Multi-target laser pointing system and control method thereof

Abstract

The application relates to the technical field of machine vision, and provides a multi-target laser pointing system and a control method thereof. The control method comprises the following steps: acquiring a training data set, training a first sub-network model and a second sub-network model by using the training data set, determining a first real-time angle, and controlling a galvanometer assembly to rotate at the first real-time angle; determining a second real-time angle; and controlling the galvanometer assembly to rotate at the second real-time angle, so that a laser point of laser deflected by the galvanometer assembly coincides with the coordinates of a target to be measured. The control method has the advantages of long distance, high precision and small error, is suitable for multi-target dynamic tracking and accurate positioning in a complex environment, and effectively improves the automation level of laser measurement and processing; through real-time feedback and closed-loop control of a deep learning model, angle correction can be completed within a millisecond-level response time, the laser point is ensured to be stably focused on the coordinates of the target to be measured, and the demand for human intervention is significantly reduced.

Description

Technical Field

[0001] This application relates to the field of machine vision technology, and in particular to a multi-target laser pointing system and its control method. Background Technology

[0002] Multi-target laser pointing systems can simultaneously acquire distance information of multiple targets within the field of view relative to the observation point. They are one of the core technologies in the field of modern intelligent sensing. Traditional single-laser systems can only measure one target at a time, while multi-target laser pointing systems, by combining machine vision and laser scanning technology, achieve rapid and accurate distance measurement of multiple targets. They are widely used in various fields such as intelligent monitoring and industrial inspection.

[0003] However, in multi-target laser pointing systems, the complex spatial distribution of multiple targets easily leads to ranging interference problems caused by laser echo signal aliasing and overlapping fields of view. This is especially problematic in long-range multi-target scenarios, where existing detection methods struggle to effectively distinguish the reflected light signals from adjacent targets, resulting in misjudgments of distance and reduced detection accuracy. Furthermore, the system's measurement range is limited by the response speed of its iterative adjustment and feedback mechanisms, making it difficult to simultaneously meet the requirements of a large field of view and high precision, thus limiting its real-time tracking capability for multiple targets in dynamic environments. Inefficient calibration methods, reliant on manual intervention, are ill-suited for rapid deployment in complex conditions, leading to significant system errors.

[0004] Therefore, there is an urgent need for a long-range, high-precision, and low-error multi-target laser pointing system. Summary of the Invention

[0005] This application provides a multi-target laser pointing system and its control method to solve the technical problems of limited distance, low accuracy and large error in existing detection methods.

[0006] The control method for a multi-target laser pointing system provided in the first aspect of this application includes: acquiring a training dataset; the training dataset includes a first training video, a second training video, and a training distance; the first training video is obtained by a first camera component capturing images of a training target, and the second training video is obtained by a second camera component capturing images of the training target, wherein both the first and second training videos include the pixel coordinates of the training target; a laser and a galvanometer component are disposed between the first and second camera components, and the optical axis of the second camera component is coaxial with the 0-point optical axis of the galvanometer component; the laser is configured to generate laser light, which is reflected by the galvanometer component to the training target; the training distance is obtained from the return light generated after the training target receives the laser light; a first sub-network model and a second sub-network model are trained using the training dataset; the input of the trained first sub-network model is the first training video. The first training angle is the output of the first sub-network model after training, and the second training angle is the output of the second sub-network model after training, which is the input of the second training video and the training distance. Both the first and second training angles are angles that need to be controlled to rotate the galvanometer assembly. The second training angle is a correction angle of the first training angle. The first real-time angle is determined based on the first real-time video, the real-time distance, and the first sub-network model. The first real-time video is obtained by the first camera assembly capturing images of the target under test. The galvanometer assembly is controlled to rotate at the first real-time angle. The second real-time angle is determined based on the second real-time video, the real-time distance, and the second sub-network model. The second real-time video is obtained by the second camera assembly capturing images of the target under test. The galvanometer assembly is controlled to rotate at the second real-time angle so that the laser point of the laser deflected by the galvanometer assembly coincides with the coordinates of the target under test.

[0007] In some feasible implementations, the training dataset is used to train the first sub-network model and the second sub-network model, including: processing the first training video and outputting the pixel coordinates of the training target; converting the pixel coordinates of the training target into the offset angle of the optical axis of the first camera component relative to the training target; determining the visual compensation amount based on the auxiliary distance and the training distance; the auxiliary distance is the distance between the baseline of the first camera component and the galvanometer component; and outputting the first training angle based on the offset angle and the visual compensation amount.

[0008] In some feasible implementations, training the first sub-network model and the second sub-network model using the training dataset also includes: processing the second training video and outputting the pixel deviation of the training target relative to the center of the second training video; converting the pixel deviation into an angle deviation in response to the pixel deviation being greater than a preset pixel; and outputting the second training angle based on the angle deviation and the training distance.

[0009] In some feasible implementations, training the first sub-network model and the second sub-network model using a training dataset also includes: outputting a second training angle of zero in response to a pixel deviation being less than or equal to a preset pixel.

[0010] In some feasible implementations, training the first sub-network model and the second sub-network model using a training dataset further includes: generating a first training signal based on a first training angle and sending the first training signal to a galvanometer assembly; the galvanometer assembly being configured to rotate the first training angle based on the first training signal; generating a second training signal based on a second training angle and sending the second training signal to the galvanometer assembly; the galvanometer assembly being further configured to rotate the second training angle based on the second training signal.

[0011] In some feasible implementations, obtaining the training dataset includes: obtaining training data; preprocessing the training data to obtain the training dataset; wherein, the preprocessing includes rotating, translating, scaling, and adding noise to the first training video and the second training video; the preprocessing also includes normalizing the training distance; the training dataset includes a training set, a validation set, and a test set; the ratio of the training set, validation set, and test set is 8:1:1.

[0012] In some feasible implementations, the loss function of both the first and second sub-network models is MSE; the optimizer of both the first and second sub-network models is AdamW, where the learning rate of the first sub-network model is 1e-4 and the learning rate of the second sub-network model is 5e-5.

[0013] The control method for the multi-target laser pointing system provided in the first aspect of this application, on the one hand, ensures high-quality and noise-free training data through a visualized aiming mechanism, laying a data foundation for a high-precision model. On the other hand, it actively compensates for the parallax effect caused by the physical baseline through a built-in network model, fundamentally eliminating the main error source in long-distance measurement, thereby achieving excellent overall measurement accuracy. It transforms the previously invisible laser pointing into a real-time visible process, turning the entire aiming and measurement process from a "black box" to a "white box." Operators can intuitively monitor, verify, and intervene, which not only greatly enhances user trust in the multi-target laser pointing system but also makes the debugging, maintenance, and fault diagnosis of the multi-target laser pointing system exceptionally simple.

[0014] The multi-target laser pointing system provided in the second aspect of this application includes a first camera assembly and a second camera assembly, both configured to capture video of a target under test; a laser, disposed between the first and second camera assemblies, configured to emit laser light; a galvanometer assembly, disposed in the optical path of the laser, configured to rotate relative to the laser to transmit the laser light to the target under test; the optical axis of the second camera assembly is coaxial with the 0-point optical axis of the galvanometer assembly and the optical axis of the laser; a beam splitter, disposed in the optical path of the laser, and located between the galvanometer assembly and the second camera assembly; the beam splitter is configured to reflect the laser light to the galvanometer assembly; and a processor connected to the first camera assembly, the second camera assembly, the laser, and the galvanometer assembly, configured to adjust the rotation angle of the galvanometer assembly so that the laser spot deflected by the galvanometer assembly coincides with the coordinates of the target under test.

[0015] In some feasible implementations, the laser wavelength is 1572 nm and the repetition frequency is 5 kHz.

[0016] In some feasible implementations, the multi-target laser pointing system also includes a receiver, located on the side of the second camera assembly away from the beam splitter, the receiver being configured to receive the reflected light from the laser beam after passing through the target under test.

[0017] The multi-target laser pointing system provided in the second aspect of this application establishes a unified coordinate mapping model decoupled from the distance to the target. This overcomes the inherent limitations of traditional methods, which require calibration at specific distances and are unsuitable for long distances. It achieves high-precision aiming across the entire measurement range, from near-field to ultra-far-field, greatly expanding its application scenarios. By introducing coaxial visual feedback, the inefficient "blind adjustment" iterative search process is completely eliminated. This multi-target laser pointing system boasts high aiming accuracy, short response time, and high success rate. The data acquisition and aiming process has been optimized from "minutes" to "seconds," enabling fully automated or semi-automated rapid operation and significantly improving the efficiency of system deployment and actual measurement. Attached Figure Description

[0018] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of a multi-target laser pointing system provided in an embodiment of this application;

[0020] Figure 2 This is a schematic flowchart of a control method for a multi-target laser pointing system provided in an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of a drone taking pictures in different states, provided by an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of the model training process in a specific implementation provided in this application embodiment.

[0023] Illustration markings:

[0024] 10-First camera assembly; 20-Second camera assembly; 30-Laser; 40-Galvanometer assembly; 41-Control unit; 50-Beam splitter; 60-Processor; 70-Receiver; 71-Avalanche photodiode; M-Target under test. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are all within the protection scope of this application.

[0026] In the following description, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0027] Furthermore, in this application, directional terms such as "upper," "lower," "inner," and "outer" are defined relative to the indicated placement of the components in the accompanying drawings. It should be understood that these directional terms are relative concepts, used for relative description and clarification, and can change accordingly depending on the placement of the components in the accompanying drawings.

[0028] First, the concepts mentioned in the embodiments of this application will be briefly introduced.

[0029] Laser ranging: The basic function of a laser system, which calculates distance by measuring the time difference or phase difference between the emitted and received reflected laser signals.

[0030] Parallax: The difference in viewing angle caused by the baseline between the optical center of the camera and the laser emission center is the fundamental source of long-distance aiming error.

[0031] Baseline: The physical distance between the optical center of the camera and the laser emission center, which is the direct cause of parallax.

[0032] Galvanometer: An actuator that uses electric current to drive the deflection of a reflector, thereby precisely controlling the direction of a laser beam.

[0033] Coaxial optical path: refers to the optical design in which the optical axes of the ranging laser and the auxiliary camera are completely coincident in space, which is the basis for realizing "what you see is what you shoot".

[0034] Coordinate systems: The system involves multiple coordinate systems, including the image coordinate system, camera coordinate system, galvanometer coordinate system, and world coordinate system.

[0035] Camera calibration: The process of determining camera intrinsic parameters and distortion coefficients.

[0036] Internal parameters: The camera's own attribute parameters, including focal length and principal point.

[0037] To address the technical problems of slow detection calibration speed, short distance, low accuracy, and large error in existing detection methods, this application provides a multi-target laser pointing system. This system provides a high-precision, vision-based "spatial coordinate mapping system" that enables absolute coordination between visual perception and physical beam control, allowing a controlled laser beam to automatically, quickly, and accurately hit any distant target identified by the visual sensor.

[0038] Figure 1 This is a schematic diagram of the structure of a multi-target laser pointing system provided in an embodiment of this application.

[0039] See Figure 1 The multi-target laser pointing system includes: a first camera assembly 10, a second camera assembly 20, a laser 30, a galvanometer assembly 40, a beam splitter 50, and a processor 60.

[0040] Both the first camera assembly 10 and the second camera assembly 20 are configured to capture video of the target M under test. There are multiple targets M under test.

[0041] Specifically, the first camera component 10 can directly capture video of the test area of ​​the target M along the first direction to generate a first real-time video; the second camera component 20 can capture video of the target M in the same test area along the second direction through the reflection of the galvanometer component 40 to generate a second real-time video.

[0042] The first direction can be horizontal, and the second direction can be vertical.

[0043] A laser 30 is disposed between the first camera assembly 10 and the second camera assembly 20, and is configured to emit laser light. A galvanometer assembly 40 is disposed in the optical path of the laser and is configured to rotate relative to the laser 30 to transmit the laser light to the target M to be measured.

[0044] Specifically, the galvanometer assembly 40 can be angled to change the direction of laser reflection, thereby achieving scanning coverage of the target M at different positions; the beam splitter 50 is disposed in the optical path of the laser and located between the galvanometer assembly 40 and the second camera assembly 20, and is used to reflect the laser to the galvanometer assembly 40.

[0045] The processor 60 is connected to the first camera assembly 10, the second camera assembly 20, the laser 30 and the galvanometer assembly 40 respectively. The processor 60 is configured to adjust the rotation angle of the galvanometer assembly 40 so that the laser spot of the laser deflected by the galvanometer assembly 40 coincides with the coordinates of the target M to be measured.

[0046] Specifically, the processor 60 can also receive the reflected light from the laser beam from the target M and determine the distance in real time. The processor 60 may include a ranging module, which calculates the distance to the target M based on the time difference or phase difference information of the reflected light. The processor 60 can also adjust the galvanometer assembly 40 based on the first and second real-time videos to enable the laser to capture the position of the target M in real time. That is, when there are multiple targets M, the galvanometer assembly 40 can be rotated in real time to align the targets M.

[0047] In one feasible implementation, the first camera component 10 can be a camera, specifically a 50-megapixel, 800mm focal length, F6 folding adjustable lens. Its primary function is to transmit the field of view image to the processor 60 for target recognition. By adjusting the lens focus, the number of pixels occupied by the target M is increased, enhancing feature recognition accuracy and enabling high-accuracy recognition at greater distances. The maximum distance between the camera and the target M can reach 3km, maintaining clear imaging and accurate recognition of the target M even at this distance.

[0048] In one example, the galvanometer assembly 40 includes a first galvanometer and a second galvanometer (not shown in the figure). The first galvanometer can rotate horizontally to adjust the angle between the first galvanometer and the horizontal direction, thereby changing the reflection angle of the laser in the horizontal direction. The second galvanometer can rotate vertically to adjust the angle between the second galvanometer and the vertical direction, thereby changing the reflection angle of the laser in the vertical direction. Through the coordinated rotation of the two galvanometers, rapid scanning and precise positioning of the laser in a two-dimensional plane are achieved, ensuring that the laser can accurately point to the specified position of the target M.

[0049] In some feasible implementations, the multi-target laser pointing system may further include a control unit 41, which is disposed between the processor 60 and the galvanometer assembly 40. The control unit 41 receives instructions from the processor 60 and drives the galvanometer assembly 40 to rotate according to the instructions, thereby adjusting the scanning angle of the laser. The control unit 41 may be a motor, and the beam deflection module composed of the motor and the galvanometer assembly 40 can deflect the angle of the galvanometer assembly 40 by calculating the control voltage of the multi-target laser pointing system, thereby deflecting the laser beam.

[0050] The control component 41 may include a first control component and a second control component (not shown in the figure). The first control component is connected to the first galvanometer and is used to control the rotation angle of the first galvanometer in the horizontal direction. The second control component is connected to the second galvanometer and is used to control the rotation angle of the second galvanometer in the vertical direction.

[0051] In some feasible implementations, laser 30 can be a microchip laser in a wavelength safe for the human eye, with a wavelength of 1572nm and a repetition frequency of 5kHz, meeting the light intensity requirements for long-distance ranging. This wavelength has good penetration and anti-interference capabilities, while avoiding harm to the human eye.

[0052] In some feasible implementations, the average power of the laser 30 can be 30mW; the laser beam diameter can be 30mm; and the laser beam coincides with the central optical axis of the second camera assembly 20.

[0053] The second camera component 20 can be a camera module that includes 50 megapixels, digital zoom, and optical zoom. Its optical axis is set to be coaxial with the 0-point optical axis of the galvanometer component 40 and the optical axis of the laser to ensure that the laser and the imaging field of view are always consistent.

[0054] The second camera assembly 20 is used to correct the parallax caused by the misalignment between the first camera assembly 10 and the laser window. It provides an image for visual recognition and enables fine-tuning of the angle of the galvanometer assembly 40, ensuring that the center of the second real-time video image coincides with the target M under test.

[0055] In some feasible implementations, the multi-target laser pointing system may also include a receiver 70. The receiver 70 receives the reflected light and can be positioned on the side of the second camera assembly 20 away from the beam splitter 50. This ensures that the receiver 70's position does not interfere with laser emission and effectively receives the reflected light from the target M, thereby obtaining accurate distance information. The receiver 70 possesses high sensitivity and low noise characteristics, effectively identifying weak echo signals in the reflected light and ensuring ranging stability in complex environments. This multi-target laser pointing system, through the collaborative operation of multiple modules, achieves long-distance, multi-target, and high-precision real-time ranging and position tracking functions. It can stably identify and measure the distance and position information of multiple targets M within a 3km range. The receiver 70 consists of a flexible fiber bundle and a lens; its numerical aperture can be set to >1 through design.

[0056] In some feasible implementations, the multi-target laser pointing system may also include an avalanche photodiode 71, which is positioned between the processor 60 and the receiver 70 to pre-amplify the weak light signal captured by the receiver 70, thereby improving the signal-to-noise ratio. The avalanche photodiode 71 possesses high gain and fast response characteristics, and can effectively convert light pulses into electrical signals.

[0057] The multi-target laser pointing system provided in this application establishes a unified coordinate mapping model decoupled from the distance to the target M. This overcomes the inherent limitations of traditional methods, which require calibration at specific distances and are unsuitable for long distances. It achieves high-precision aiming across the entire range from near-field to ultra-far-field (e.g., 3km), greatly expanding its application scenarios. By introducing coaxial visual feedback, such as visible light indication or a second camera component 20, the inefficient "blind adjustment" iterative search process is completely eliminated, resulting in high automation and significantly increased deployment and measurement efficiency. This multi-target laser pointing system boasts high aiming accuracy (distance error between the laser point and the target center ≤5cm@1km is acceptable), short response time (≤200ms from recognition to aiming), and high success rate (≥95% across all scenarios). The data acquisition and aiming process has been optimized from "minute-level" to "second-level," enabling fully or semi-automated rapid operation and significantly improving the efficiency of system deployment and actual measurement.

[0058] It should be emphasized that the multi-target laser pointing system provided in this application embodiment can not only realize distance detection, but also perform other types of laser detection.

[0059] Corresponding to the aforementioned embodiments of the multi-target laser pointing system, this application also provides embodiments of the control method for the multi-target laser pointing system.

[0060] Figure 2This is a schematic flowchart of a control method for a multi-target laser pointing system provided in an embodiment of this application.

[0061] See Figure 2 As shown, the control method for a multi-target laser pointing system may include the following steps S1 to S6.

[0062] Step S1: Obtain the training dataset; the training dataset includes a first training video, a second training video, and a training distance; the first training video is obtained by the first camera component 10 capturing images of the training target, the second training video is obtained by the second camera component 20 capturing images of the training target, and the training distance is obtained based on laser detection of the training target; wherein, both the first and second training videos include the pixel coordinates of the training target; the laser 30 and the galvanometer component 40 are positioned between the first camera component 10 and the second camera component 20, and the optical axis of the second camera component 20 is coaxial with the O-point optical axis of the galvanometer component 40; the laser 30 is configured to generate laser light, which is reflected by the galvanometer component 40 to the training target; the training distance is obtained from the return light generated after the training target receives the laser light.

[0063] When acquiring training data, it is necessary to clearly define the identification requirements and scene parameters, determine the target categories to be identified, and specify the shooting environment, real-time requirements, and accuracy standards. For example, if the target category is a drone, car, or person, corresponding features can be set for different target types. For instance, when identifying a car, the car is relatively large, or when identifying a drone, the drone is relatively small. This allows for differentiation of target categories based on size. Alternatively, target categories can be distinguished by features such as the size, location, and shape of the heat source, thus enabling targeted identification and improving positioning accuracy. The shooting environment can include indoor, outdoor, or lighting conditions; real-time requirements can include millisecond-level response, and accuracy requirements can include accurate identification thresholds.

[0064] Figure 3 This application provides an embodiment of a photographic illustration of a drone in different states; wherein, Figure 3 Image (a) shows a schematic diagram of a drone in operation. Figure 3 (b) and (c) are schematic diagrams of the drone being photographed in a stationary state.

[0065] In one example, see Figure 3 As shown, this is a simple explanation using drones as the target category. See also Figure 3 As shown in (a), when the UAV is operational, its rotor generates a strong airflow. This airflow disturbs the optical detection area. This disturbance manifests on the detection surface as a bright spot with blurred boundaries and a diffused outline. See also Figure 3As shown in (b), when the UAV hovers or rests near the detection area, the rotor does not generate significant airflow. At this time, the UAV body and its stationary rotor form a relatively concentrated, well-defined bright spot on the detection surface. See also... Figure 3 As shown in (c), in another static scene or a higher resolution detection setting, the light spot appears as four clearly separated bright circular bright spots. These four bright spots precisely correspond to the positions of the four rotors of the UAV and are symmetrically distributed around the intersection of the detection reference crosshairs. Step S1 may include steps S11 and S12.

[0066] Step S11: Obtain training data.

[0067] After clarifying the above requirements, training data can be collected from image or video datasets covering target categories under different scenes, angles, and postures, such as hovering drones, flying drones, cars of different models, and people in different clothing. Annotation tools such as LabelImg are used to label the training data with target categories and location information. The first and second training videos can be captured under various conditions.

[0068] To better understand the control method of the multi-target laser pointing system provided in this application, the specific process of acquiring training data is described below.

[0069] During the acquisition of training data, calibration is first performed using the lens intrinsic parameters of the first camera component 10 and the second camera component 20. During calibration, a checkerboard pattern can be used to calibrate the lens intrinsic parameters, where the lens intrinsic parameters may include the coordinates of the principal focal point (…). , Parameters such as the distortion coefficient are used to convert "pixel coordinates" into "angle offset". Next, the parameters of the galvanometer assembly 40 are calibrated, and the mapping relationship between the galvanometer assembly 40's "control signal (such as voltage / pulse)" and the "actual deflection angle" is recorded. This mapping relationship can be linear or nonlinear. After determining the mapping relationship, the angle output by artificial intelligence (AI) can be directly converted into the corresponding galvanometer assembly 40 control signal. Finally, the baseline distance between the optical axis of the first camera assembly 10 and the laser exit of the galvanometer assembly 40 is measured, and a beam combiner is used to ensure that the optical axis of the second camera assembly 20 is completely coaxial with the laser beam. This can be verified by illuminating a distant target and observing whether the target center of the second camera assembly 20 coincides with the laser point, thus achieving baseline and coaxial calibration.

[0070] Before acquiring the training dataset, preparation is needed. The training target can be a reflective target with a Quick Response Code (QR). A drone carrying a 50×50cm high-reflectivity target sign with a QR code is used to iterate through scene changes such as "distance D, angle (azimuth / pitch)".

[0071] Table 1

[0072]

[0073] See Table 1 for the variable types and specific dimension settings during the training data acquisition process, which are used to systematically collect training data in multiple scenarios.

[0074] Step S12: Preprocess the training data to obtain the training dataset.

[0075] The training data is preprocessed to enhance sample diversity and ensure that the first and second sub-network models have good generalization ability in complex environments.

[0076] In some feasible implementations, preprocessing may include one or more data augmentation techniques, such as rotation, translation, scaling, and noise addition, to augment the training dataset on the first and second training videos.

[0077] Specifically, preprocessing can enhance the images in the first and second training videos.

[0078] The preprocessing step involves randomly rotating the images from the first and second training videos by an angle ranging from ±3° to ±7°, such as ±5°.

[0079] Alternatively, preprocessing can involve translating the images of the first and second training videos within a range of ±8 to ±12 pixels, such as ±10 pixels.

[0080] Alternatively, preprocessing can involve adding noise to the first and second training videos, such as adding Gaussian noise.

[0081] It can also simulate slight movement of the training target and environmental disturbances to improve the adaptability of the first and second sub-network models to complex environments.

[0082] In some feasible implementations, preprocessing also includes normalizing the training distance.

[0083] Specifically, the training distance D is normalized to [0, 1]. Normalization can be based on distance, such as 3km being 1 when the training distance D is 3km. The angle label is normalized to the deflection range of the galvanometer assembly 40, such as [-30°, 30°] → [-1, 1].

[0084] After preprocessing the training data, the training dataset can be divided into a training set, a validation set, and a test set in a ratio of 8:1:1. After the division, the training set is used for fitting, the validation set for parameter tuning, and the test set for final evaluation. It is crucial to ensure that the same scene does not cross sets; for example, samples within a 3km radius should not be simultaneously in the training and test sets.

[0085] In some feasible implementations, during the training process, the data is iteratively optimized. For samples that fail in the test set, such as those caused by occlusion leading to incorrect shooting and recognition by the first camera component 10, or insufficient near-range parallax compensation, similar data is collected to fine-tune the model. If the fine-tuning accuracy is insufficient, the image resolution of the second camera component 20 can be increased or the feature extraction layer of the convolutional neural network (CNN) can be optimized.

[0086] Step S2: Train the first sub-network model and the second sub-network model using the training dataset; the input of the trained first sub-network model is the first training video and the training distance, and the output of the trained first sub-network model is the first training angle; the input of the trained second sub-network model is the second training video and the training distance, and the output of the trained second sub-network model is the second training angle; both the first training angle and the second training angle are angles that need to be controlled to rotate the galvanometer assembly 40; the second training angle is a correction angle of the first training angle.

[0087] In the inputs of the first sub-network model and the second sub-network model, the first training video serves as the original video and is used for the recognition and coarse localization of the training target. The pixel coordinates of the training target can be obtained from the first training video. , During the extraction process, data can be extracted manually or using traditional algorithms. After model training is complete, AI can be used to replace manual methods. The ranging module in processor 60 outputs the training distance D, which is used to calculate disparity compensation. The second training video is used as the original video for fine-tuning feedback. This fully automated data acquisition method greatly improves efficiency, enhances the inherent accuracy of the model through high-quality data sources, and establishes a universal coordinate mapping model decoupled from distance, which has strong universality and achieves the goal of "one-time calibration, full-domain measurement".

[0088] In the outputs of the first and second sub-network models, the first training angle is used as the initial coarse adjustment angle. , The angle calculated based on the first training video can be used as an intermediate label for phased training; the second training angle serves as the final precise deflection angle. , ): That is, "the actual deflection angle of the galvanometer assembly 40 when the laser accurately illuminates the target", which is recorded when the training target is in the center of the image in the second training video.

[0089] During model training, deep learning frameworks such as PyTorch and TensorFlow can be used to load training datasets, set parameters such as learning rate and number of iterations, and monitor model performance in real time using validation sets. If overfitting occurs, methods such as dropout and regularization can be used to adjust the model until the class recognition accuracy and positioning accuracy on the training dataset meet the standards. Then, the model is deployed and connected to the camera component for data integration. The trained model is converted into a format suitable for hardware operation, such as the TensorRT engine, and deployed to the processor 60 in this application. The processor 60 can be an edge computing device or a server. The camera components are connected via a Software Development Kit (SDK) interface to achieve frame extraction and data preprocessing (such as size normalization and pixel normalization) of the real-time video streams from the first camera component 10 and the second camera component 20. Real-time inference and result output are then performed. The model performs inference calculations on the preprocessed video frames, outputting the category label, confidence score, and bounding box coordinates for each target. The results are visualized (e.g., overlaying the target box and category text onto the video frame). Simultaneously, the recognition results can be stored in a database as needed. Finally, continuous iterative optimization is performed. Samples of model recognition errors (such as misidentification cases under occlusion or severe weather) are collected during actual use, added to the training dataset, and the model is retrained to continuously improve the robustness of multi-target category recognition.

[0090] During model training, the first training angle is used to adjust the position of the galvanometer assembly 40. Rotating the galvanometer assembly 40 by this first training angle achieves coarse adjustment of its position. The second training angle is used to fine-tune the final pointing of the galvanometer assembly 40, ensuring that the laser or observation equipment can accurately align with the center of the training target. Through the collaborative work of the first and second sub-network models, the first sub-network model first completes large-scale target localization and coarse angle adjustment output, and then the second sub-network model performs high-precision identification and fine-tuning angle calculation in local areas based on more refined features, thus achieving a fast and accurate two-stage adjustment mechanism in multi-target scenarios. The entire process achieves a balance between real-time performance and accuracy, making it suitable for continuous tracking and response needs in complex dynamic environments.

[0091] Here, the first training angle and the second training angle need to be associated with the control instructions for the galvanometer assembly 40. Specifically, for the adjustment of the galvanometer assembly 40, the processor 60 may also include a galvanometer assembly 40 control algorithm module, which is used to generate corresponding galvanometer assembly 40 drive signals according to the first training angle and the second training angle, and control the deflection of the galvanometer assembly 40 in the horizontal and vertical directions.

[0092] In step S2, it is necessary to first establish the first sub-network model and the second sub-network model.

[0093] The first sub-network model can be a mature pre-trained model, such as the YOLO series, Faster R-CNN, or SSD, to improve the recognition of multiple target types. Alternatively, ResNet or MobileNet can be used, with fine-tuning of the structure and matching of target features such as the fuselage of a drone or the contour features of a person, based on the model's backbone network.

[0094] Specifically, the working process of the first sub-network model and the second sub-network model is as follows: the first camera component 10 takes a picture → the first sub-network model coarsely adjusts the angle and outputs the first real-time angle → the galvanometer component 40 initially deflects the first training angle → the second camera component 20 takes a picture → the second sub-network model finely adjusts the angle and outputs the second real-time angle → the galvanometer component 40 precisely deflects the second training angle.

[0095] After establishing the first and second sub-network models, they are trained using the training dataset.

[0096] In some feasible implementations, specifically, the coarse localization process of the first sub-network model can eliminate most of the disparity, with the input being a first training video (containing image features of the training target) and the training distance. The output is a first training angle (…). , ),in, This is the coarse adjustment angle in the horizontal direction. This is the coarse adjustment angle in the vertical direction. Based on the target perception and training distance D of the first camera component 10, the first sub-network model outputs the "initial deflection angle" of the galvanometer component 40, allowing the laser to roughly illuminate the training target, providing a prerequisite for fine-tuning the second sub-network model.

[0097] Training the first sub-network model and the second sub-network model using training data includes training the first sub-network model using a first training video and a training distance, and training the second sub-network model using a second training video and a training distance.

[0098] Specifically, the first sub-network model is trained using the first training video and training distance, including:

[0099] Step S21: Process the first training video and output the pixel coordinates of the training target.

[0100] The model architecture of the first sub-network model includes feature extraction and geometry + learning fusion.

[0101] Feature extraction may involve processing images from a first training video using a lightweight CNN (YOLO-Nano) to output the pixel coordinates of the training target. , " and eigenvectors, where, Let x be the x-coordinate pixel of the training target in the first training video. The feature vector, which represents the vertical coordinate pixel of the training target in the first video, replaces traditional image recognition and can improve robustness.

[0102] Step S22: Convert the pixel coordinates of the training target into the offset angle of the optical axis of the first camera component 10 relative to the training target.

[0103] In the integration of geometry and learning, the calibrated intrinsic parameters are first used to... , ) is converted to "the angle offset of the optical axis of the first camera component 10 relative to the training target ( , )",in, The offset is in the horizontal direction. The offset is in the vertical direction. , ; , These are the principal point coordinates of the first camera component 10; d is the pixel size; and f is the focal length.

[0104] Step S23: Determine the visual compensation amount based on the auxiliary distance and the training distance. The auxiliary distance is the distance between the baseline of the first camera component 10 and the galvanometer component 40.

[0105] Calculate the disparity compensation amount by combining the auxiliary distance b and the training distance D. , ),in, This represents the parallax compensation amount in the horizontal direction. This represents the parallax compensation amount in the vertical direction. There is baseline parallax only in the horizontal direction; it is negligible in the vertical direction. Thus, the coarse adjustment angle (first training angle) = original offset angle + disparity compensation amount.

[0106] Step S24: Output the first training angle based on the offset angle and visual compensation amount.

[0107] Fitting "with fully connected layers" " The mapping of "" may exhibit nonlinearity due to mechanical errors, making neural networks more accurate than pure geometric calculations. The output of the first sub-network model is: .

[0108] Step S25: Generate a first training signal based on the first training angle and send the first training signal to the galvanometer assembly 40; the galvanometer assembly 40 is configured to rotate the first training angle based on the first training signal.

[0109] In this step, the first training signal may include command pulses controlling the rotation of the galvanometer assembly 40 in the horizontal and vertical directions, and the controller 41 drives the galvanometer assembly 40 to precisely adjust the reflecting mirrors to... and The corresponding angular position.

[0110] Galvanometer assembly 40 driver: The software in processor 60 automatically ( , The signal is converted into a control signal, such as voltage / pulse, for the galvanometer assembly 40, which drives the galvanometer assembly 40 to complete the coarse adjustment based on the previously calibrated angle-signal mapping.

[0111] The second sub-network model is trained using the second training video and training distance, including:

[0112] Step S26: Process the second training video and output the pixel deviation of the training target relative to the center of the second training video.

[0113] The fine-tuning process of the second sub-network model can be based on coaxial feedback, with the input being the second training video (which contains image features of the training target) and the training distance. The output is the second training angle ( , ).in, The offset is in the horizontal direction. The offset is in the vertical direction, meaning the actual adjustment angle is the sum of the coarse adjustment angle and the fine adjustment angle. , .

[0114] In some feasible implementations, the model architecture of the second sub-network model includes: bias perception and angle regression.

[0115] Bias perception: A CNN, such as a lightweight version of ResNet-18, can be used to process the images in the second camera component 20, outputting the "pixel deviation" (or "pixel offset") of the center of the training target relative to the center of the images in the second training video. , )".

[0116] Step S27: In response to a pixel deviation greater than a preset pixel, convert the pixel deviation into an angle deviation.

[0117] Angle regression: When the pixel deviation exceeds a preset pixel value, fine-tuning is required. , ) is converted into "angle deviation" through intrinsic parameters. , )".

[0118] Step S28: Output the second training angle based on the angle deviation and training distance.

[0119] In this step, the training distance D (far distance bias has less impact on the angle) is used to output a fine-tuned angle using a regression head (fully connected layer + ReLU). , ).

[0120] In one specific implementation, the fine-tuning angle calculation can be performed using the intrinsic parameters of the second camera component 20 to calculate the pixel deviation ( , ) converted to the fine-tuning angle required for the galvanometer assembly 40 ( , The formula is: , ; , and These are the pixel size and focal length of the second camera component 20, respectively.

[0121] Note: Because the second camera component 20 is coaxial with the laser, , The model primarily fits the "correction of distance-bias-angle mapping".

[0122] Step S29: Generate a second training signal based on the second training angle and send the second training signal to the galvanometer assembly 40; the galvanometer assembly 40 is also configured to rotate the second training angle based on the second training signal.

[0123] In this step, the second training signal may include command pulses controlling the rotation of the galvanometer assembly 40 in the horizontal and vertical directions, and the control unit 41 may drive the galvanometer assembly 40 to precisely adjust the reflecting mirrors to... and Corresponding angular position. Fine-tuning and cycling of the galvanometer assembly 40: The software in the processor 60 automatically sends fine-tuning signals (current angle) to the galvanometer assembly 40. , The galvanometer assembly 40 then responds and completes attitude correction, so that the laser beam is precisely aligned with the center of the target.

[0124] In some feasible implementations, the loss function for both the first and second sub-network models is MSE, with the loss function of the first sub-network model used for prediction. The mean squared error of the ground truth can be used as the optimizer, which can be AdamW with a learning rate of 1e-4. During training, the CNN feature layers are first frozen for pre-training (using ImageNet weights), and then the fully connected layers are fine-tuned.

[0125] The loss function of the second sub-network model is used for prediction. The mean squared error of the true value can be used as the optimizer, which can be AdamW with a learning rate of 5e-5. During training, the focus is on fitting samples smaller than the nearest distance (<1km) because the residual error of near distance disparity is large.

[0126] In some feasible implementations, step S271 may also be included.

[0127] Step S271: In response to a pixel deviation being less than or equal to a preset pixel, the second training angle is output as zero. In this step, if the pixel deviation is less than or equal to the preset pixel, it indicates that the multi-target laser pointing system has determined that the laser is accurately aligned with the target, and the multi-target laser pointing system terminates the fine-tuning process. The multi-target laser pointing system enters a stable working state, and the laser continues to be aligned with the training target. Therefore, in this state, no fine-tuning is required, and the second training angle is zero.

[0128] In some feasible implementations, the core of the training process of the first sub-network model and the second sub-network model is to take advantage of the hardware advantage of the second camera component 20 being coaxial with the laser. Therefore, when the training target is in the center of the image in the second training video, the laser is in a state of accurately illuminating the training target. Thus, during the training process, a closed-loop control algorithm for fine-tuning the image deviation control galvanometer component 40 can be realized, achieving fully automatic fine-tuning and automated closed-loop control.

[0129] Specifically, this includes: the second camera component 20 automatically extracts the pixel coordinates of the training target ( , ), calculate its relationship with the image center ( , Pixel deviation, including horizontal pixel deviation Vertical pixel deviation .

[0130] For example, in deviation judgment, if the horizontal pixel deviation... Pixels, vertical pixel deviation If the pixel count (preset precision threshold, corresponding to an angle deviation <0.1mrad) is reached, then fine-tuning is complete, and the second training angle is zero.

[0131] Fine-tuning and cycling of galvanometer assembly 40: The software automatically sends fine-tuning signals (current angle) to galvanometer assembly 40. , Then repeat this step 1 to 3 times until the deviation meets the threshold (usually convergence is achieved in 1 to 3 cycles). To avoid control oscillations, a proportional-integral-derivative (PID) control algorithm can be used instead of "pure proportional fine-tuning". By adjusting the PID parameters (Kp, Ki, Kd), the deviation can be brought to a fast convergence without overshoot.

[0132] In some feasible implementations, after fine-tuning is complete, the software needs to automatically and synchronously collect and save all "input features" and "labels" to form training samples. The core is "trigger signal + structured storage". The triggering condition includes that when the fine-tuning closed-loop control satisfies "deviation ≤ threshold", the software generates a "collection trigger signal" for recording.

[0133] In some feasible implementations, key optimization techniques may include hard sample mining, disparity modular learning, and quantization compression.

[0134] Difficult sample mining includes automatically selecting samples with the "top 10% of loss values" during training, such as close-range, large-angle, and complex lighting scenes, and repeatedly feeding them into the training to improve the model's adaptability to extreme scenes.

[0135] The disparity modular learning includes setting up a separate "disparity compensation submodule" in the first sub-network model, using known b and D to calculate theoretical disparity as supervision, allowing the model to quickly learn the disparity rules.

[0136] Quantization compression includes performing INT8 quantization on the model after training to reduce inference latency, thereby ensuring the real-time performance of laser aiming with an inference time of <50ms.

[0137] By positioning the first camera component 10, the laser is roughly aligned with the target, and full automation can be achieved based on the pre-programmed geometric logic of "pixel coordinates → angle calculation → galvanometer component 40 control".

[0138] It should be emphasized that the first sub-network model and the second sub-network model can be two independent model structures, or they can be two sub-model parts within a single model.

[0139] After the first and second sub-network models are trained, they are deployed in processor 60 to form a closed loop of "perception-decision-execution-feedback", and the model is optimized by feeding back feedback through actual testing.

[0140] Specifically, the termination conditions for the closed loop include: if the deviation between the training target and the image center in the second training video is ≤1 pixel (corresponding to an angular deviation <0.1mrad, satisfying precise aiming), the deflection stops; otherwise, the fine-tuning is repeated (maximum 2 times to avoid redundancy). When the termination conditions are met, the multi-target laser pointing system automatically locks the galvanometer assembly at a 40° angle and outputs an aiming completion signal.

[0141] This application provides a control method for a multi-target laser pointing system, which has the advantages of long-range, high precision, and small error. It is suitable for dynamic tracking and precise positioning of multiple targets in complex environments, effectively improving the automation level of laser measurement and processing. Through real-time feedback and closed-loop control of a deep learning model, angle correction can be completed within a millisecond response time, ensuring that the laser point is stably focused on the coordinates of the target under test, significantly reducing the need for human intervention.

[0142] Figure 4 This is a schematic diagram of the model training process in a specific implementation provided in this application embodiment.

[0143] In a specific implementation, see Figure 4 As shown, model training may include steps S301 to S307.

[0144] Step S301: Hardware pre-calibration.

[0145] In this step, the first camera assembly 10, the second camera assembly 20, and the galvanometer assembly 40 can be jointly calibrated.

[0146] Step S302: Data set collection.

[0147] Steps S301 and S302 can be performed by referring to step S11.

[0148] Step S303: Artificial intelligence model design.

[0149] Choose the appropriate neural network architecture based on the task requirements.

[0150] Step S304: Establishment of artificial intelligence model.

[0151] Step S305: Model training.

[0152] Step S306: Evaluation.

[0153] Step S307: Model output.

[0154] Steps S303 to S307 can refer to step S2 above.

[0155] Step S3: Determine the first real-time angle based on the first real-time video, real-time distance, and first sub-network model; the first real-time video is obtained by the first camera component 10 capturing images of the target M to be tested.

[0156] After training the first sub-network model and the second sub-network model, the entire multi-target laser pointing system is put into practical application. The first real-time video is the real-time video of the target M to be tested captured by the first camera component 10, and the real-time distance is the real-time value fed back by the laser ranging module. In this step, the first sub-network model performs initial positioning of the target M to be tested based on the first real-time video and the real-time distance, outputs a coarse adjustment angle command, drives the galvanometer component 40 to quickly align with the center area of ​​the target M to be tested, and enters the fine adjustment stage after the coarse adjustment is completed.

[0157] Step S4: Control the galvanometer assembly 40 to rotate at the first real-time angle.

[0158] In this step, the first real-time angle can include either a horizontal or a vertical angle. The galvanometer assembly 40 adjusts its horizontal and vertical orientation according to the first real-time angle to align the target area with the center of the field of view of the second camera assembly 20. After receiving the first real-time angle command, the galvanometer assembly 40 quickly adjusts its posture to bring the target area into the field of view of the second camera assembly 20.

[0159] Step S5: Determine the second real-time angle based on the second real-time video, real-time distance, and second sub-network model; the second real-time video is obtained by the second camera component 20 capturing images of the target M under test.

[0160] In this step, the second camera component 20 can capture a second real-time video containing high-resolution local images for fine feature extraction. Acquiring the second real-time video ensures clear representation of the local features of the target M, providing high-quality image data for subsequent accurate ranging. The second real-time video and real-time distance are input into the second sub-network model to determine the second real-time angle; the second real-time angle is a correction angle to the first real-time angle.

[0161] Step S6: Control the galvanometer assembly 40 to rotate at the second real-time angle so that the laser spot of the laser deflected by the galvanometer assembly 40 coincides with the coordinates of the target M to be measured.

[0162] In this step, the laser spot can be aligned with the target M to be measured.

[0163] In some feasible implementations, the control method for a multi-target laser pointing system may also include step S7.

[0164] Step S7: In response to the pixel deviation of the target M relative to the center of the second real-time video being greater than a preset pixel, step S5 is repeated. This effectively ensures the alignment accuracy of the laser. The repetition is less than or equal to three times, with each repetition recalculating the second real-time angle based on the updated second real-time video and real-time distance, until the pixel deviation meets the preset condition or the maximum number of repetitions is reached. If accurate alignment is not achieved within three repetitions, the multi-target laser pointing system determines that the target M is out of lock and triggers an alarm, ensuring measurement reliability. This enhances the transparency of human-computer interaction and system credibility, facilitating on-site debugging and troubleshooting.

[0165] The control method for the multi-target laser pointing system provided in this application, on the one hand, ensures high-quality and noise-free training data through a visualized aiming mechanism, laying a data foundation for high-precision models. On the other hand, the device actively compensates for the parallax effect caused by the physical baseline through a built-in mathematical model, fundamentally eliminating the main error source in long-distance measurements, thereby achieving excellent overall measurement accuracy. The process is transparent and controllable, and the multi-target laser pointing system boasts high reliability and ease of use, transforming the previously invisible laser pointing into a real-time visible process, turning the entire aiming and measurement process from a "black box" to a "white box." Operators can intuitively monitor, verify, and intervene, which not only greatly enhances user trust in the multi-target laser pointing system but also makes debugging, maintenance, and fault diagnosis of the system exceptionally simple.

[0166] It should be noted that, upon considering the specification and practicing the application disclosed herein, those skilled in the art will readily conceive of other embodiments of this application. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0167] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The true scope is indicated by this application.

Claims

1. A control method for a multi-target laser pointing system, characterized in that, include: Obtain the training dataset; The training dataset includes a first training video, a second training video, and a training distance; The first training video is obtained by capturing the training target using a first camera component, and the second training video is obtained by capturing the training target using a second camera component. Both the first and second training videos include the pixel coordinates of the training target. A laser and a galvanometer assembly are positioned between the first and second camera components, with the optical axis of the second camera component coaxial with the zero-point optical axis of the galvanometer assembly. The laser is configured to generate laser light, which is reflected by the galvanometer assembly to the training target. The training distance is obtained from the return light generated after the training target receives the laser light. The training dataset is used to train a first sub-network model and a second sub-network model. The first sub-network model, after training, takes the first training video and the training distance as input and outputs a first training angle as output. The second sub-network model, after training, takes the second training video and the training distance as input and outputs a second training angle as output. Both the first and second training angles are angles that need to be used to control the rotation of the galvanometer assembly. The second training angle is a correction angle to the first training angle. The first real-time angle is determined based on the first real-time video, the real-time distance, and the first sub-network model; the first real-time video is obtained by the first camera component capturing images of the target under test; Control the galvanometer assembly to rotate at the first real-time angle; The second real-time angle is determined based on the second real-time video, the real-time distance, and the second sub-network model; the second real-time video is obtained by the second camera component capturing images of the target under test. The galvanometer assembly is controlled to rotate at the second real-time angle so that the laser spot of the laser deflected by the galvanometer assembly coincides with the coordinates of the target to be measured. Training the first sub-network model and the second sub-network model using the training dataset includes: The first training video is processed to output the pixel coordinates of the training target; The pixel coordinates of the training target are converted into the offset angle of the optical axis of the first camera component relative to the training target; The visual compensation amount is determined based on the auxiliary distance and the training distance; the auxiliary distance is the distance between the baselines of the first camera component and the galvanometer component. The first training angle is output based on the offset angle and the visual compensation amount.

2. The control method for the multi-target laser pointing system according to claim 1, characterized in that, Training the first sub-network model and the second sub-network model using the aforementioned training dataset further includes: The second training video is processed to output the pixel deviation of the training target relative to the center of the second training video; In response to the pixel deviation being greater than a preset pixel, the pixel deviation is converted into an angle deviation; The second training angle is output based on the angle deviation and the training distance.

3. The control method for the multi-target laser pointing system according to claim 2, characterized in that, Training the first sub-network model and the second sub-network model using the aforementioned training dataset further includes: In response to the pixel deviation being less than or equal to the preset pixel, the second training angle is output as zero.

4. The control method for the multi-target laser pointing system according to claim 1, characterized in that, Training the first sub-network model and the second sub-network model using the aforementioned training dataset further includes: A first training signal is generated based on the first training angle, and the first training signal is sent to the galvanometer assembly; the galvanometer assembly is configured to rotate the first training angle based on the first training signal. A second training signal is generated based on the second training angle, and the second training signal is sent to the galvanometer assembly; the galvanometer assembly is also configured to rotate the second training angle based on the second training signal.

5. The control method for the multi-target laser pointing system according to claim 1, characterized in that, Obtain the training dataset, including: Obtain training data; The training data is preprocessed to obtain a training dataset; The preprocessing includes rotating, translating, scaling, and adding noise to the first training video and the second training video; the preprocessing also includes normalizing the training distance; the training dataset includes a training set, a validation set, and a test set; the ratio of the training set, the validation set, and the test set is 8:1:

1.

6. The control method for the multi-target laser pointing system according to claim 4, characterized in that, The loss function for both the first sub-network model and the second sub-network model is MSE; The optimizers for both the first and second sub-network models are AdamW, with the learning rate of the first sub-network model being 1e-4 and the learning rate of the second sub-network model being 5e-5.

7. A multi-target laser pointing system, characterized in that, The control method of the multi-target laser pointing system as described in any one of claims 1 to 6, wherein the multi-target laser pointing system comprises: Both the first and second camera components are configured to capture video of the target under test; A laser is disposed between the first camera assembly and the second camera assembly, and the laser is configured to emit laser light; A galvanometer assembly is disposed in the optical path of the laser, and the galvanometer assembly is configured to rotate relative to the laser to transmit the laser to the target under test; the optical axis of the second camera assembly is coaxially arranged with the 0-point optical axis of the galvanometer assembly and the optical axis of the laser. A beam splitter is disposed in the optical path of the laser, and the beam splitter is located between the galvanometer assembly and the second camera assembly; the beam splitter is configured to reflect the laser to the galvanometer assembly; A processor is connected to the first camera assembly, the second camera assembly, the laser, and the galvanometer assembly. The processor is configured to adjust the rotation angle of the galvanometer assembly so that the laser spot of the laser deflected by the galvanometer assembly coincides with the coordinates of the target under test.

8. The multi-target laser pointing system according to claim 7, characterized in that, The wavelength of the laser is 1572nm; the repetition frequency is 5kHz.

9. The multi-target laser pointing system according to claim 7 or 8, characterized in that, Also includes: A receiver is disposed on the side of the second camera assembly away from the beam splitter, and the receiver is configured to receive the reflected light of the laser beam after passing through the target under test.

Images