Intelligent automatic firearm calibration method and system based on three-dimensional modeling and laser measurement

Abstract

The invention discloses an intelligent automatic firearm calibration method and system based on three-dimensional modeling and laser measurement. The intelligent automatic firearm calibration method comprises the following steps that S1, point cloud data and two-dimensional image data of key parts of firearms are collected; s2, performing feature extraction and spatial relationship analysis on the original measurement data set; s3, comparing the three-dimensional structure model with a standard three-dimensional model in a historical database; s4, performing spatial feature extraction on the position information by using a convolutional layer; s5, if the current calibration deviation value is not in the preset standard range, submitting the high-dimensional position feature vector to a cross-level feature interaction unit; and S6, when the calibration deviation value is within a preset standard range, adjustment is stopped. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement is realized by combining a convolutional self-attention network, laser high-precision measurement and a three-dimensional point cloud processing technology.

Description

Technical Field

[0001] This invention relates to the field of intelligent firearm calibration technology, and in particular to an intelligent automatic firearm calibration method and system based on three-dimensional modeling and laser measurement. Background Technology

[0002] With the continuous improvement of the automation level of military equipment, the precision of maintenance and use of firearms and other precision weapons places higher demands on combat effectiveness. Traditional firearm calibration methods mostly rely on manual operation, using visual comparison of the positions of the sights, front sight, and muzzle, supplemented by mechanical adjustments to achieve the adjustment of the geometric relationships of various parts of the firearm. This manual calibration method requires operators with rich experience and superb skills in practical applications, but its operation process inevitably suffers from strong subjectivity, poor repeatability, low efficiency, and limited accuracy. Furthermore, in rapid deployment and high-intensity use environments, manual calibration is difficult to meet the demands of high-frequency, batch, and high-precision use. In addition, the manual calibration process is difficult to standardize and trace, leading to significant differences in calibration results between different operators, affecting the consistency of the overall performance of the weapon system and its operational reliability.

[0003] In recent years, with the rapid development of technologies such as 3D imaging, laser measurement, and automatic control, more and more automatic calibration systems have been applied to the inspection and maintenance of weaponry. For example, some systems use coordinate measuring machines to measure the spatial coordinates of key parts of firearms, and then use motor-driven mechanical adjustment devices to make quantitative displacement adjustments to the sights or rear sights. These semi-automatic or automatic calibration systems have improved the accuracy and efficiency of calibration to some extent, but several technical bottlenecks still exist in practical applications. First, most existing systems can only acquire geometric data from a single perspective, and cannot accurately reproduce the complex spatial structural relationships of key firearm components, resulting in a certain deviation between the model and the actual situation, making it difficult to achieve adaptive optimization of high-order spatial relationships. Second, existing systems mostly use traditional geometric fitting and spatial registration algorithms, lacking in-depth mining and utilization of complementary information between different spatial scales and complex structures. Especially when dealing with batch firearms or complex structures in special industrial environments, registration errors are prone to accumulate or the model's generalization ability is insufficient, affecting calibration quality and equipment versatility.

[0004] Furthermore, while laser-based automatic calibration schemes have been increasingly applied in industrial inspection in recent years, they still face numerous limitations in firearm calibration. For instance, relying solely on laser measurement to obtain the spatial coordinates of key components results in significant data noise and limited point cloud density and accuracy, especially under conditions of ambient light interference, varying surface reflection characteristics, and complex structures of the measured parts. Traditional laser measurement equipment often processes data only at the level of spatial straightness analysis or local geometric reconstruction, making it difficult to achieve detailed modeling and global feature optimization of the overall spatial structure of the firearm. Although some existing systems incorporate high-precision laser 3D scanners, they lack intelligent and automated methods for data fusion, multi-view registration, and error iterative optimization, relying on manual intervention and offline analysis, and failing to achieve efficient, closed-loop adaptive adjustment during automatic adjustment.

[0005] Therefore, how to provide an intelligent automatic firearm calibration method based on 3D modeling and laser measurement is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent automatic firearm calibration method and system based on 3D modeling and laser measurement. This invention combines convolutional self-attention networks, high-precision laser measurement, and 3D point cloud processing technology. Through multi-source spatial data fusion and depth feature extraction, it realizes an intelligent automatic firearm calibration method based on 3D modeling and laser measurement. The system uses high-density 3D point clouds and multi-view images to spatially model key parts of the firearm, and utilizes a laser measurement unit to acquire real-time structural parameters. Combined with a convolutional self-attention network, it performs multi-scale fusion and dynamic weight allocation of high-dimensional spatial features, achieving high-precision analysis and closed-loop optimization adjustment of calibration residuals. This method not only improves the automation and intelligence level of calibration but also possesses strong generalization ability and real-time adaptability. It can effectively meet the high-precision, high-efficiency automatic calibration requirements of complex equipment under varying operating conditions, significantly improving the level of firearm maintenance and operational reliability.

[0007] The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to an embodiment of the present invention includes the following steps: S1. Collect point cloud data and two-dimensional image data of key parts of the firearm to generate the original measurement dataset; S2. Perform feature extraction and spatial relationship analysis on the original measurement dataset to establish a three-dimensional structural model including the gun's rear sight, front sight, and muzzle, and obtain the spatial coordinate information of each key component; S3. Using the spatial coordinate information of each key component, the three-dimensional structural model is compared with the standard three-dimensional model in the historical database to analyze the spatial coordinate deviation. Based on the analysis results, the rear sight and front sight are automatically adjusted by a precision robotic arm, and the adjustment position information is collected in real time during the adjustment process. S4. Input the real-time adjustment position information into the high-precision laser measurement unit, use a convolutional self-attention network, use the convolutional layer to extract spatial features of the position information, output the current calibration deviation, and determine whether it is within the preset standard range. S5. If the current calibration deviation is not within the preset standard range, the high-dimensional position feature vector is submitted to the cross-level feature interaction unit. By dynamically associating different spatial resolutions with multi-dimensional context information, the calibration error is analyzed in depth and finely adjusted. S6. When the calibration deviation is within the preset standard range, stop the adjustment and store the actual spatial coordinate information, calibration deviation, and adjustment parameters in the historical database.

[0008] Optionally, the original measurement dataset specifically includes high-density three-dimensional point cloud coordinate data of the gun's rear sight, front sight, and muzzle, multi-view two-dimensional images of the corresponding parts, reflection intensity values ​​of each point cloud, spatial pose parameters of the calibration plate, and ambient lighting information during acquisition.

[0009] Optionally, the real-time position adjustment information specifically includes the spatial displacement, rotation angle, adjustment direction, adjustment speed, and current spatial coordinate changes of the rear sight and front sight recorded by the precision robotic arm during the adjustment of the rear sight and front sight.

[0010] Optionally, S2 specifically includes: S21. Preprocess the original measurement dataset and output the preprocessed point cloud data and two-dimensional image data; S22. Segment the purified point cloud data, extract the point cloud regions of the rear sight, front sight and muzzle, and output the point cloud subsets of each key part. S23. Extract features from the point cloud subsets of each key part to obtain the spatial geometric feature parameters of each key part, and output the feature set of key parts containing spatial location and structural features. S24. Combine two-dimensional image data to perform fusion processing on the feature sets of key parts, enhance the spatial feature expression capability, and output the fused spatial feature data. S25. Based on the fused spatial feature data, establish a three-dimensional structural model including the rear sight, front sight, and muzzle, and output complete three-dimensional structural model data. S26. Perform spatial analysis on the complete three-dimensional structural model data to obtain the spatial coordinate information of the rear sight, front sight, and muzzle.

[0011] Optionally, S3 specifically includes: S31. Obtain the spatial coordinate information of key components of the three-dimensional structural model after spatial filtering optimization, and output the optimized current spatial coordinate data; S32. Based on the standard 3D models under different working conditions in the historical database, the multi-model fusion algorithm is used to extract the spatial coordinates of standard key components and output the fused standard spatial coordinate data. S33. Perform adaptive weighted registration on the optimized current spatial coordinate data and the fused standard spatial coordinate data, and output the adaptively registered spatial coordinate comparison data. S34. Perform fitting analysis on the adaptive registration spatial coordinate comparison data, predict the spatial coordinate deviation trend of each key component, and output the spatial coordinate deviation trend data. S35. Based on the spatial coordinate deviation trend data, generate a dynamic adjustment path, control the precision robotic arm to make adaptive step-by-step adjustments to the gate and the crosshair, record the adjustment path information in real time, and output the dynamic adjustment process data. S36. During the adaptive step-by-step adjustment process, the real-time adjustment position information is intelligently calibrated by combining sensor feedback and historical adjustment data, and the calibrated real-time adjustment position information is output.

[0012] Optionally, S4 specifically includes: S41. Collect and organize the real-time adjusted position information after calibration, and output standardized position information data; S42. Input the standardized location information data into the high-precision laser measurement unit, perform data fusion processing in combination with the laser measurement data, and output the fused location information data. S43. Spatial grid encoding is performed on the fused location information data. By mapping the original location information to a multi-dimensional spatial grid, the statistical features of each grid cell are extracted and combined with the surrounding grid information for feature enhancement, and high-dimensional location feature data containing spatial structure relationships are output. S44. The processed high-dimensional location feature data is input into a convolutional self-attention network. Spatial features under different receptive fields are extracted through a multi-layer convolutional structure. A multi-scale feature fusion layer is introduced, and then the multi-scale features output from each convolutional layer are fed into the multi-scale feature fusion layer. The features at different scales are weighted and fused through a learnable dynamic weight allocation mechanism. The fused features are then used to reconstruct spatial relationships. Finally, the reconstructed multi-scale fused features are input into the self-attention mechanism layer, and a high-dimensional spatial feature vector is output. ; in, For the first Output high-dimensional positional feature vectors For the first The first scale Each input location feature vector For the first Scale fusion weighting coefficients, For the first The first scale Spatial relationships can be learned weights. For the first The normalization factor for each output, where S is the total number of scales for the multi-scale features, and N is the number of input location features. The output feature index has a value range of 100. , The input feature index has a value range of 100. , This is a scale index, with a value range of [value range missing]. ; S45. Based on high-dimensional space feature vectors, feature weight allocation and correlation analysis are performed through a self-attention mechanism to output the current calibration deviation data. S46. Compare the current calibration deviation data with the preset standard range and output the calibration status judgment result.

[0013] Optionally, S5 specifically includes: S51. If the current calibration deviation is not within the preset standard range, the current high-dimensional position feature vector is input to the cross-level feature interaction unit to obtain preliminary fused feature information. S52. Perform multi-scale convolution operations on the preliminary fused feature information to extract context features at different spatial resolutions, and then splice and integrate the context features obtained at each scale to form a comprehensive feature expression containing multi-level spatial relationships and context information. S53. Based on multi-level contextual feature representation, spatial correlation weights are dynamically constructed and updated in association with the preliminary fused feature information to generate error analysis feature representations. S54. Using error analysis feature representation, fine-tune the high-dimensional position feature vector and output the optimized high-dimensional position feature vector. S55. The optimized high-dimensional position feature vector is fed back to the calibration unit. Based on the optimized high-dimensional position feature vector, combined with the currently acquired sensor raw data, historical calibration data and current calibration deviation, multiple rounds of feature comparison and deviation residual analysis are used to further refine the calibration results and update the error compensation mechanism of the calibration unit in real time.

[0014] Optionally, S6 specifically includes: S61. When the calibration deviation is within the preset standard range, stop adjusting the high-dimensional position feature vector and collect the current actual spatial coordinate information. S63. Integrate the current actual spatial coordinate information with the current calibration deviation to generate the current calibration status data; S64. Collect all adjustment parameters used in the current calibration process and associate them with the current calibration status data to form calibration record entries; S65. Format the calibration record entries according to the historical database data structure to ensure that they can be correctly identified and retrieved by the database. S66. Write the formatted calibration record entries into the historical database.

[0015] The intelligent automatic firearm calibration system based on three-dimensional modeling and laser measurement according to an embodiment of the present invention includes the following modules: The 3D data acquisition and standardization module is used to acquire point cloud data and multi-view 2D image data of key parts of firearms. The spatial structure modeling and feature analysis module is used for extracting spatial features of key parts, analyzing spatial relationships, and modeling three-dimensional structures. The automatic adjustment module is used to compare the 3D structural model with the standard 3D model in the historical database, and to automatically adjust the rear sight and front sight through a precision robotic arm. The high-precision laser measurement and calibration judgment module is used to input real-time adjustment position information into the high-precision laser measurement unit, extract and fuse spatial features based on a convolutional self-attention network, and output the current calibration deviation. The multi-level error analysis and fine calibration module is used to dynamically analyze calibration errors and make fine adjustments when the calibration deviation does not meet the standard requirements, by using multi-scale convolution and contextual feature fusion. The calibration result storage and historical database management module is used to integrate and format calibration status data and adjustment parameters, and write them into the historical database.

[0016] The beneficial effects of this invention are: Compared to existing manual and semi-automatic firearm calibration techniques, this invention proposes an intelligent automatic firearm calibration method and system based on 3D modeling and laser measurement, significantly improving the accuracy, efficiency, and intelligence of calibration. Through the comprehensive acquisition and deep fusion of 3D point cloud and multi-view image data, the spatial structure and geometric features of key parts such as the rear sight, front sight, and muzzle are effectively restored and meticulously depicted, overcoming problems such as modeling bias and accumulated spatial coordinate errors inherent in traditional single data sources. During calibration, the system utilizes a high-precision laser measurement unit to collect adjustment information and extract spatial features in real time. Furthermore, a convolutional self-attention network enables multi-scale analysis and dynamic feature optimization of high-dimensional positional information, greatly improving the accuracy and precision of spatial calibration residual analysis and adjustment.

[0017] Furthermore, the innovative introduction of an automatic control unit and a precision robotic arm enables adaptive, step-by-step automatic adjustment of the rear and front sights, fundamentally eliminating subjective errors and consistency issues caused by manual adjustments. During calibration, the system utilizes multi-level error analysis modules and cross-level feature interaction mechanisms to deeply analyze and intelligently compensate for spatial relationships and error sources in complex calibration environments, possessing closed-loop adaptive optimization capabilities. Compared to traditional technologies, this invention enables efficient, consistent, and traceable automatic calibration operations under varying operating conditions and with mass production equipment support, significantly improving the accuracy, maintenance efficiency, and safety reliability of firearms equipment, and laying a solid foundation for subsequent intelligent maintenance and data-driven performance management. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0019] Figure 1 This is a flowchart of the intelligent automatic firearm calibration method based on 3D modeling and laser measurement proposed in this invention; Figure 2 This is a schematic diagram of the intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement proposed in this invention; Figure 3 This is a data flow diagram of the intelligent automatic firearm calibration system based on 3D modeling and laser measurement proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1-2 The intelligent automatic firearm calibration method based on 3D modeling and laser measurement includes the following steps: S1. Collect point cloud data and two-dimensional image data of key parts of the firearm to generate the original measurement dataset; S2. Perform feature extraction and spatial relationship analysis on the original measurement dataset to establish a three-dimensional structural model including the gun's rear sight, front sight, and muzzle, and obtain the spatial coordinate information of each key component; S3. Using the spatial coordinate information of each key component, the three-dimensional structural model is compared with the standard three-dimensional model in the historical database to analyze the spatial coordinate deviation. Based on the analysis results, the rear sight and front sight are automatically adjusted by a precision robotic arm, and the adjustment position information is collected in real time during the adjustment process. S4. Input the real-time adjustment position information into the high-precision laser measurement unit, use a convolutional self-attention network, use the convolutional layer to extract spatial features of the position information, output the current calibration deviation, and determine whether it is within the preset standard range. S5. If the current calibration deviation is not within the preset standard range, the high-dimensional position feature vector is submitted to the cross-level feature interaction unit. By dynamically associating different spatial resolutions with multi-dimensional context information, the calibration error is analyzed in depth and finely adjusted. S6. When the calibration deviation is within the preset standard range, stop the adjustment and store the actual spatial coordinate information, calibration deviation, and adjustment parameters in the historical database.

[0022] This invention achieves accurate 3D modeling and spatial coordinate analysis of key parts by fusing point cloud and image multi-source data. Combined with convolutional self-attention network and cross-level feature interaction mechanism, it dynamically judges and corrects calibration deviations. With the help of robotic arm automatic adjustment, it effectively improves the intelligence, accuracy and real-time performance of automatic firearm calibration, and realizes data storage and adaptive closed-loop optimization throughout the process.

[0023] In this embodiment, the original measurement dataset specifically includes high-density three-dimensional point cloud coordinate data of the gun's rear sight, front sight, and muzzle, multi-view two-dimensional images of the corresponding parts, reflection intensity values ​​of each point cloud, spatial pose parameters of the calibration plate, and ambient lighting information during acquisition.

[0024] This invention achieves high-precision, multi-dimensional characterization of the spatial features of a target area by acquiring high-density 3D point clouds, reflection intensity, multi-view images, ambient lighting, and the spatial pose of a calibration board from the rear sight, front sight, and muzzle of a firearm. This method effectively improves the integrity and recognizability of the 3D model, optimizes the accuracy of feature extraction and spatial relationship analysis, and provides a more comprehensive and accurate data foundation for subsequent automatic calibration and intelligent judgment.

[0025] In this embodiment, the real-time position adjustment information specifically includes the spatial displacement, rotation angle, adjustment direction, adjustment speed, and current spatial coordinate changes of the rear sight and front sight recorded by the precision robotic arm during the adjustment of the rear sight and front sight.

[0026] This invention achieves quantitative tracking of the entire adjustment process by recording the spatial displacement, rotation angle, direction, speed, and coordinate changes of the rear and front sights of a precision robotic arm during adjustment. This method improves the controllability and traceability of the adjustment action, facilitates accurate analysis of parameter influences during the adjustment process, optimizes calibration algorithms, and further enhances the intelligence and accuracy of firearm calibration.

[0027] In this embodiment, S2 specifically includes: S21. Preprocess the original measurement dataset and output the preprocessed point cloud data and two-dimensional image data; S22. Segment the purified point cloud data, extract the point cloud regions of the rear sight, front sight and muzzle, and output the point cloud subsets of each key part. S23. Extract features from the point cloud subsets of each key part to obtain the spatial geometric feature parameters of each key part, and output the feature set of key parts containing spatial location and structural features. S24. Combine two-dimensional image data to perform fusion processing on the feature sets of key parts, enhance the spatial feature expression capability, and output the fused spatial feature data. S25. Based on the fused spatial feature data, establish a three-dimensional structural model including the rear sight, front sight, and muzzle, and output complete three-dimensional structural model data. S26. Perform spatial analysis on the complete three-dimensional structural model data to obtain the spatial coordinate information of the rear sight, front sight, and muzzle.

[0028] This invention achieves high-precision reconstruction of the spatial structure of the rear sight, front sight, and muzzle by preprocessing point cloud and image data in stages, segmenting and extracting key parts, and fusing features and creating 3D models. Combined with multi-dimensional feature fusion, it improves spatial representation and discrimination capabilities, ensuring the accuracy and completeness of spatial coordinate extraction, providing a solid data foundation for subsequent automatic calibration, and significantly enhancing system robustness and calibration reliability.

[0029] In this embodiment, S3 specifically includes: S31. Obtain the spatial coordinate information of key components of the three-dimensional structural model after spatial filtering optimization, and output the optimized current spatial coordinate data; S32. Based on the standard 3D models under different working conditions in the historical database, the multi-model fusion algorithm is used to extract the spatial coordinates of standard key components and output the fused standard spatial coordinate data. S33. Perform adaptive weighted registration on the optimized current spatial coordinate data and the fused standard spatial coordinate data, and output the adaptively registered spatial coordinate comparison data. S34. Perform fitting analysis on the adaptive registration spatial coordinate comparison data, predict the spatial coordinate deviation trend of each key component, and output the spatial coordinate deviation trend data. S35. Based on the spatial coordinate deviation trend data, generate a dynamic adjustment path, control the precision robotic arm to make adaptive step-by-step adjustments to the gate and the crosshair, record the adjustment path information in real time, and output the dynamic adjustment process data. S36. During the adaptive step-by-step adjustment process, the real-time adjustment position information is intelligently calibrated by combining sensor feedback and historical adjustment data, and the calibrated real-time adjustment position information is output.

[0030] This invention extracts standard coordinates through multi-model fusion, adaptively weighted registration, and predicts spatial deviation trends to achieve dynamic optimization and adjustment of key components. By combining sensor feedback and historical data, the robotic arm's adjustment path is calibrated in real time, improving the accuracy and adaptability of the adjustment process. This method significantly enhances the adaptive capability and intelligent level of spatial calibration, providing strong support for highly reliable and accurate calibration.

[0031] In this embodiment, S4 specifically includes: S41. Collect and organize the real-time adjusted position information after calibration, and output standardized position information data; S42. Input the standardized location information data into the high-precision laser measurement unit, perform data fusion processing in combination with the laser measurement data, and output the fused location information data. S43. Spatial grid encoding is performed on the fused location information data. By mapping the original location information to a multi-dimensional spatial grid, the statistical features of each grid cell are extracted and combined with the surrounding grid information for feature enhancement, and high-dimensional location feature data containing spatial structure relationships are output. S44. The processed high-dimensional location feature data is input into a convolutional self-attention network. Spatial features under different receptive fields are extracted through a multi-layer convolutional structure. A multi-scale feature fusion layer is introduced, and then the multi-scale features output from each convolutional layer are fed into the multi-scale feature fusion layer. The features at different scales are weighted and fused through a learnable dynamic weight allocation mechanism. The fused features are then used to reconstruct spatial relationships. Finally, the reconstructed multi-scale fused features are input into the self-attention mechanism layer, and a high-dimensional spatial feature vector is output. ; in, For the first Output high-dimensional positional feature vectors For the first The first scale Each input location feature vector For the first Scale fusion weighting coefficients, For the first The first scale Spatial relationships can be learned weights. For the first The normalization factor for each output, where S is the total number of scales for the multi-scale features, and N is the number of input location features. The output feature index has a value range of 100. , The input feature index has a value range of 100. , This is a scale index, with a value range of [value range missing]. ; S45. Based on high-dimensional space feature vectors, feature weight allocation and correlation analysis are performed through a self-attention mechanism to output the current calibration deviation data. S46. Compare the current calibration deviation data with the preset standard range and output the calibration status judgment result.

[0032] This invention extracts and enhances high-dimensional positional features of key components through high-precision laser measurement and spatial gridding encoding. Combined with a convolutional self-attention network, it achieves multi-scale spatial feature fusion and reconstruction, effectively improving the accuracy and robustness of spatial relationship analysis and calibration deviation determination. Learnable weights are used to dynamically optimize feature representation, enabling intelligent and rapid determination of calibration status, ensuring the system adapts to varying operating conditions and precise calibration requirements.

[0033] In this embodiment, S5 specifically includes: S51. If the current calibration deviation is not within the preset standard range, the current high-dimensional position feature vector is input to the cross-level feature interaction unit to obtain preliminary fused feature information. S52. Perform multi-scale convolution operations on the preliminary fused feature information to extract context features at different spatial resolutions, and then splice and integrate the context features obtained at each scale to form a comprehensive feature expression containing multi-level spatial relationships and context information. S53. Based on multi-level contextual feature representation, spatial correlation weights are dynamically constructed and updated in association with the preliminary fused feature information to generate error analysis feature representations. S54. Using error analysis feature representation, fine-tune the high-dimensional position feature vector and output the optimized high-dimensional position feature vector. S55. The optimized high-dimensional position feature vector is fed back to the calibration unit. Based on the optimized high-dimensional position feature vector, combined with the currently acquired sensor raw data, historical calibration data and current calibration deviation, multiple rounds of feature comparison and deviation residual analysis are used to further refine the calibration results and update the error compensation mechanism of the calibration unit in real time.

[0034] This invention extracts multi-level contextual information through cross-level feature interaction and multi-scale convolution, dynamically constructing spatial correlation weights to achieve accurate analysis of error features and fine adjustment of high-dimensional features. By combining multi-round feature comparison with historical data to optimize the calibration process, it enhances the calibration system's adaptability and correction capabilities to complex spatial biases, ensuring intelligent, efficient, and highly accurate calibration.

[0035] In this embodiment, S6 specifically includes: S61. When the calibration deviation is within the preset standard range, stop adjusting the high-dimensional position feature vector and collect the current actual spatial coordinate information. S63. Integrate the current actual spatial coordinate information with the current calibration deviation to generate the current calibration status data; S64. Collect all adjustment parameters used in the current calibration process and associate them with the current calibration status data to form calibration record entries; S65. Format the calibration record entries according to the historical database data structure to ensure that they can be correctly identified and retrieved by the database. S66. Write the formatted calibration record entries into the historical database.

[0036] This invention collects spatial coordinates and calibration parameters in real time after successful calibration, and integrates them to generate calibration records in a standardized manner, ensuring data format compatibility with the database. This method systematically records key data from each calibration process, enabling orderly archiving and efficient retrieval of calibration history. This facilitates subsequent data analysis, experience optimization, and intelligent parameter learning, improving the system's continuous improvement capabilities and traceability efficiency.

[0037] refer to Figure 3 The intelligent automatic firearm calibration system based on 3D modeling and laser measurement includes the following modules: The 3D data acquisition and standardization module is used to acquire point cloud data and multi-view 2D image data of key parts of firearms. The spatial structure modeling and feature analysis module is used for extracting spatial features of key parts, analyzing spatial relationships, and modeling three-dimensional structures. The automatic adjustment module is used to compare the 3D structural model with the standard 3D model in the historical database, and to automatically adjust the rear sight and front sight through a precision robotic arm. The high-precision laser measurement and calibration judgment module is used to input real-time adjustment position information into the high-precision laser measurement unit, extract and fuse spatial features based on a convolutional self-attention network, and output the current calibration deviation. The multi-level error analysis and fine calibration module is used to dynamically analyze calibration errors and make fine adjustments when the calibration deviation does not meet the standard requirements, by using multi-scale convolution and contextual feature fusion. The calibration result storage and historical database management module is used to integrate and format calibration status data and adjustment parameters, and write them into the historical database.

[0038] Example 1: To verify the feasibility of this invention in practice, it was applied to a weapons and equipment support workshop of a military unit, where a batch of Type 95 automatic rifles were automatically calibrated. The actual effects of the automatic calibration scheme of this invention and traditional manual calibration in terms of calibration accuracy, efficiency, and consistency were compared. The experimental setting was a centralized weapons maintenance warehouse, with standardized workstations, a temperature of 22°C, and a humidity of 45%, located in the equipment maintenance center of a military factory.

[0039] In the traditional manual calibration mode, experienced firearms technicians use conventional visual comparison, mechanical calibration tools, and manual data reading to calibrate an average of 20 rifles per day, with each rifle taking approximately 28 minutes to calibrate. Manual calibration primarily relies on visual inspection, measuring the geometric axes and positions of the muzzle, front sight, and rear sight, and then manually adjusting them. This method has significant limitations in batch operations and calibrating minor errors, as it is highly subjective and results in inconsistent calibration. Often, due to factors such as technician operating habits and visual fatigue, the calibration accuracy of some rifles deviates significantly. Subsequent statistics show a pass rate of approximately 93.5%, a calibration variance of 0.28 mm², and some individual rifles even exhibiting a maximum error of 0.65 mm.

[0040] Introducing the system of this invention into the same maintenance scenario, the system first utilizes 3D point cloud scanning and multi-view cameras to comprehensively acquire spatial data of key parts of the firearm to be calibrated. A laser rangefinder is then used to perform high-precision positioning of key structural data points such as the front sight, rear sight, and muzzle, automatically converting the data into a 3D model. Subsequently, a convolutional self-attention network is used to perform deep feature fusion and automatic error distribution analysis on the point cloud and spatial data. A precision robotic arm automatically adjusts the positions of the rear sight and front sight based on system feedback, automatically monitoring the adjustment error at each step and re-checking the spatial accuracy after adjustment, forming a continuous closed loop.

[0041] Table 1. Comparison of Optimization Effects of Intelligent Customer Service Systems Based on Reinforcement Learning

[0042] Table 1 compares the performance of automatic and manual calibration, including key indicators such as sample size, average calibration time per gun, maximum and minimum errors, average error, calibration variance, pass rate, number of rework requests, and error dispersion. As can be seen from the table, the automatic calibration system of this invention significantly outperforms the traditional manual method in calibration efficiency. The average calibration time is reduced to 9.6 minutes, nearly doubling the efficiency. Simultaneously, the maximum and average errors are significantly reduced, the calibration variance decreases from 0.28 mm² in manual calibration to 0.056 mm², and the pass rate significantly increases from 93.5% to 99.4%. Furthermore, the automatic calibration system requires far fewer rework requests than manual calibration and exhibits less error dispersion, demonstrating higher operational consistency and data reliability, significantly validating the beneficial effects and technical advantages of this invention.

[0043] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent automatic firearm calibration method based on 3D modeling and laser measurement, characterized in that, Includes the following steps: S1. Collect point cloud data and two-dimensional image data of key parts of the firearm to generate the original measurement dataset; S2. Perform feature extraction and spatial relationship analysis on the original measurement dataset to establish a three-dimensional structural model including the gun's rear sight, front sight, and muzzle, and obtain the spatial coordinate information of each key component; S3. Using the spatial coordinate information of each key component, the three-dimensional structural model is compared with the standard three-dimensional model in the historical database to analyze the spatial coordinate deviation. Based on the analysis results, the rear sight and front sight are automatically adjusted by a precision robotic arm, and the adjustment position information is collected in real time during the adjustment process. S4. Input the real-time adjustment position information into the high-precision laser measurement unit, use a convolutional self-attention network, use the convolutional layer to extract spatial features of the position information, output the current calibration deviation, and determine whether it is within the preset standard range. S5. If the current calibration deviation is not within the preset standard range, the high-dimensional position feature vector is submitted to the cross-level feature interaction unit. By dynamically associating different spatial resolutions with multi-dimensional context information, the calibration error is analyzed in depth and finely adjusted. S6. When the calibration deviation is within the preset standard range, stop the adjustment and store the actual spatial coordinate information, calibration deviation, and adjustment parameters in the historical database.

2. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to claim 1, characterized in that, The original measurement dataset specifically includes high-density three-dimensional point cloud coordinate data of the gun's rear sight, front sight, and muzzle, multi-view two-dimensional images of the corresponding parts, reflection intensity values ​​of each point cloud, spatial pose parameters of the calibration plate, and ambient lighting information during acquisition.

3. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to claim 1, characterized in that, The real-time position adjustment information specifically includes the spatial displacement, rotation angle, adjustment direction, adjustment speed, and current spatial coordinate changes of the rear sight and front sight recorded by the precision robotic arm during the adjustment of the rear sight and front sight.

4. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to claim 1, characterized in that, S2 specifically includes: S21. Preprocess the original measurement dataset and output the preprocessed point cloud data and two-dimensional image data; S22. Segment the purified point cloud data, extract the point cloud regions of the rear sight, front sight and muzzle, and output the point cloud subsets of each key part. S23. Extract features from the point cloud subsets of each key part to obtain the spatial geometric feature parameters of each key part, and output the feature set of key parts containing spatial location and structural features. S24. Combine two-dimensional image data to perform fusion processing on the feature sets of key parts, enhance the spatial feature expression capability, and output the fused spatial feature data. S25. Based on the fused spatial feature data, establish a three-dimensional structural model including the rear sight, front sight, and muzzle, and output complete three-dimensional structural model data. S26. Perform spatial analysis on the complete three-dimensional structural model data to obtain the spatial coordinate information of the rear sight, front sight, and muzzle.

5. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to claim 1, characterized in that, S3 specifically includes: S31. Obtain the spatial coordinate information of key components of the three-dimensional structural model after spatial filtering optimization, and output the optimized current spatial coordinate data; S32. Based on the standard 3D models under different working conditions in the historical database, the multi-model fusion algorithm is used to extract the spatial coordinates of standard key components and output the fused standard spatial coordinate data. S33. Perform adaptive weighted registration on the optimized current spatial coordinate data and the fused standard spatial coordinate data, and output the adaptively registered spatial coordinate comparison data. S34. Perform fitting analysis on the adaptive registration spatial coordinate comparison data, predict the spatial coordinate deviation trend of each key component, and output the spatial coordinate deviation trend data. S35. Based on the spatial coordinate deviation trend data, generate a dynamic adjustment path, control the precision robotic arm to make adaptive step-by-step adjustments to the gate and the crosshair, record the adjustment path information in real time, and output the dynamic adjustment process data. S36. During the adaptive step-by-step adjustment process, the real-time adjustment position information is intelligently calibrated by combining sensor feedback and historical adjustment data, and the calibrated real-time adjustment position information is output.

6. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to claim 1, characterized in that, S4 specifically includes: S41. Collect and organize the real-time adjusted position information after calibration, and output standardized position information data; S42. Input the standardized location information data into the high-precision laser measurement unit, perform data fusion processing in combination with the laser measurement data, and output the fused location information data. S43. Spatial grid encoding is performed on the fused location information data. By mapping the original location information to a multi-dimensional spatial grid, the statistical features of each grid cell are extracted and combined with the surrounding grid information for feature enhancement, and high-dimensional location feature data containing spatial structure relationships are output. S44. The processed high-dimensional location feature data is input into a convolutional self-attention network. Spatial features under different receptive fields are extracted through a multi-layer convolutional structure. A multi-scale feature fusion layer is introduced, and then the multi-scale features output from each convolutional layer are fed into the multi-scale feature fusion layer. The features at different scales are weighted and fused through a learnable dynamic weight allocation mechanism. The fused features are then used to reconstruct spatial relationships. Finally, the reconstructed multi-scale fused features are input into the self-attention mechanism layer, and a high-dimensional spatial feature vector is output. ； in, For the first Output high-dimensional positional feature vectors For the first The first scale Each input location feature vector For the first Scale fusion weighting coefficients, For the first The first scale Spatial relationships can be learned weights. For the first The normalization factor for each output, where S is the total number of scales for the multi-scale features, and N is the number of input location features. The output feature index has a value range of 100. , The input feature index has a value range of 100. , This is a scale index, with a value range of [value range missing]. ; S45. Based on high-dimensional space feature vectors, feature weight allocation and correlation analysis are performed through a self-attention mechanism to output the current calibration deviation data. S46. Compare the current calibration deviation data with the preset standard range and output the calibration status judgment result.

7. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to claim 1, characterized in that, S5 specifically includes: S51. If the current calibration deviation is not within the preset standard range, the current high-dimensional position feature vector is input to the cross-level feature interaction unit to obtain preliminary fused feature information. S52. Perform multi-scale convolution operations on the preliminary fused feature information to extract context features at different spatial resolutions, and then splice and integrate the context features obtained at each scale to form a comprehensive feature expression containing multi-level spatial relationships and context information. S53. Based on multi-level contextual feature representation, spatial correlation weights are dynamically constructed and updated in association with the preliminary fused feature information to generate error analysis feature representations. S54. Using error analysis feature representation, fine-tune the high-dimensional position feature vector and output the optimized high-dimensional position feature vector. S55. The optimized high-dimensional position feature vector is fed back to the calibration unit. Based on the optimized high-dimensional position feature vector, combined with the currently acquired sensor raw data, historical calibration data and current calibration deviation, multiple rounds of feature comparison and deviation residual analysis are used to further refine the calibration results and update the error compensation mechanism of the calibration unit in real time.

8. The intelligent automatic firearm calibration method based on three-dimensional modeling and laser measurement according to claim 1, characterized in that, S6 specifically includes: S61. When the calibration deviation is within the preset standard range, stop adjusting the high-dimensional position feature vector and collect the current actual spatial coordinate information. S63. Integrate the current actual spatial coordinate information with the current calibration deviation to generate the current calibration status data; S64. Collect all adjustment parameters used in the current calibration process and associate them with the current calibration status data to form calibration record entries; S65. Format the calibration record entries according to the historical database data structure to ensure that they can be correctly identified and retrieved by the database. S66. Write the formatted calibration record entries into the historical database.

9. An intelligent automatic firearm calibration system based on 3D modeling and laser measurement, comprising the intelligent automatic firearm calibration method based on 3D modeling and laser measurement as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The 3D data acquisition and standardization module is used to acquire point cloud data and multi-view 2D image data of key parts of firearms. The spatial structure modeling and feature analysis module is used for extracting spatial features of key parts, analyzing spatial relationships, and modeling three-dimensional structures. The automatic adjustment module is used to compare the 3D structural model with the standard 3D model in the historical database, and to automatically adjust the rear sight and front sight through a precision robotic arm. The high-precision laser measurement and calibration judgment module is used to input real-time adjustment position information into the high-precision laser measurement unit, extract and fuse spatial features based on a convolutional self-attention network, and output the current calibration deviation. The multi-level error analysis and fine calibration module is used to dynamically analyze calibration errors and make fine adjustments when the calibration deviation does not meet the standard requirements, by using multi-scale convolution and contextual feature fusion. The calibration result storage and historical database management module is used to integrate and format calibration status data and adjustment parameters, and write them into the historical database.

Images