pixel offset, occlusion degree, and image gray offset) under different environment combinations (which can include muzzle blast intensity, optical visibility, light abrupt change rate, camera posture data, temperature and humidity, air pressure, etc.) can be collected, and a machine learning algorithm (such as a random forest algorithm or a LightGBM algorithm) is used to train the collected image error data to obtain a multi-dimensional environment-image error mapping model, that is, a multi-source error prediction model, so that the model can output corresponding image error parameters (that is, the prediction error of the actual image relative to the reference image) according to the input multi-source data.

[0105] The collected image error data can be obtained by computer simulation, and when the computer simulation is used for labeling, a simulation environment of antiaircraft gun shooting can be constructed, and environment parameters can be simulated to generate simulation images under different environment combinations, and then corresponding image error data such as pixel offset, occlusion degree, and image gray offset can be obtained, and then each set of environment parameters is used as input features, and its corresponding image error data is used as output labels, so as to constitute a labeled data set for training the multi-source error prediction model.

[0106] After obtaining the labeled data set, the labeled data set can be divided into a training set, a validation set, and a test set according to a ratio of 7:2:1, a mean square error can be used as a loss function, and a training termination condition can be that the number of iterations is less than or equal to 500, and the mean square error of the validation set does not decrease (the decrease amplitude is less than 0.0001) for 10 consecutive iterations, or the mean square error of the validation set is less than 0.001.

[0107] The embodiment of the present application obtains the target image in the zero flight test process collected by the camera fixed on the barrel of the high-powered gun in the ideal environment as the reference image, and obtains the muzzle blast intensity, optical visibility, illumination mutation rate and camera posture data of the high-powered gun at present, can make error prediction on the reference image according to the muzzle blast intensity, optical visibility, illumination mutation rate and camera posture data from the current environment of the zero flight error detection, so as to obtain the accurate prediction error of the actual image of the high-powered gun at present relative to the reference image, and then correct the actual image according to the prediction error, and detect the zero flight error according to the corrected actual image, so as to realize the accurate detection of the zero flight error in different zero flight test environments and improve the environmental adaptability.

[0108] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0109] The following is the device embodiment of the present application, and for the details not described in detail, reference can be made to the corresponding method embodiments described above.

[0110] Figure 2 The structure schematic diagram of the high-powered gun zero flight error detection device provided by the embodiment of the present application is shown, only the part related to the embodiment of the present application is shown for the convenience of description, and the details are as follows:

[0111] As Figure 2 shown, the high-powered gun zero flight error detection device comprises:

[0112] The first acquisition module 21 is used for acquiring the reference image, and the reference image is the target image in the zero flight test process collected by the camera fixed on the barrel of the high-powered gun in the ideal environment.

[0113] The second acquisition module 22 is used for acquiring the muzzle blast intensity, optical visibility, illumination mutation rate, camera posture data and actual image collected by the camera when the actual zero flight test of the high-powered gun is performed.

[0114] The error prediction module 23 is used for making error prediction on the reference image according to the muzzle blast intensity, optical visibility, illumination mutation rate and camera posture data, and obtaining the prediction error of the actual image relative to the reference image.

[0115] The error correction module 24 is used for correcting the actual image according to the prediction error, and detecting the zero flight error according to the corrected actual image.

[0116] The embodiment of the present application obtains the target image in the zero flight test process collected by the camera fixed on the high gun barrel in the ideal environment as a reference image, and obtains the muzzle blast intensity, optical visibility, illumination mutation rate, camera posture data and actual image collected by the camera in the actual zero flight test of the high gun, can make error prediction on the reference image according to the muzzle blast intensity, optical visibility, illumination mutation rate and camera posture data from the current environment of the zero flight error detection, so as to obtain the accurate prediction error of the actual image of the high gun relative to the reference image, and then correct the actual image according to the prediction error, and detect the zero flight error according to the corrected actual image, so as to realize the accurate detection of the zero flight error in different zero flight test environments and improve the environmental adaptability.

[0117] In a possible implementation, the error prediction module 23 is specifically configured to:

[0118] According to the muzzle blast intensity, a first error prediction result and a first error prediction confidence of the reference image are obtained.

[0119] According to the optical visibility, a second error prediction result and a second error prediction confidence of the reference image are obtained.

[0120] According to the illumination mutation rate, a third error prediction result and a third error prediction confidence of the reference image are obtained.

[0121] According to the camera posture data, a fourth error prediction result and a fourth error prediction confidence of the reference image are obtained.

[0122] The first error prediction confidence, the second error prediction confidence, the third error prediction confidence and the fourth error prediction confidence are normalized to obtain a normalized first confidence, a normalized second confidence, a normalized third confidence and a normalized fourth confidence.

[0123] The same type of error prediction result in the first error prediction result, the second error prediction result, the third error prediction result and the fourth error prediction result is fused according to the normalized first confidence, the normalized second confidence, the normalized third confidence and the normalized fourth confidence, and the prediction error of the actual image relative to the reference image is obtained according to the fusion result.

[0124] In a possible implementation, the error prediction module 23 is specifically configured to:

[0125] According to the muzzle blast intensity and the preset "blast intensity-pixel offset amount" mapping model, a first target pixel offset amount corresponding to the muzzle blast intensity is obtained, and the first target pixel offset amount is taken as the first error prediction result.

[0126] According to the muzzle blast intensity and a first preset error prediction confidence function, a first error prediction confidence corresponding to the muzzle blast intensity is obtained.

[0127] In a possible implementation, the error prediction module 23 is specifically configured to:

[0128] According to the optical visibility, a shielding degree of the current visibility environment of the antiaircraft gun to the reference image is determined, and the shielding degree is taken as a second error prediction result.

[0129] According to the optical visibility and a second preset error prediction confidence function, a second error prediction confidence corresponding to the optical visibility is obtained.

[0130] In a possible implementation, the error prediction module 23 is specifically configured to:

[0131] According to the light mutation rate and a preset “light mutation-image gray offset” mapping model, a target image gray offset corresponding to the light mutation rate is obtained, and the target image gray offset is taken as a third error prediction result.

[0132] According to the light mutation rate and a third preset error prediction confidence function, a third error prediction confidence corresponding to the light mutation rate is obtained.

[0133] In a possible implementation, the error prediction module 23 is specifically configured to:

[0134] According to the camera pose data and a preset “pose angle change-pixel offset amount” mapping model, a second target pixel offset amount corresponding to the camera pose data is obtained, and the second target pixel offset amount is taken as a fourth error prediction result.

[0135] According to the camera pose data and a fourth preset error prediction confidence function, a fourth error prediction confidence corresponding to the camera pose data is obtained.

[0136] In a possible implementation, the error prediction module 23 is specifically configured to:

[0137] The muzzle blast intensity, the optical visibility, the light mutation rate and the camera pose data are input into a multi-source error prediction model, and a prediction error of a current actual image of the antiaircraft gun relative to the reference image is obtained.

[0138] The multi-source error prediction model is obtained by training image error data under different environment combinations, and the different environment combinations are combinations of different muzzle blast intensities, optical visibilities, light mutation rates and camera pose data.

[0139] Figure 3 is a schematic diagram of an electronic device provided by an embodiment of the present application. Figure 3As shown, the electronic device 3 of the embodiment includes a processor 30 and a memory 31. The memory 31 stores a computer program 32. The processor 30 implements the steps in the above method embodiments when executing the computer program 32. Alternatively, the processor 30 implements the functions of the modules / units in the above apparatus embodiments when executing the computer program 32.

[0140] For example, the computer program 32 can be divided into one or more modules / units, which are stored in the memory 31 and executed by the processor 30 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program 32 in the electronic device 3.

[0141] The electronic device 3 can include, but is not limited to, the processor 30 and the memory 31. Those skilled in the art can understand that the electronic device 3 can include more or fewer components, or combine certain components, or include different components, such as an input / output device, a network access device, a bus, etc. Figure 3 The electronic device 3 is only an example and does not constitute a limitation on the electronic device 3, which can include more or fewer components than those shown, or combine certain components, or include different components, for example, the electronic device 3 can also include an input / output device, a network access device, a bus, etc.

[0142] The processor 30 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0143] The memory 31 can be an internal storage unit of the electronic device 3, such as a hard disk or a memory of the electronic device 3. The memory 31 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. provided on the electronic device 3. Further, the memory 31 can include both the internal storage unit and the external storage device of the electronic device 3. The memory 31 is used to store the computer program 32 and other programs and data required by the electronic device 3. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0144] For the convenience and brevity of description, only the above-mentioned division of each functional module / unit is exemplified, and in actual application, the above-mentioned functions can be completed by different functional modules / units according to needs. The above-mentioned modules / units can be realized in the form of hardware, software or a combination of hardware and software.

[0145] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method in each method embodiment described above is realized.

[0146] The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), electric carrier wave signal, telecommunication signal and software distribution medium, etc.

[0147] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments. If there is no special description and logical conflict, the terms and / or descriptions of different embodiments are consistent and can be mutually referred to, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0148] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for detecting the zero-flight error of anti-aircraft guns, characterized in that, include: Acquire a reference image, which is a target image captured by a camera fixed on the barrel of an anti-aircraft gun under ideal conditions during a zero-flight test. The muzzle shock wave intensity, optical visibility, illumination change rate, camera attitude data, and actual images captured by the camera were obtained during the actual zero-flight test of the anti-aircraft gun. Based on the muzzle shock wave intensity, optical visibility, illumination change rate, and camera attitude data, error prediction is performed on the reference image to obtain the prediction error of the actual image relative to the reference image. The actual image is corrected based on the prediction error, and zero-flight error is detected based on the corrected actual image. The method involves performing error prediction on the reference image based on the muzzle shock wave intensity, optical visibility, illumination change rate, and camera attitude data to obtain the prediction error of the actual image relative to the reference image, including: Based on the intensity of the muzzle shock wave, the first error prediction result and the first error prediction confidence level of the reference image are obtained; Based on the optical visibility, a second error prediction result and a second error prediction confidence level for the reference image are obtained; Based on the illumination change rate, the third error prediction result and the third error prediction confidence level of the reference image are obtained; Based on the camera pose data, the fourth error prediction result and the fourth error prediction confidence level of the reference image are obtained; The first error prediction confidence, the second error prediction confidence, the third error prediction confidence, and the fourth error prediction confidence are normalized to obtain normalized first confidence, normalized second confidence, normalized third confidence, and normalized fourth confidence. Based on the normalized first confidence level, the normalized second confidence level, the normalized third confidence level, and the normalized fourth confidence level, error prediction results of the same type in the first error prediction result, the second error prediction result, the third error prediction result, and the fourth error prediction result are fused, and the prediction error of the actual image relative to the reference image is obtained based on the fusion result.

2. The method for detecting the zero-flight error of anti-aircraft guns according to claim 1, characterized in that, Based on the muzzle shock wave intensity, a first error prediction result and a first error prediction confidence level for the reference image are obtained, including: Based on the muzzle shock wave intensity and the preset "shock wave intensity-pixel offset" mapping model, the first target pixel offset corresponding to the muzzle shock wave intensity is obtained, and the first target pixel offset is used as the first error prediction result; The first error prediction confidence level corresponding to the muzzle shock wave intensity is obtained based on the muzzle shock wave intensity and the first preset error prediction confidence function.

3. The method for detecting the zero-flight error of anti-aircraft guns according to claim 1, characterized in that, Based on the optical visibility, a second error prediction result and a second error prediction confidence level for the reference image are obtained, including: Based on the optical visibility, determine the degree of occlusion of the reference image by the current visibility environment of the anti-aircraft gun, and use the degree of occlusion as the second error prediction result; The second error prediction confidence level corresponding to the optical visibility is obtained based on the optical visibility and the second preset error prediction confidence level function.

4. The method for detecting the zero-flight error of anti-aircraft guns according to claim 1, characterized in that, Based on the illumination abrupt change rate, the third error prediction result and the third error prediction confidence level of the reference image are obtained, including: Based on the illumination change rate and the preset "illumination change-image grayscale shift" mapping model, the target image grayscale shift corresponding to the illumination change rate is obtained, and the target image grayscale shift is used as the third error prediction result; Based on the light change rate and the third preset error prediction confidence function, the third error prediction confidence level corresponding to the light change rate is obtained.

5. The method for detecting the zero-flight error of anti-aircraft guns according to claim 1, characterized in that, Based on the camera pose data, the fourth error prediction result and the fourth error prediction confidence level of the reference image are obtained, including: Based on the camera pose data and the preset "pose angle change - pixel offset" mapping model, the second target pixel offset corresponding to the camera pose data is obtained, and the second target pixel offset is used as the fourth error prediction result; Based on the camera pose data and the fourth preset error prediction confidence function, the fourth error prediction confidence level corresponding to the camera pose data is obtained.

6. The method for detecting the zero-flight error of anti-aircraft guns according to claim 1, characterized in that, Error prediction is performed on the reference image based on the muzzle shock wave intensity, optical visibility, illumination change rate, and camera attitude data to obtain the prediction error of the actual image relative to the reference image, including: The muzzle shock wave intensity, optical visibility, illumination change rate, and camera attitude data are input into a multi-source error prediction model to obtain the prediction error of the actual image relative to the reference image. The multi-source error prediction model is trained using image error data under different environmental combinations, which are combinations of different muzzle shock wave intensities, optical visibility, illumination change rates, and camera attitude data.

7. A device for detecting the zero-flight error of anti-aircraft guns, characterized in that, include: The first acquisition module is used to acquire a reference image, which is a target image captured by a camera fixed on the barrel of an anti-aircraft gun under ideal conditions during a zero-flight test. The second acquisition module is used to acquire the muzzle shock wave intensity, optical visibility, illumination change rate, camera attitude data, and actual images captured by the camera during the actual zero-flight test of the anti-aircraft gun. An error prediction module is used to perform error prediction on the reference image based on the muzzle shock wave intensity, the optical visibility, the illumination change rate, and the camera attitude data, so as to obtain the prediction error of the actual image relative to the reference image. An error correction module is used to correct the actual image based on the prediction error, and to detect zero-flight errors based on the corrected actual image. The error prediction module is specifically used for: Based on the intensity of the muzzle shock wave, the first error prediction result and the first error prediction confidence level of the reference image are obtained; Based on the optical visibility, a second error prediction result and a second error prediction confidence level for the reference image are obtained; Based on the illumination change rate, the third error prediction result and the third error prediction confidence level of the reference image are obtained; Based on the camera pose data, the fourth error prediction result and the fourth error prediction confidence level of the reference image are obtained; The first error prediction confidence, the second error prediction confidence, the third error prediction confidence, and the fourth error prediction confidence are normalized to obtain normalized first confidence, normalized second confidence, normalized third confidence, and normalized fourth confidence. Based on the normalized first confidence level, the normalized second confidence level, the normalized third confidence level, and the normalized fourth confidence level, error prediction results of the same type in the first error prediction result, the second error prediction result, the third error prediction result, and the fourth error prediction result are fused, and the prediction error of the actual image relative to the reference image is obtained based on the fusion result.

8. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.