A high-precision three-dimensional vision system based on multi-camera time-sharing acquisition to improve acquisition frequency

Abstract

The application discloses a high-precision three-dimensional vision system based on multi-camera time-sharing acquisition to improve acquisition frequency, and relates to the technical field of three-dimensional vision, comprising: a time-sharing synchronous control module, which is used for generating and outputting a group of trigger signals with precise time sequence offset; an image acquisition unit array, which is composed of at least three spatially distributed image acquisition units, each unit responds to the trigger signal, and sequentially opens its image sensor for image acquisition within a unified system acquisition cycle, and the acquisition windows of each unit do not overlap in time; a time-sharing acquisition cooperative scheduling module, which is used for dynamically solving and assigning the mutually exclusive acquisition window parameters of each image acquisition unit within the acquisition cycle according to scene motion speed, illumination change and system state. The application solves the problem that the existing three-dimensional vision system is difficult to consider high frequency and high precision use scene.

Description

Technical Field

[0001] This invention relates to the field of 3D vision technology, specifically a high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency. Background Technology

[0002] There is a significant demand for high-frequency, high-precision real-time measurement of the three-dimensional shape of moving objects in fields such as industrial inspection, robot guidance, and motion analysis. Therefore, three-dimensional vision systems are required. However, existing three-dimensional vision systems still have some shortcomings.

[0003] Chinese patent application CN113496523A discloses a system and method for a 3D calibration vision system, eliminating the need for accurate pre-calibration of the target. The system and method acquire images of multi-layered 3D calibration targets at different spatial locations and times, and calculate the orientation difference of the 3D calibration target between two acquisitions. This technology can be used to perform vision-based inspection and monitoring of single-plane orientation repeatability. By applying this technology to an assembly work plane, vision-based inspection and monitoring of the orientation repeatability of the assembly work plane can be performed. Combined with a moving robot end effector, this technology can provide vision-based inspection and monitoring of the robot end effector orientation repeatability. Visual guidance adjustment of two planes can be achieved to achieve parallelism. The system and method operate to perform precise vision-guided robot settings to achieve parallelism between the robot end effector and the assembly work plane.

[0004] While the aforementioned systems and methods can achieve visually guided adjustment, they struggle to meet the demands of high-frequency and high-precision applications. Existing technologies, employing a single high-speed camera, cannot directly acquire 3D information, limiting measurement accuracy and completeness. When using active 3D sensors such as structured light, their hardware frame rates are low, making it difficult to meet the intensive sampling requirements of high-speed motion processes such as vibration analysis and high-speed assembly. Summary of the Invention

[0005] The purpose of this invention is to provide a high-precision 3D vision system based on multi-camera time-division acquisition to improve the acquisition frequency, in order to solve the problem that existing 3D vision systems are unable to simultaneously handle high-frequency and high-precision application scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency, comprising:

[0007] The time-sharing synchronization control module is used to generate and output a set of trigger signals with precise timing offset;

[0008] The image acquisition unit array consists of at least three spatially distributed image acquisition units, each responding to the trigger signal within a unified system acquisition cycle. Inside, the image sensors of each unit are turned on sequentially to acquire images, and the acquisition windows of each unit do not overlap in time.

[0009] The time-sharing acquisition and collaborative scheduling module is used to dynamically solve and allocate mutually exclusive acquisition window parameters for each image acquisition unit within the acquisition cycle based on scene movement speed, illumination changes, and system status. The start time is... Duration is The solution process uses a window utility function that combines data quality, motion blur, and temporal coordination. For the goal;

[0010] The time-aware 3D reconstruction module is used to receive image sequences with precise timestamps output by each unit. When performing feature matching and triangulation, this module explicitly compensates for the micro-time differences caused by time-division acquisition and finally fuses them to generate a high frame rate 3D point cloud.

[0011] The point cloud refinement and verification module is used to clean and optimize the initially reconstructed point cloud based on geometric and topological consistency, and output a 3D model with quality assessment information.

[0012] As a further aspect of the present invention: the time-sharing acquisition and collaborative scheduling module determines the acquisition window parameters by solving the following optimization problem: , in, This represents the total number of image acquisition units. For scene motion velocity estimation, This is for illumination estimation.

[0013] As a further aspect of the present invention: the window utility function Constructed as: , in, For the term characterizing the signal-to-noise ratio of an image, These are the weighting coefficients. >1, the second term is a nonlinear penalty term designed to address the cumulative effect of motion fuzziness under time-division acquisition.

[0014] As a further aspect of the present invention: the time-aware 3D reconstruction module includes a motion-guided feature matching unit, which, when performing matching, for features at time... Images acquired The feature points in the data, at another time Images acquired When searching for feature points, the search area is adjusted from a fixed neighborhood centered on the point projection to a strip-shaped area extending along the predicted motion trajectory.

[0015] As a further aspect of the present invention: the time-series-aware 3D reconstruction module further includes a time-varying point cloud fusion unit, which fuses point clouds at different times... Preliminary calculation of the three-dimensional point set When merging into a unified model, a valid timestamp is added to each point, for points that are at different times due to motion. The time-varying point cloud fusion unit merges the same physical point that is repeatedly observed into a three-dimensional point trace and records the sequence of its position change over time.

[0016] As a further aspect of the present invention, the system also includes an acquisition consistency self-test module. The acquisition consistency self-test module analyzes the stability of images acquired by the same image acquisition unit in the same phase within multiple consecutive acquisition cycles, monitors long-term drift of time-division synchronization or performance jitter of a single camera, and triggers recalibration or alarm when an anomaly is detected.

[0017] As a further aspect of the present invention, the specific monitoring indicator of the acquisition consistency self-test module is the variance of the consistency measure of feature point positions or grayscale distribution between images of the same phase.

[0018] As a further aspect of the present invention, the time-division synchronization control module includes an anti-interference signal synthesis unit.

[0019] As a further aspect of the present invention: when performing optimization, the time-sharing acquisition and collaborative scheduling module not only considers maximizing the utility of the current acquisition cycle, but also adopts a simple rolling time-domain optimization strategy: that is, when solving the window parameters of the current cycle, the expected acquisition requirements of some cameras in the next cycle are taken into consideration as soft constraints.

[0020] As a further aspect of the present invention: the system ultimately reconstructs each three-dimensional point Calculate and attach a time-sharing reconstruction confidence score, which includes a temporal consistency factor and a motion compatibility factor. The temporal consistency factor is based on the concentration of timestamps from multiple images used to reconstruct the point. The motion compatibility factor is based on whether the positional changes of the point and its neighbors at multiple times conform to a consistent local motion model.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. By employing a time-division acquisition architecture, multiple cameras are triggered sequentially along the timeline, enabling the system to acquire multiple frames of images with interleaved timestamps within a single acquisition cycle. The system's equivalent output frame rate for 3D point clouds can reach N times the frame rate of a single camera (where N is the number of cameras), breaking through the frame rate bottleneck of single-camera or synchronous multi-camera systems and enabling the capture of continuous attitude changes of high-speed moving objects.

[0023] 2. Through a time-sharing acquisition and collaborative scheduling module and optimization model, the system can dynamically allocate the optimal acquisition window to each camera. The introduced nonlinear motion blur penalty term actively suppresses image degradation caused by high-speed motion, improving the quality of single-frame images from the source. Through a motion-guided feature matching unit, the algorithm uses optical flow to predict the position of feature points after micro-hour differences, reducing the 2D matching search range from a surface to a strip region, improving the success rate and accuracy of feature matching in temporally inconsistent images. Through a time-varying point cloud fusion unit, the system can output 3D point traces containing motion trajectories, rather than blurred static point clouds, more realistically reflecting dynamic scenes and providing direct data support for motion analysis.

[0024] 3. Through the data acquisition consistency self-check module, the system can monitor long-term drift of the synchronization signal and changes in camera performance online, providing timely warnings or triggering calibration to ensure the long-term stable operation of the time-sharing mechanism and enable reliable application in industrial settings. Through the anti-interference signal synthesis unit, the system can proactively adapt to complex industrial electromagnetic environments, generating trigger sequences with stronger anti-interference capabilities, ensuring the accuracy and stability of time-sharing synchronization. The rolling time-domain optimization strategy ensures a smooth and stable scheduling scheme, avoiding drastic fluctuations in camera resource allocation and guaranteeing the continuity and stability of the 3D data stream. Through the time-sharing reconstruction confidence assessment mechanism, the system can provide quantified reliability indicators for each 3D point. Downstream applications such as precision measurement and defect interpretation can use this confidence level to filter or weight data, thereby increasing the frequency while ensuring the accuracy and reliability of the final output results. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of the present invention;

[0026] Figure 2 This is a schematic diagram of the time-sharing synchronization control module of the present invention;

[0027] Figure 3 This is a schematic diagram of the time-sharing acquisition and collaborative scheduling module of the present invention;

[0028] Figure 4 This is a flowchart of the time-aware 3D reconstruction process of the present invention;

[0029] Figure 5 This is a schematic diagram illustrating the quality monitoring and quality assessment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1:

[0032] like Figures 1-5 As shown, this embodiment proposes a high-precision 3D vision system based on multi-camera time-division acquisition to improve the acquisition frequency, including:

[0033] The time-sharing synchronization control module is used to generate and output a set of trigger signals with precise timing offset;

[0034] An image acquisition unit array consists of at least three spatially distributed image acquisition units, each responding to a trigger signal within a unified system acquisition cycle. Inside, the image sensors of each unit are turned on sequentially to acquire images, and the acquisition windows of each unit do not overlap in time.

[0035] The time-sharing acquisition and collaborative scheduling module is used to dynamically solve and allocate mutually exclusive acquisition window parameters for each image acquisition unit within its acquisition cycle based on scene movement speed, illumination changes, and system status. The start time is... Duration is The solution process uses a window utility function that combines data quality, motion blur, and temporal coordination. For the goal;

[0036] The time-aware 3D reconstruction module is used to receive image sequences with precise timestamps output by each unit. When performing feature matching and triangulation, this module explicitly compensates for the micro-time differences caused by time-division acquisition and finally fuses them to generate a high frame rate 3D point cloud.

[0037] The point cloud refinement and verification module is used to clean and optimize the initially reconstructed point cloud based on geometric and topological consistency, and output a 3D model with quality assessment information.

[0038] By employing time-sharing synchronous control and collaborative scheduling, multiple cameras are forced to acquire data at staggered times, increasing the equivalent sampling frequency of the hardware system to N times that of a single camera (where N is the number of cameras). Time-series-aware 3D reconstruction is crucial for addressing the time difference introduced by time-sharing acquisition, ensuring that high-frequency acquired data can be accurately fused into a high-precision 3D model. The point cloud refinement and verification module is responsible for improving the quality and reliability of the output model.

[0039] Example 2:

[0040] The solution in Example 1 will be further described below with reference to its specific working method.

[0041] As a preferred implementation, based on the above method, the time-sharing acquisition and collaborative scheduling module further determines the acquisition window parameters by solving the following optimization problem: , in, This represents the total number of image acquisition units. For scene motion velocity estimation, This is for illumination estimation.

[0042] The scheduling problem is formalized as a constrained optimization model. The objective function seeks to maximize the overall utility. The first constraint ensures that each acquisition window is within the period. The second constraint is the core physical limitation of time-sharing acquisition, which forces that the effective acquisition periods of all cameras be completely separated in time and do not overlap with each other, so as to achieve interference-free time-sharing acquisition and improve the overall frame rate of the system.

[0043] As a preferred implementation, based on the above method, the window utility function is further... Constructed as: , in, For the term characterizing the signal-to-noise ratio of an image, These are the weighting coefficients. >1, the second term is a nonlinear penalty term designed to address the cumulative effect of motion fuzziness under time-division acquisition.

[0044] Unlike traditional exposure control, this system takes into account the characteristic that objects may move between cameras in time-division acquisition scenarios. It enables the scheduling algorithm to actively select shorter and more frequent acquisition windows in high-speed scenarios, thereby achieving an optimal balance between motion blur and signal strength and ensuring the image clarity of moving objects.

[0045] In a preferred embodiment, based on the above method, the time-aware 3D reconstruction module further includes a motion-guided feature matching unit. When performing matching, the motion-guided feature matching unit, for the feature at time... Images acquired The feature points in the data, at another time Images acquired When searching for feature points, the search area is adjusted from a fixed neighborhood centered on the point projection to a strip-shaped region extending along the predicted motion trajectory. The predicted motion trajectory is based on... The local optical flow field and time difference calculated in the image ( This is obtained through extrapolation. Traditional methods search within a fixed neighborhood and are prone to matching failures when objects are moving. This scheme introduces motion guidance, using optical flow to predict the movement trajectory of feature points within a time difference, upgrading the two-dimensional search to a strip-shaped region search along the trajectory. This significantly improves the success rate and accuracy of time-spanning image feature matching in moving scenes.

[0046] As a preferred embodiment, based on the above method, the time-series-aware 3D reconstruction module further includes a time-varying point cloud fusion unit, which fuses point clouds from different times... Preliminary calculation of the three-dimensional point set When merging into a unified model, a valid timestamp is added to each point, for points that are at different times due to motion. The time-varying point cloud fusion unit merges the same physical point that is repeatedly observed into a three-dimensional point trace and records the sequence of its position changes over time, rather than a static point, thereby directly outputting a dynamic three-dimensional representation containing motion information.

[0047] Traditional systems output static snapshot-style point clouds, while this system features time-division multiplexing and high frame rate acquisition. Leveraging this advantage, the positions of the same physical point at different times are correlated into a continuous point trace. The system's final output is not a model of a single instant, but a model that includes the object's motion history, providing a direct data foundation for advanced applications such as high-speed motion analysis and trajectory prediction.

[0048] As a preferred embodiment, based on the above method, the system further includes an acquisition consistency self-test module. The acquisition consistency self-test module analyzes the stability of images acquired by the same image acquisition unit at the same phase (i.e., the relatively same time in each cycle) within multiple consecutive acquisition cycles, monitors long-term drift of time-division synchronization or performance jitter of a single camera, and triggers recalibration or alarm when an anomaly is detected.

[0049] The accuracy of time-division acquisition is highly dependent on the long-term stability of clock synchronization. The system's self-monitoring mechanism ensures that the static scene captured by each camera at the same relative moment in each acquisition cycle should be exactly the same. By continuously comparing the consistency of these images, problems such as slight changes in camera pose caused by clock drift or vibration, or performance degradation of individual cameras can be keenly detected, enabling predictive maintenance and ensuring long-term stable operation of the system in complex industrial environments.

[0050] As a preferred implementation, based on the above method, the specific monitoring indicator of the acquisition consistency self-test module is the variance of the consistency measure of feature point positions or grayscale distribution between images with the same phase. If the variance exceeds the threshold, the synchronization state is determined to be abnormal.

[0051] Using the variance of feature point locations can directly and sensitively detect minute changes in the camera's optical center or attitude; using variance measures of grayscale distribution consistency, such as histogram correlation and structural similarity (SSIM), can detect problems such as light source stability, sensor gain drift, or lens smudges.

[0052] As a preferred embodiment, based on the above method, the time-division synchronization control module further includes an anti-interference signal synthesis unit. The anti-interference signal synthesis unit does not simply superimpose random jitter on the clock signal, but dynamically generates a set of pseudo-random trigger edge sequences orthogonal to the frequency spectrum of electromagnetic interference in the current main environment, so that the energy of the synthesized trigger signal in the key frequency band is minimized, thereby reducing the probability of being interfered with.

[0053] The system first analyzes the main frequency components of interference in the environment, and then purposefully generates a trigger signal so that its energy spectrum is orthogonal to the interference frequency band, thereby effectively avoiding interference and is more targeted than traditional methods.

[0054] As a preferred implementation, based on the above method, the time-sharing acquisition and collaborative scheduling module, when performing optimization, not only considers maximizing the utility of the current acquisition cycle, but also adopts a simple rolling time-domain optimization strategy: that is, when solving the window parameters of the current cycle, the expected acquisition demand of some cameras in the next cycle is taken into consideration as a soft constraint, so as to avoid drastic fluctuations in the scheduling scheme of adjacent cycles, which would cause some cameras to be allocated to unfavorable acquisition times for a long time.

[0055] In existing technologies, greedy algorithms only optimize the current cycle, which can lead to some cameras being scheduled to collect data in multiple consecutive cycles at times with poor lighting or severe motion blur, affecting their lifespan and data balance. The introduction of rolling temporal optimization, which uses the subsequent limited 1-2 cycles as a preview window for overall planning, although slightly increases the amount of computation, can ensure a smooth transition of the scheduling scheme in time, improving the overall performance and fairness of the system.

[0056] As a preferred embodiment, based on the above method, the system further refines the process for each reconstructed 3D point. Calculate and attach a time-sharing reconstruction confidence score, which includes a temporal consistency factor and a motion compatibility factor. The temporal consistency factor is based on the concentration of timestamps from multiple images used to reconstruct the point. The motion compatibility factor is based on whether the positional changes of the point and its neighbors at multiple times conform to a consistent local motion model. Points with low confidence will be filtered or marked in subsequent applications.

[0057] The time consistency factor assesses whether the evidence on which the point is reconstructed is concentrated in time. If the timestamps are scattered, it means that the point may be reconstructed from the mixed positions of objects at different times, resulting in low confidence. The motion compatibility factor assesses whether the motion of the point is consistent with the motion of its surrounding points. If they are inconsistent, it may be a matching error or a noise point. This enables downstream applications such as precision measurement and defect detection to identify and trust high-quality data and automatically remove suspicious data, thereby ensuring the accuracy and reliability of the final results in practical applications.

[0058] Example 3:

[0059] The solutions in Embodiments 1 and 2 will be further described below with reference to their specific working methods.

[0060] Specifically, in a high-precision 3D vision system based on multi-camera time-sharing acquisition to increase the acquisition frequency, each industrial camera completes power-on and self-test, and the time-sharing synchronization control module initializes, preparing to generate a high-precision trigger signal sequence. Simultaneously, the system utilizes the first acquisition cycle or pre-scan to quickly analyze the initial scene and obtain a preliminary estimate of the scene's motion velocity. and ambient light level These parameters will serve as optimization inputs.

[0061] In each system data collection cycle Initially, the time-sharing data acquisition and collaborative scheduling module begins operation. This module aims to maximize the overall data acquisition efficiency of the system by solving a constrained optimization problem, specifically the window utility function within the optimization problem. The signal-to-noise ratio was taken into account. and nonlinear penalty for motion fuzziness To ensure that shorter acquisition windows are automatically allocated to cameras in high-speed scenarios and to suppress blur, the module employs a rolling temporal optimization strategy. When optimizing parameters for the current cycle, it proactively considers the needs of the next cycle, ensuring the stability of the scheduling scheme and preventing certain cameras from being at a prolonged acquisition disadvantage. After solving the problem, the module outputs a set of optimal acquisition window parameters. Each window strictly adheres to the constraint of non-overlapping time. The time-sharing synchronization control module receives window parameters from the scheduling module. Its internal anti-interference signal synthesis unit dynamically generates a set of pseudo-random trigger edge sequences orthogonal to the current environmental electromagnetic interference spectrum, using these to modulate the basic clock signal and synthesize a final trigger signal with strong anti-interference capability. This module is based on... Precisely at the moment To the Each image acquisition unit sends a trigger signal to control its... The sensor is activated to acquire images within a specified time period. In this way, each unit in the image acquisition unit array periodically... The system is sequentially awakened and put into hibernation, capturing a sequence of images of the same dynamic scene from different perspectives and at different minute moments, and stamping them with precise timestamps.

[0062] The time-aware 3D reconstruction module first performs feature extraction. In the feature matching stage, its motion-guided feature matching unit matches features for each time step. When matching images, feature points in the image no longer undergo global or fixed neighborhood search, but are instead determined based on... The local optical flow field calculated from the image predicts the search region as a band-shaped area extending along the direction of motion, improving the matching speed and accuracy in dynamic scenes. After matching is completed, the module initially calculates a series of 3D points through triangulation. Subsequently, the time-varying point cloud fusion unit begins to work. It does not simply merge all points into a static model, but identifies and associates them at different times. The observed physical point is fused into a time-varying 3D point trace, directly outputting a dynamic 3D representation containing object motion information, with continuous motion state of the object acquired in a time-division manner. The initially reconstructed point cloud is then sent to the point cloud refinement and verification module for optimization. This module filters and optimizes based on geometric consistency such as surface smoothness and multi-view photometric consistency, removing obvious outliers.

[0063] Meanwhile, the data acquisition consistency self-check module continues to run, comparing images acquired at the same relative phase time across different acquisition cycles (e.g., the first camera image triggered in each cycle) to calculate the variance of feature point positions or grayscale distribution. If the variance increases abnormally, it indicates potential drift in time-division synchronization or instability in the performance of a particular camera, triggering an alarm or initiating a recalibration process. Ultimately, the system calculates the variance for each output 3D point. A time-division reconstruction confidence score is calculated, which is jointly determined by a temporal consistency factor reflecting the concentration of timestamps in the images used for reconstruction and a motion compatibility factor reflecting the degree of agreement between the motion model of the point and that of surrounding points. Low-confidence points can be marked or filtered, thereby providing high-quality, reliable 3D data for subsequent applications such as precision measurement and defect detection.

[0064] The accuracy assessment results of 3D reconstruction, the confidence distribution of point clouds, and the status information of the self-testing module can all be used as feedback signals and input into the time-sharing acquisition and collaborative scheduling module to dynamically adjust the weight coefficients in the optimization problem. Alternatively, the scheduling strategy can be changed to enable the system to continuously adapt to complex and changing industrial site environments.

[0065] The above description is merely 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 high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency, characterized in that, include: The time-sharing synchronization control module is used to generate and output a set of trigger signals with precise timing offset; The image acquisition unit array consists of at least three spatially distributed image acquisition units, each responding to the trigger signal within a unified system acquisition cycle. Inside, the image sensors of each unit are turned on sequentially to acquire images, and the acquisition windows of each unit do not overlap in time. The time-sharing acquisition and collaborative scheduling module is used to dynamically solve and allocate mutually exclusive acquisition window parameters for each image acquisition unit within the acquisition cycle based on scene movement speed, illumination changes, and system status. The start time is... Duration is The solution process uses a window utility function that combines data quality, motion blur, and temporal coordination. For the goal; The time-aware 3D reconstruction module is used to receive image sequences with precise timestamps output by each unit. When performing feature matching and triangulation, this module explicitly compensates for the micro-time differences caused by time-division acquisition and finally fuses them to generate a high frame rate 3D point cloud. The point cloud refinement and verification module is used to clean and optimize the initially reconstructed point cloud based on geometric and topological consistency, and output a 3D model with quality assessment information.

2. The high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, The time-sharing acquisition and collaborative scheduling module determines the acquisition window parameters by solving the following optimization problem: ， in, This represents the total number of image acquisition units. For scene motion velocity estimation, This is for illumination estimation.

3. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, The window utility function Constructed as: ， in, For the term characterizing the signal-to-noise ratio of an image, These are the weighting coefficients. >1, the second term is a nonlinear penalty term designed to address the cumulative effect of motion fuzziness under time-division acquisition.

4. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, The time-aware 3D reconstruction module includes a motion-guided feature matching unit. When performing matching, the motion-guided feature matching unit matches features at time... Images acquired The feature points in the data, at another time Images acquired When searching for feature points, the search area is adjusted from a fixed neighborhood centered on the point projection to a strip-shaped area extending along the predicted motion trajectory.

5. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency, as described in claim 1 or 4, characterized in that, The time-series-aware 3D reconstruction module also includes a time-varying point cloud fusion unit, which fuses point clouds from different times... Preliminary calculation of the three-dimensional point set When merging into a unified model, a valid timestamp is added to each point, for points that are at different times due to motion. The time-varying point cloud fusion unit merges the same physical point that is repeatedly observed into a three-dimensional point trace and records the sequence of its position change over time.

6. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, The system also includes an acquisition consistency self-test module. The acquisition consistency self-test module analyzes the stability of images acquired by the same image acquisition unit in the same phase within multiple consecutive acquisition cycles, monitors long-term drift of time-division synchronization or performance jitter of a single camera, and triggers recalibration or alarm when an anomaly is detected.

7. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, The specific monitoring indicator of the acquisition consistency self-test module is the variance of the consistency measure of feature point positions or grayscale distribution between images with the same phase.

8. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, The time-division synchronization control module includes an anti-interference signal synthesis unit.

9. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, When performing optimization, the time-sharing acquisition and collaborative scheduling module not only considers maximizing the utility of the current acquisition cycle, but also adopts a simple rolling time-domain optimization strategy: that is, when solving the window parameters of the current cycle, the expected acquisition requirements of some cameras in the next cycle are taken into consideration as soft constraints.

10. A high-precision 3D vision system based on multi-camera time-division acquisition to improve acquisition frequency according to claim 1, characterized in that, The system ultimately reconstructs each 3D point Calculate and attach a time-sharing reconstruction confidence score, which includes a temporal consistency factor and a motion compatibility factor. The temporal consistency factor is based on the concentration of timestamps from multiple images used to reconstruct the point. The motion compatibility factor is based on whether the positional changes of the point and its neighbors at multiple times conform to a consistent local motion model.

Images