Digital intelligent gun calibration method based on three-dimensional reconstruction

Abstract

The invention discloses a digital intelligent gun calibration method based on three-dimensional reconstruction. The method comprises the following steps that S1, two-dimensional image data obtained through a photoelectric sensor and acoustic ranging data obtained through an acoustic sensor are collected; s2, bullet hole detection is carried out on the two-dimensional image data, and bullet hole two-dimensional pixel coordinates are output; s3, associating the acoustic ranging data with the two-dimensional pixel coordinates of the bullet hole to obtain three-dimensional space coordinates of the bullet hole; s4, inputting the three-dimensional space coordinates of the bullet holes into a third-stage network of the cascaded neural network; s5, optimizing search of optimal parameters in the third-level network of the cascaded neural network in the training process by adopting a squirrel search algorithm; and S6, displaying the three-dimensional coordinates of the final aiming point on the target surface through an aiming point accurate positioning indication module. According to the invention, a squirrel search algorithm, a cascade neural network and a target surface real-time positioning indication technology are combined, and digital intelligent gun calibration based on three-dimensional reconstruction is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent gun calibration technology, and in particular to a digital intelligent gun calibration method based on three-dimensional reconstruction. Background Technology

[0002] Existing shooting calibration and aiming point optimization technologies are widely used in military, police, and sports fields, playing an irreplaceable role, especially in weapon calibration, shooter training, and ballistic analysis. Traditional weapon calibration systems generally rely on manual observation and mechanical aiming aids, with shooters manually adjusting the aiming point position based on experience, and judging the calibration effect by combining bullet marks on the target surface and observation equipment. While these methods were practical in early applications, with the continuous improvement of shooting accuracy requirements and the increasing complexity of shooting environments, traditional calibration methods have gradually revealed many limitations.

[0003] First, traditional weapon calibration systems are highly dependent on the shooter's experience. Significant subjective differences exist among shooters in adjusting aiming points, controlling the calibration process, and interpreting data, easily leading to fluctuations in results and affecting the accuracy and consistency of the final calibration. Simultaneously, manual recording and calibration processes are often labor-intensive, time-consuming, and inefficient, making it difficult to meet the demands of high-intensity, high-frequency shooting training and weapon verification. Especially in complex situations such as changing ambient light and varied shooting postures, the subjective errors of manual calibration become even more pronounced, failing to provide a stable and reliable reference standard for subsequent shooting.

[0004] Secondly, while some existing automated gun-calibrating systems have incorporated auxiliary equipment such as cameras and sensors to achieve bullet hole recognition and preliminary aiming point positioning, most still remain at the level of two-dimensional image analysis and simple geometric correction. These systems primarily identify bullet hole locations through image processing algorithms and calculate aiming point deviations using traditional methods such as least squares and geometric fitting, making it difficult to fully utilize the multi-dimensional features and spatial information of the ballistic trajectory. Due to the lack of modeling and optimization for complex errors in three-dimensional space, the final accuracy of the calculated aiming point is limited, making it difficult to meet the demands of high-precision and high-complexity shooting. Furthermore, existing automated gun-calibrating systems are relatively simple in terms of data processing and optimization algorithms, lacking intelligent and multi-model fusion capabilities, and unable to dynamically and adaptively adjust to different shooting conditions and individual differences.

[0005] In terms of image acquisition and data management, existing systems typically use independent camera equipment to record the entire gun calibration process. However, the integration of image data and aiming point data is rather crude, lacking targeted data integration and a mechanism for tracing the entire gun calibration process. As a result, even if image data of the gun calibration process can be obtained, it is difficult to achieve synchronous correlation between image content and key parameters such as specific aiming point adjustments and shooting results, causing inconvenience for subsequent gun calibration effect analysis and shooter training evaluation. In addition, existing systems often operate in isolation between image acquisition, data storage, and display control, failing to form a highly integrated gun calibration information management platform and reducing the overall level of automation and informatization.

[0006] In recent years, emerging technologies such as artificial intelligence and deep learning have made rapid progress in fields such as shooting assistance and target recognition. Some studies have attempted to introduce neural networks into processes such as bullet hole recognition and aiming point optimization. However, these attempts mostly focus on the training and application of single network models, limiting the algorithm optimization capabilities and making it difficult to cope with complex nonlinear errors and multi-source noise interference in shooting scenarios. At the same time, automated gun calibration systems often neglect key indicators such as algorithm convergence and robustness, leading to problems such as slow training speed, unstable optimization results, and sensitivity to extreme samples during actual deployment. Especially in the areas of multi-dimensional spatial coordinate optimization and high-precision aiming point calculation, existing methods lack a systematic optimization framework and cannot achieve efficient and intelligent parameter search and adjustment.

[0007] Furthermore, existing technologies have certain shortcomings in the physical display of aiming point information and shooter interaction. Traditional systems often use simple laser pointers or LED indicators for target marking, which have limited display accuracy and real-time performance and are greatly affected by environmental interference. While some systems can project data to the terminal display, the information synchronization and interactivity are poor, failing to provide shooters with intuitive, real-time, and accurate aiming adjustment guidelines. The various information flows during the aiming process fail to achieve efficient collaborative transmission, affecting the shooter's operating experience and aiming efficiency.

[0008] Therefore, how to provide a digital intelligent gun calibration method based on 3D reconstruction is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] One objective of this invention is to propose a digital intelligent gun-calibrating method based on 3D reconstruction. This invention efficiently extracts bullet hole feature data and reconstructs 3D space using a multi-level neural network, integrates historical gun-calibrating data to intelligently calculate the optimal aiming point, and employs a squirrel search algorithm to dynamically and globally optimize network parameters, continuously improving the accuracy of aiming point coordinates and system robustness. Furthermore, it integrates a real-time target surface positioning and indication module with a full-process image acquisition and data fusion mechanism, which can accurately map the optimized 3D aiming point onto the target surface, provide real-time feedback on the gun-calibrating status, and display the entire gun-calibrating process information on a terminal. This effectively improves the shooter's gun-calibrating efficiency and shooting accuracy, possessing significant advantages in terms of high automation, intelligence, and traceability.

[0010] A digital intelligent gun calibration method based on three-dimensional reconstruction according to an embodiment of the present invention includes the following steps: S1. Collect two-dimensional image data containing bullet holes obtained through photoelectric sensor and acoustic ranging data obtained through acoustic sensor; S2. Input the two-dimensional image data into the first-level network of the cascaded neural network, perform bullet hole detection on the two-dimensional image data and output the two-dimensional pixel coordinates of the bullet holes; S3. Associate the acoustic ranging data with the two-dimensional pixel coordinates of the bullet hole, input the data into the second-level network of the cascaded neural network, and perform three-dimensional spatial reconstruction to obtain the three-dimensional spatial coordinates of the bullet hole. S4. Input the three-dimensional spatial coordinates of the bullet hole into the third-level network of the cascaded neural network, and combine it with historical gun calibration data to generate the optimal three-dimensional coordinates of the aiming point. S5. Based on the optimal aiming point three-dimensional coordinates, the squirrel search algorithm is used to optimize the search for the optimal parameters in the training process of the third-level network of the cascaded neural network. A behavior exploration unit is configured for each squirrel. Based on the path selection and foraging behavior of all squirrels and exploration modules, the final aiming point three-dimensional coordinates are generated. S6. The three-dimensional coordinates of the final aiming point are displayed on the target surface through the aiming point precision positioning indicator module for the shooter to aim and adjust. The entire aiming process is captured and stored by the aiming point image recording unit, and the target point information is displayed through the shooter display control terminal.

[0011] Optionally, the two-dimensional image data specifically includes an overall target surface image, a bullet hole area image, bullet hole edge features, target surface coordinate markers, ambient lighting information, and pixel grayscale and color related to the bullet hole.

[0012] Optionally, the target information specifically includes the bullet hole position, the optimal aiming point position, the recommended crosshair adjustment amount, and the status of the aiming process.

[0013] Optionally, S2 specifically includes: S21. Perform image denoising and enhancement processing on the acquired two-dimensional image data containing bullet holes to obtain enhanced image data; S22. Perform image segmentation on the enhanced image data to extract candidate regions that may contain bullet holes; S23. Input the candidate region into the first-level network of the cascaded neural network to extract the feature information of the bullet hole region; S24. Based on the feature information output by the first-level network, the candidate regions are judged for bullet holes, and the bullet hole target regions are selected. S25. Accurately locate the selected bullet hole target area and output the two-dimensional pixel coordinates of the bullet hole.

[0014] Optionally, S3 specifically includes: S31. Perform a preliminary spatial estimation of the two-dimensional pixel coordinates of the bullet hole to obtain the initial spatial position of the bullet hole in the image coordinate system. S32. Use acoustic ranging data to perform depth correction on the initial spatial position, and combine distance information to compensate for spatial deviation of two-dimensional pixel coordinates. S33. Input the corrected spatial position and the original input data into the second-level network of the cascaded neural network to extract the spatial feature information of the bullet hole; S34. Based on the spatial feature information of the bullet hole output by the second-level network, perform three-dimensional coordinate optimization on the spatial position to obtain the three-dimensional spatial coordinates of the bullet hole: ; in, Represents the three-dimensional spatial coordinates of the bullet hole. This represents the spatial coordinate data from the i-th source. This represents the weight coefficient corresponding to the i-th source, where i represents the source number of the spatial coordinate data, and N represents the total number of sources.

[0015] Optionally, S4 specifically includes: S41. The three-dimensional spatial coordinates of the bullet holes are fused with historical gun-tracking data to generate fused feature data. S42. Input the fused feature data into the third stage of the cascaded neural network to obtain the network output feature data: ; Where Q represents the network output feature data, Represents the three-dimensional spatial coordinates of the bullet hole. Group j of historical gun-calibration spatial data The weight coefficient corresponding to the j-th group of historical gun calibration data, where j represents the index of the historical gun calibration data and M represents the total number of historical gun calibration data groups; S43. Based on the network output feature data, combined with the three-dimensional spatial coordinates of bullet holes and historical gun calibration data, generate a set of candidate three-dimensional coordinates for aiming points; S44. Filter and sort the candidate three-dimensional coordinates of the aiming point, calculate the mean Euclidean distance from each candidate aiming point to all bullet hole three-dimensional coordinates, sort them in ascending order of mean distance, select the coordinate with the smallest mean distance as the optimal aiming point, and output the optimal aiming point three-dimensional coordinates.

[0016] Optionally, S5 specifically includes: S51. Based on the three-dimensional coordinates of the optimal aiming point, construct the search space of the squirrel search algorithm and initialize the squirrel body. Configure a behavior exploration unit for each squirrel body. S52. Configure the parameters of each squirrel body as the training parameters of the third-level network of the cascaded neural network, input the training samples, and obtain the network output feature set. S53. Calculate the fitness value of each squirrel body based on the network output feature set: ; in, This indicates the fitness value of the squirrel's body. This represents the three-dimensional coordinate components of the k-th candidate aiming point in Q. This represents the three-dimensional coordinate components of the k-th candidate aiming point. This represents the weight of the k-th candidate point. This represents the total number of candidate coordinates, and k represents the index of the three-dimensional coordinate components of the candidate aiming point. S54. Based on the fitness values ​​of all squirrel bodies, simulate path selection and foraging behavior, update the parameter configuration and exploration path of the squirrel bodies, and generate a new generation of squirrel bodies; S55. Continue to input the parameter configuration of the new generation squirrel body into the third level of the cascaded neural network, repeat S52 to S54 until the number of iterations reaches 500 or the fitness value increases by less than 0.0001 in 50 consecutive iterations, then stop the iteration. The current squirrel body parameter configuration is the optimal squirrel body parameter configuration. S56. Based on the optimal squirrel body parameter configuration, output the three-dimensional coordinates of the final aiming point.

[0017] Optionally, S6 specifically includes: S61. Input the final three-dimensional coordinate data of the aiming point into the aiming point precise positioning indicator module to generate the spatial position information of the aiming point; S62. Based on the spatial position information of the aiming point, control the positioning indicator module to display the aiming point at the corresponding physical position on the target surface; S63. Synchronously transmit the spatial position information of the aiming point and the target display status to the gun calibration image recording unit. S64. The gun-calibrating image recording unit continuously captures images of the shooter's entire gun-calibrating process based on the received spatial position information of the aiming point and the target display status. S65. Integrate the acquired image data and aiming point spatial location information to form a complete data file containing the entire process image and related aiming point information; S66. The data file containing the entire process image and the spatial position information of the aiming point is displayed in real time through the shooter's display and control terminal to display the target information.

[0018] The beneficial effects of this invention are: This invention achieves high-precision automatic optimization of the three-dimensional coordinates of the aiming point by introducing an optimization method combining a squirrel search algorithm and a cascaded neural network, effectively improving the positioning accuracy and intelligence level of the shooting calibration system. Through complex high-order error weighting processing of the candidate aiming point coordinate components, the system can adaptively adjust to multi-dimensional spatial errors, greatly improving the robustness and convergence speed of aiming point optimization. Finally, the aiming point is accurately displayed on the target surface in real time via a positioning indicator module. Combined with image recording and data integration, the entire gun calibration process is fully recorded, achieving process traceability and verifiable results, effectively supporting the shooter's rapid calibration and accurate shooting. The overall solution not only improves the automation and intelligence level of gun calibration operations and reduces the uncertainty caused by human intervention, but also provides strong data support and technical assurance for shooting training and weapon certification, demonstrating significant engineering application value and promising prospects for widespread application. Attached Figure Description

[0019] 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:

[0020] Figure 1 This is a flowchart of a digital intelligent gun calibration method based on three-dimensional reconstruction proposed in this invention; Figure 2 This is a schematic diagram of a digital intelligent gun calibration method based on three-dimensional reconstruction proposed in this invention; Figure 3 This is a data flow diagram of a digital intelligent gun calibration method based on three-dimensional reconstruction proposed in this invention. Detailed Implementation

[0021] 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.

[0022] refer to Figure 1-3 A digital intelligent gun calibration method based on 3D reconstruction includes the following steps: S1. Collect two-dimensional image data containing bullet holes obtained through photoelectric sensor and acoustic ranging data obtained through acoustic sensor; S2. Input the two-dimensional image data into the first-level network of the cascaded neural network, perform bullet hole detection on the two-dimensional image data and output the two-dimensional pixel coordinates of the bullet holes; S3. Associate the acoustic ranging data with the two-dimensional pixel coordinates of the bullet hole, input the data into the second-level network of the cascaded neural network, and perform three-dimensional spatial reconstruction to obtain the three-dimensional spatial coordinates of the bullet hole. S4. Input the three-dimensional spatial coordinates of the bullet hole into the third-level network of the cascaded neural network, and combine it with historical gun calibration data to generate the optimal three-dimensional coordinates of the aiming point. S5. Based on the optimal aiming point three-dimensional coordinates, the squirrel search algorithm is used to optimize the search for the optimal parameters in the training process of the third-level network of the cascaded neural network. A behavior exploration unit is configured for each squirrel. Based on the path selection and foraging behavior of all squirrels and exploration modules, the final aiming point three-dimensional coordinates are generated. S6. The three-dimensional coordinates of the final aiming point are displayed on the target surface through the aiming point precision positioning indicator module for the shooter to aim and adjust. The entire aiming process is captured and stored by the aiming point image recording unit, and the target point information is displayed through the shooter display control terminal.

[0023] This invention achieves three-dimensional spatial reconstruction of bullet holes and intelligent optimization of aiming points by integrating cascaded neural networks, a squirrel search algorithm, and multi-sensor data acquisition. The system automatically completes bullet hole detection, coordinate calculation, and real-time target display, improving the accuracy and efficiency of gun calibration. Furthermore, through full-process image acquisition and target information display, it enables traceability and intelligent management of the gun calibration process.

[0024] In this embodiment, the two-dimensional image data specifically includes the overall target surface image, bullet hole area image, bullet hole edge features, target surface coordinate markers, ambient lighting information, and pixel grayscale and color related to the bullet hole.

[0025] This invention improves the accuracy and robustness of bullet hole recognition by collecting data including the entire target surface, bullet hole areas, edge features, coordinate markers, ambient lighting, and pixel grayscale and color. By integrating detailed information from multiple sources, it effectively reduces recognition errors caused by factors such as lighting and complex backgrounds, laying a high-quality data foundation for subsequent bullet hole localization and 3D reconstruction, and significantly enhancing the system's adaptability and detection accuracy.

[0026] In this embodiment, the target information specifically includes the bullet hole position, the optimal aiming point position, the recommended crosshair adjustment amount, and the status of the gun calibration process.

[0027] This invention achieves comprehensive monitoring of shooting errors, aiming optimization, and the aiming process by outputting target information including bullet hole positions, optimal aiming point positions, recommended front sight adjustments, and the status of the aiming process. This multi-dimensional target information helps shooters adjust their aim promptly, improving aiming efficiency and accuracy. It also provides detailed data for subsequent data analysis and shooting training, greatly enhancing the system's intelligent decision-making and support capabilities.

[0028] In this embodiment, S2 specifically includes: S21. Perform image denoising and enhancement processing on the acquired two-dimensional image data containing bullet holes to obtain enhanced image data; S22. Perform image segmentation on the enhanced image data to extract candidate regions that may contain bullet holes; S23. Input the candidate region into the first-level network of the cascaded neural network to extract the feature information of the bullet hole region; S24. Based on the feature information output by the first-level network, the candidate regions are judged for bullet holes, and the bullet hole target regions are selected. S25. Accurately locate the selected bullet hole target area and output the two-dimensional pixel coordinates of the bullet hole.

[0029] This invention achieves high-precision screening and localization of bullet hole regions by denoising, enhancing, and segmenting bullet hole image data, combined with hierarchical feature extraction and target discrimination using a cascaded neural network. This method effectively improves the accuracy of bullet hole detection, reduces the impact of background interference on the detection results, provides accurate and reliable basic data for subsequent 3D reconstruction and aiming point optimization, and significantly enhances the system's adaptability and detection stability in complex environments.

[0030] In this embodiment, S3 specifically includes: S31. Perform a preliminary spatial estimation of the two-dimensional pixel coordinates of the bullet hole to obtain the initial spatial position of the bullet hole in the image coordinate system. S32. Use acoustic ranging data to perform depth correction on the initial spatial position, and combine distance information to compensate for spatial deviation of two-dimensional pixel coordinates. S33. Input the corrected spatial position and the original input data into the second-level network of the cascaded neural network to extract the spatial feature information of the bullet hole; S34. Based on the spatial feature information of the bullet hole output by the second-level network, perform three-dimensional coordinate optimization on the spatial position to obtain the three-dimensional spatial coordinates of the bullet hole: ; in, Represents the three-dimensional spatial coordinates of the bullet hole. This represents the spatial coordinate data from the i-th source. This represents the weight coefficient corresponding to the i-th source, where i represents the source number of the spatial coordinate data, and N represents the total number of sources.

[0031] This invention achieves depth correction of the spatial position by combining the two-dimensional pixel coordinates of bullet holes with acoustic ranging data, and optimizes the three-dimensional spatial coordinates of bullet holes by fusing multi-source features from a cascaded neural network. Employing a weighted fusion mechanism, it fully utilizes spatial data from different sources, effectively reducing coordinate estimation errors and improving the accuracy and robustness of 3D reconstruction, providing a highly reliable spatial positioning foundation for ballistic analysis and intelligent gun aiming.

[0032] In this embodiment, S4 specifically includes: S41. The three-dimensional spatial coordinates of the bullet holes are fused with historical gun-tracking data to generate fused feature data. S42. Input the fused feature data into the third stage of the cascaded neural network to obtain the network output feature data: ; Where Q represents the network output feature data, Represents the three-dimensional spatial coordinates of the bullet hole. Group j of historical gun-calibration spatial data The weight coefficient corresponding to the j-th group of historical gun calibration data, where j represents the index of the historical gun calibration data and M represents the total number of historical gun calibration data groups; S43. Based on the network output feature data, combined with the three-dimensional spatial coordinates of bullet holes and historical gun calibration data, generate a set of candidate three-dimensional coordinates for aiming points; S44. Filter and sort the candidate three-dimensional coordinates of the aiming point, calculate the mean Euclidean distance from each candidate aiming point to all bullet hole three-dimensional coordinates, sort them in ascending order of mean distance, select the coordinate with the smallest mean distance as the optimal aiming point, and output the optimal aiming point three-dimensional coordinates.

[0033] This invention integrates the three-dimensional spatial coordinates of bullet holes with historical gun-calibrating data, and utilizes a cascaded neural network for deep feature extraction to achieve intelligent filtering and sorting of candidate aiming point coordinates. By minimizing the mean of Euclidean distance, the optimal aiming point is accurately determined, effectively improving the scientific rigor and accuracy of aiming point selection. It also considers historical data and current bullet hole distribution patterns, significantly enhancing the system's adaptability and gun-calibrating decision-making efficiency.

[0034] In this embodiment, S5 specifically includes: S51. Based on the three-dimensional coordinates of the optimal aiming point, construct the search space of the squirrel search algorithm and initialize the squirrel body. Configure a behavior exploration unit for each squirrel body. S52. Configure the parameters of each squirrel body as the training parameters of the third-level network of the cascaded neural network, input the training samples, and obtain the network output feature set. S53. Calculate the fitness value of each squirrel body based on the network output feature set: ; in, This indicates the fitness value of the squirrel's body. This represents the three-dimensional coordinate components of the k-th candidate aiming point in Q. This represents the three-dimensional coordinate components of the k-th candidate aiming point. This represents the weight of the k-th candidate point. This represents the total number of candidate coordinates, and k represents the index of the three-dimensional coordinate components of the candidate aiming point. S54. Based on the fitness values ​​of all squirrel bodies, simulate path selection and foraging behavior, update the parameter configuration and exploration path of the squirrel bodies, and generate a new generation of squirrel bodies; S55. Continue to input the parameter configuration of the new generation squirrel body into the third level of the cascaded neural network, repeat S52 to S54 until the number of iterations reaches 500 or the fitness value increases by less than 0.0001 in 50 consecutive iterations, then stop the iteration. The current squirrel body parameter configuration is the optimal squirrel body parameter configuration. S56. Based on the optimal squirrel body parameter configuration, output the three-dimensional coordinates of the final aiming point.

[0035] This invention employs a squirrel search algorithm combined with the output features of a cascaded neural network to dynamically optimize the three-dimensional coordinates of the aiming point by simulating foraging behavior and path selection. A fitness evaluation and multi-round iterative update mechanism effectively enhances global optimization capabilities and avoids getting trapped in local optima. This method achieves intelligent adaptive optimization of aiming point parameters, significantly improving aiming accuracy and system robustness, and meeting the needs for efficient gun calibration in complex scenarios.

[0036] In this embodiment, S6 specifically includes: S61. Input the final three-dimensional coordinate data of the aiming point into the aiming point precise positioning indicator module to generate the spatial position information of the aiming point; S62. Based on the spatial position information of the aiming point, control the positioning indicator module to display the aiming point at the corresponding physical position on the target surface; S63. Synchronously transmit the spatial position information of the aiming point and the target display status to the gun calibration image recording unit. S64. The gun-calibrating image recording unit continuously captures images of the shooter's entire gun-calibrating process based on the received spatial position information of the aiming point and the target display status. S65. Integrate the acquired image data and aiming point spatial location information to form a complete data file containing the entire process image and related aiming point information; S66. The data file containing the entire process image and the spatial position information of the aiming point is displayed in real time through the shooter's display and control terminal to display the target information.

[0037] This invention automates aiming point positioning, display, and full-process image acquisition by integrating the three-dimensional spatial coordinates of the aiming point with the target surface display, gun calibration image recording, and data. The synchronous integration of full-process images and spatial information facilitates real-time review and analysis of the gun calibration process for shooters, effectively improving the traceability, operational transparency, and training efficiency of gun calibration, and providing high-quality basic data support for subsequent intelligent diagnosis and data mining.

[0038] Example 1: To verify the feasibility of this invention in practice, it was applied to a field test project of an "Intelligent Shooting Correction Assistance System" organized by an armed police force at a field firing range. This test aimed to solve problems such as large errors in aiming points, difficulty in tracing and analyzing the correction process, and low training efficiency associated with traditional manual aiming correction. Ten sets of standard rifle targets were set up at the firing range, each target measuring 50cm × 50cm, with the long-range shooting point set at 100 meters. Fifteen shooters participated in the test, each completing three rounds of shooting, firing 15 rounds of ammunition per round. Throughout the test, the system automatically collected bullet hole images and acoustic data, and based on this, intelligently optimized the aiming point, displayed the results, and recorded the entire process.

[0039] At the testing site, the system first captured images of each target using a high-precision front-end camera. Due to the complex environment, variations in lighting, wind, and target paper reflections all affected the clarity of the bullet hole images. Therefore, the image denoising and enhancement algorithm described in this invention was employed to effectively suppress noise interference and make the bullet hole edges clearer. After enhancement, the system automatically segmented all candidate bullet hole regions and then used the first layer of a cascaded neural network to perform feature analysis and bullet hole discrimination on each candidate region, accurately identifying the real bullet holes and outputting the two-dimensional pixel coordinates of the bullet holes. For example, in the first round of shooting on the morning of October 14, 2024, a total of 14 bullet hole targets were detected, with a false positive rate of only 2.3%, an improvement of approximately 12.5 percentage points compared to traditional manual identification.

[0040] Subsequently, the system performs depth correction on the spatial position of each bullet hole based on the two-dimensional pixel coordinates of each bullet hole, combined with data from the acoustic sensor. This step not only compensates for spatial deviations caused by lens distortion and changes in shooting angle, but also greatly improves the accuracy of bullet hole positioning. Taking the third set of shots as an example, the average error of the three-dimensional spatial coordinates of the bullet holes was only 1.2mm, and the maximum error did not exceed 3.4mm. After multiple rounds of comparison, the repeatability of the three-dimensional spatial coordinates of the automatically positioned bullet holes is better than 98%, meeting the requirements for high-precision shooting and gun calibration.

[0041] Next, the current three-dimensional spatial coordinates of bullet holes were integrated with the unit's three previous gun-calibration data. The third layer of a cascaded neural network was used to deeply extract and fuse features. The weights for fusing historical gun-calibration data were dynamically adjusted based on the data's age and correlation, ensuring that the output aiming point candidate set considered both the current firing distribution and historical firing deviation patterns. For example, in the eighth test conducted on the afternoon of October 14, 2024, the system-generated aiming point candidate set, sorted by the mean of Euclidean distance, ultimately selected an optimal aiming point with a deviation rate approximately 17% lower than that obtained using manual experience, effectively avoiding systematic errors introduced by subjective judgment.

[0042] To further optimize the spatial coordinates of the aiming point, this invention introduces a squirrel search algorithm. A search space is constructed near the optimal aiming point, with 40 squirrels representing a specific parameter configuration. In each iteration, the system simulates foraging and path selection based on the network output features and fitness function. After a maximum of 500 iterations, the algorithm converges to the optimal squirrel parameters. Actual test data shows that the global optimization capability of the aiming point spatial coordinates is significantly improved after adopting the squirrel search algorithm. Compared with the traditional genetic algorithm, the root mean square error of the aiming point is reduced by 9.2%, and the convergence speed is increased by approximately 34%.

[0043] Table 1. Comparison of Optimization Effects of Intelligent Customer Service Systems Based on Reinforcement Learning

[0044] Table 1 demonstrates the high adaptability and robustness of this invention under various lighting, wind, and shooter skill levels. Statistics show that the average accuracy rate for automatic bullet hole identification is 96.4%, the average spatial positioning error is 1.3 mm, and the average time for automatic aiming point optimization and indication is 3.7 seconds. The overall gun-calibrating process is approximately 41% faster than the manual process, and the entire process is recorded synchronously with video and position data without frame drops or missed targets. Shooter satisfaction surveys indicate that 92% of shooters believe this invention significantly improves gun-calibrating efficiency and training quality.

[0045] 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. A digital intelligent gun calibration method based on three-dimensional reconstruction, characterized in that, Includes the following steps: S1. Collect two-dimensional image data containing bullet holes obtained through photoelectric sensor and acoustic ranging data obtained through acoustic sensor; S2. Input the two-dimensional image data into the first-level network of the cascaded neural network, perform bullet hole detection on the two-dimensional image data and output the two-dimensional pixel coordinates of the bullet holes; S3. Associate the acoustic ranging data with the two-dimensional pixel coordinates of the bullet hole, input the data into the second-level network of the cascaded neural network, and perform three-dimensional spatial reconstruction to obtain the three-dimensional spatial coordinates of the bullet hole. S4. Input the three-dimensional spatial coordinates of the bullet hole into the third-level network of the cascaded neural network, and combine it with historical gun calibration data to generate the optimal three-dimensional coordinates of the aiming point. S5. Based on the optimal aiming point three-dimensional coordinates, the squirrel search algorithm is used to optimize the search for the optimal parameters in the training process of the third-level network of the cascaded neural network. A behavior exploration unit is configured for each squirrel. Based on the path selection and foraging behavior of all squirrels and exploration modules, the final aiming point three-dimensional coordinates are generated. S6. The three-dimensional coordinates of the final aiming point are displayed on the target surface through the aiming point precision positioning indicator module for the shooter to aim and adjust. The entire aiming process is captured and stored by the aiming point image recording unit, and the target point information is displayed through the shooter display control terminal.

2. The digital intelligent gun calibration method based on three-dimensional reconstruction according to claim 1, characterized in that, The two-dimensional image data specifically includes the overall target surface image, bullet hole area image, bullet hole edge features, target surface coordinate markers, ambient lighting information, and pixel grayscale and color related to the bullet hole.

3. The digital intelligent gun calibration method based on three-dimensional reconstruction according to claim 1, characterized in that, The target information specifically includes bullet hole location, optimal aiming point location, recommended crosshair adjustment amount, and gun calibration process status.

4. The digital intelligent gun calibration method based on three-dimensional reconstruction according to claim 1, characterized in that, S2 specifically includes: S21. Perform image denoising and enhancement processing on the acquired two-dimensional image data containing bullet holes to obtain enhanced image data; S22. Perform image segmentation on the enhanced image data to extract candidate regions that may contain bullet holes; S23. Input the candidate region into the first-level network of the cascaded neural network to extract the feature information of the bullet hole region; S24. Based on the feature information output by the first-level network, the candidate regions are judged for bullet holes, and the bullet hole target regions are selected. S25. Accurately locate the selected bullet hole target area and output the two-dimensional pixel coordinates of the bullet hole.

5. The digital intelligent gun calibration method based on three-dimensional reconstruction according to claim 1, characterized in that, S3 specifically includes: S31. Perform a preliminary spatial estimation of the two-dimensional pixel coordinates of the bullet hole to obtain the initial spatial position of the bullet hole in the image coordinate system. S32. Use acoustic ranging data to perform depth correction on the initial spatial position, and combine distance information to compensate for spatial deviation of two-dimensional pixel coordinates. S33. Input the corrected spatial position and the original input data into the second-level network of the cascaded neural network to extract the spatial feature information of the bullet hole; S34. Based on the spatial feature information of the bullet hole output by the second-level network, perform three-dimensional coordinate optimization on the spatial position to obtain the three-dimensional spatial coordinates of the bullet hole: ； in, Represents the three-dimensional spatial coordinates of the bullet hole. This represents the spatial coordinate data from the i-th source. This represents the weight coefficient corresponding to the i-th source, where i represents the source number of the spatial coordinate data, and N represents the total number of sources.

6. The digital intelligent gun calibration method based on three-dimensional reconstruction according to claim 1, characterized in that, S4 specifically includes: S41. The three-dimensional spatial coordinates of the bullet holes are fused with historical gun-tracking data to generate fused feature data. S42. Input the fused feature data into the third stage of the cascaded neural network to obtain the network output feature data: ； Where Q represents the network output feature data, Represents the three-dimensional spatial coordinates of the bullet hole. Group j of historical gun-calibration spatial data The weight coefficient corresponding to the j-th group of historical gun calibration data, where j represents the index of the historical gun calibration data and M represents the total number of historical gun calibration data groups; S43. Based on the network output feature data, combined with the three-dimensional spatial coordinates of bullet holes and historical gun calibration data, generate a set of candidate three-dimensional coordinates for aiming points; S44. Filter and sort the candidate three-dimensional coordinates of the aiming point, calculate the mean Euclidean distance from each candidate aiming point to all bullet hole three-dimensional coordinates, sort them in ascending order of mean distance, select the coordinate with the smallest mean distance as the optimal aiming point, and output the optimal aiming point three-dimensional coordinates.

7. The digital intelligent gun calibration method based on three-dimensional reconstruction according to claim 1, characterized in that, S5 specifically includes: S51. Based on the three-dimensional coordinates of the optimal aiming point, construct the search space of the squirrel search algorithm and initialize the squirrel body. Configure a behavior exploration unit for each squirrel body. S52. Configure the parameters of each squirrel body as the training parameters of the third-level network of the cascaded neural network, input the training samples, and obtain the network output feature set. S53. Calculate the fitness value of each squirrel body based on the network output feature set: ； in, This indicates the fitness value of the squirrel's body. This represents the three-dimensional coordinate components of the k-th candidate aiming point in Q. This represents the three-dimensional coordinate components of the k-th candidate aiming point. This represents the weight of the k-th candidate point. This represents the total number of candidate coordinates, and k represents the index of the three-dimensional coordinate components of the candidate aiming point. S54. Based on the fitness values ​​of all squirrel bodies, simulate path selection and foraging behavior, update the parameter configuration and exploration path of the squirrel bodies, and generate a new generation of squirrel bodies; S55. Continue to input the parameter configuration of the new generation squirrel body into the third level of the cascaded neural network, repeat S52 to S54 until the number of iterations reaches 500 or the fitness value increases by less than 0.0001 in 50 consecutive iterations, then stop the iteration. The current squirrel body parameter configuration is the optimal squirrel body parameter configuration. S56. Based on the optimal squirrel body parameter configuration, output the three-dimensional coordinates of the final aiming point.

8. The digital intelligent gun calibration method based on three-dimensional reconstruction according to claim 1, characterized in that, S6 specifically includes: S61. Input the final three-dimensional coordinate data of the aiming point into the aiming point precise positioning indicator module to generate the spatial position information of the aiming point; S62. Based on the spatial position information of the aiming point, control the positioning indicator module to display the aiming point at the corresponding physical position on the target surface; S63. Synchronously transmit the spatial position information of the aiming point and the target display status to the gun calibration image recording unit. S64. The gun-calibrating image recording unit continuously captures images of the shooter's entire gun-calibrating process based on the received spatial position information of the aiming point and the target display status. S65. Integrate the acquired image data and aiming point spatial location information to form a complete data file containing the entire process image and related aiming point information; S66. The data file containing the entire process image and the spatial position information of the aiming point is displayed in real time through the shooter's display and control terminal to display the target information.

Images