Three-dimensional vision measurement system based on multi-view structured light

Abstract

The invention relates to the technical field of vision measurement, and discloses a three-dimensional vision measurement system based on multi-view structured light, and the system comprises a data collection module which obtains image data, system data and point cloud data through a sensor and a system; the synchronous link module is used for generating a synchronous pulse interval based on the system data; calculating a cross correlation coefficient to judge interference and finish construction of a synchronous link; the three-dimensional measurement module is used for adjusting and calibrating by utilizing a composite structure, adjusting brightness through double modulation, constructing a target section data representation model for iterative optimization, analyzing to obtain an absolute phase value, constructing an energy function, obtaining a matching point pair based on image data and calculating a depth error; the data architecture module is used for dividing point cloud data, carrying out gridding discretization on an integral region, carrying out parallel calculation on energy values, summarizing the energy values, and dynamically adjusting the order of multi-pole expansion; and the enhanced anti-interference module is used for carrying out anti-interference processing on the image data, adjusting measurement parameters and adjusting the projection light intensity based on the predicted light intensity.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional vision measurement technology, and more specifically to a three-dimensional vision measurement system based on multi-view structured light. Background Technology

[0002] 3D vision measurement is an advanced measurement technology that is based on optical imaging and uses computer vision and geometric measurement principles to acquire three-dimensional information about objects. The working principle of 3D vision measurement is mainly to simulate human eyes and use multi-angle image data to reconstruct the spatial structure of the object in order to obtain the three-dimensional shape information of the object being measured.

[0003] In existing technologies, traditional structured light methods are limited by fluctuations in projection brightness, large cumulative errors in phase unwrapping, manual calibration, and slow convergence. Traditional K-nearest neighbor normal vector estimation can obtain the normal vector information of each point, but the error is large in sparse point cloud regions and computational redundancy in dense regions. In dynamic scenes, sudden changes in ambient light can easily cause image overexposure. Traditional filtering algorithms can resist the interference of environmental factors such as changes in illumination and noise on the image, but when faced with irregular parts, they can easily lead to problems such as few labeled samples in the target point cloud, missing information between points, high computational complexity of the segmentation model, slow inference speed, irregular geometric shape, and difficulty in fitting curved surfaces.

[0004] Multi-view structured light method is a high-precision 3D measurement method that combines multi-view vision and structured light technology. It utilizes multiple cameras and projectors working collaboratively to acquire 3D information of objects through structured light projection and capture. Compared to traditional structural methods, multi-view structured light method can cover a larger measurement range and significantly reduce environmental interference with the measurement results. However, this method requires multiple cameras and structured light projectors to work together, demanding high timing synchronization accuracy. Signal interference or synchronization deviations can lead to phase errors. Simultaneous calibration of camera intrinsic and extrinsic parameters, structured light projector pose, and relative positions among multiple cameras can easily cause calibration error accumulation, reducing the accuracy of 3D measurement. In multi-view structured light methods, the traditional Gray code method can ensure accurate recognition of coded patterns by multiple cameras in complex environments. However, due to the need for multi-frame projection, motion artifacts are easily generated in dynamic scenes. The data fusion computation of multi-view systems is large, making it difficult to meet the requirements of high-speed detection. In strong light environments, the signal-to-noise ratio of structured light drops significantly, easily leading to overlapping shadow areas from different viewpoints, and significant error amplification at depth discontinuities. Furthermore, changes in ambient temperature can easily cause thermocouple signal drift, and the static compensation model cannot adapt to rapid environmental changes, resulting in the accumulation of ADC conversion errors and timing deviations exceeding the preset range, affecting synchronization accuracy. In shared frequency bands, it is susceptible to co-frequency interference, cannot dynamically cope with sudden strong interference, and has a high packet loss rate.

[0005] In view of this, the present invention provides a three-dimensional vision measurement system based on multi-view structured light, which aims to solve the problems of brightness fluctuation and phase error accumulation in traditional structured light projection, and reduce environmental interference, computational complexity and insufficient accuracy. Summary of the Invention

[0006] The purpose of this invention is to provide a three-dimensional vision measurement system based on multi-view structured light. To solve the aforementioned problems in the prior art, this invention achieves this through the following technical solution: The multi-view structured light three-dimensional vision measurement system provided in this embodiment of the invention specifically includes the following modules: Data collection module: Acquires image data, system data, and point cloud data through sensors and systems; Synchronous Link Module: Performs temperature drift compensation based on system data, synchronously outputs two pulse width modulation (PWM) signals, analyzes and determines strong interference, and triggers a dynamic reassembly mechanism if it is strong interference. 3D measurement module: It uses a composite structure for adjustment and calibration, adjusts brightness through dual modulation, constructs a target cross-section data representation model, obtains matching point pairs based on image data, calculates depth error and performs iterative optimization, analyzes and obtains absolute phase value, constructs energy function and summarizes the calculation. Data architecture module: Divide the point cloud data, discretize the integration region into a grid, calculate and summarize the energy values ​​in parallel; compensate for phase jump errors and depth distortion based on an adaptive error control mechanism, and dynamically adjust the order of the multipole expansion; Enhanced anti-interference module: Performs anti-interference processing on image data, adjusts measurement parameters, and adjusts the projected light intensity based on the predicted light intensity.

[0007] The beneficial effects of this invention are: 1. Utilizing multiple cameras and projectors working collaboratively, this method acquires 3D information of objects through structured light projection and capture, covering a wider measurement range, reducing environmental interference with measurement results, improving timing synchronization accuracy, and minimizing signal interference or synchronization deviation. It also reduces calibration error accumulation and improves 3D measurement accuracy. A temperature-timing closed-loop compensation system is formed using thermocouples, an FPGA-built-in ADC, and a PID algorithm to dynamically bind ambient temperature and timer parameters, breaking through the traditional static compensation mode. A triple-combination mechanism of crystal oscillator clock source, wireless frequency band time slot allocation, and synchronization pulse calibration is used to construct a low-conflict synchronization network of heterogeneous nodes. Strong interference detection based on cross-correlation coefficients and dynamic channel switching ensure data transmission stability. Combining pulse width-amplitude dual modulation technology and curvature adaptive phase unwrapping algorithm solves the problems of brightness fluctuation and phase error accumulation in traditional structured light projection. 2. A hybrid optimization approach combining sparse Jacobian matrix approximation and the trust region-steepest descent method is employed to address the non-convexity problem in camera-projector calibration. A lightweight CNN model and a parallel optical flow calculation engine based on row buffering and gradient covariance matrix are embedded to achieve precise and rapid measurement of the surface morphology of non-standard irregular parts. Octree-based block statistical point cloud density is used to dynamically adjust the multipole expansion order, enabling rapid and accurate segmentation of the part's point cloud. Median filtering and Kalman filtering are cascaded and combined with an LSTM network to predict light intensity sequences, achieving adaptive compensation for ambient light intensity. A list of idle channels based on wireless spectrum scanning enables real-time switching of interference channels and reconstruction of time slot tables, ensuring accurate recognition of coded patterns by multiple cameras in complex environments. Multi-frame projection adapts to rapid environmental changes in dynamic scenes, reducing ADC conversion error accumulation and timing deviations, and increasing synchronization accuracy. In shared frequency bands, it is susceptible to co-channel interference; dynamic responses to sudden strong interference reduce packet loss rate. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the structure of the three-dimensional vision measurement system based on multi-view structured light provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of the time slot allocation and synchronization process of the synchronization link module system node of the three-dimensional vision measurement system based on multi-view structured light provided in Embodiment 1 of the present invention. Figure 3 This is a flowchart of the enhanced anti-interference module of the three-dimensional vision measurement system based on multi-view structured light provided in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the structural working principle of a 3D vision measurement system; Figure 5 This is a cross-sectional diagram of a 3D vision measurement system. Figure 1 ; Figure 6 This is a cross-sectional diagram of a 3D vision measurement system. Figure 2 ; Figure 7 This is a cross-sectional diagram of a 3D vision measurement system. Figure 7 . Detailed Implementation

[0010] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0011] Example 1 like Figure 1 As shown in the embodiment of the present invention, the multi-view structured light three-dimensional vision measurement system specifically includes the following modules: Data collection module: Acquires image data, system data, and point cloud data through sensors and systems; Temperature signals are acquired in real time by thermocouple sensors, converted into digital signals by the built-in analog-to-digital converter (ADC) module of the FPGA, and dynamically calculated by a PID algorithm, which records the compensation value to a register to obtain system data. System data includes, but is not limited to: ambient temperature value, PID compensation value, and projection frame rate. , mode switching cycle number k, timer parameter adjustment record; It should be noted that FPGA stands for Field Programmable Gate Array, which is an integrated circuit that can be programmed to implement specific digital circuit functions; PID algorithm is a classic control algorithm based on feedback mechanism, which adjusts the output by calculating the proportional, integral and derivative terms of the error, so that the system can stably reach the target state. The FPGA timer module generates pulse parameters, and the on-chip logic analyzer (ILA) captures waveform characteristics in real time to obtain pulse interval, timing deviation and PWM signal phase difference. It should be noted that PWM signal stands for Pulse Width Modulation signal, which is a technique that controls the average output power by adjusting the duty cycle of the pulse signal; The wireless module scans the spectrum, the FPGA storage unit saves the time slot allocation table, and the SPI protocol distributes the data to the system nodes. Based on the octree block statistical density, the normal vector is output through PointNet inference, and the FMM solver records the order adjustment log to obtain point cloud data. The point cloud data includes, but is not limited to: point cloud density distribution map, normal vector correction amount and multipole expansion order. It should be noted that PointNet is a pioneering deep learning model for processing point cloud data; Acquire image data using a camera; Synchronization Link Module: Generates synchronization pulse intervals based on system data and performs temperature drift compensation; calculates cross-correlation coefficients to determine interference; if it is strong interference, it triggers a dynamic reassembly mechanism to re-synchronize and complete the construction of the synchronization link. Based on the acquired system data, the Field Programmable Gate Array (FPGA) determines the projection frame rate. Number of mode switching cycles Through formula Generate synchronization pulse interval ,in, This represents the temperature drift compensation amount, which is obtained by real-time acquisition of ambient temperature by thermocouples and dynamic adjustment of timer parameters through the built-in PID algorithm of FPGA; The FPGA synchronously outputs two pulse width modulation (PWM) signals. The first channel controls the pattern switching of the structured light projector's DLP module; It should be noted that DLP modules are an optical projection technology based on digital micromirror devices; The second path triggers the exposure of the camera's CMOS array; It should be noted that the camera's CMOS array is the core component of a modern digital camera, used to convert light signals into electrical signals and ultimately generate digital images; Clock coherence is achieved through on-chip clock manager (CMT), and timing deviation is controlled within a preset range. During the system startup phase, the 2.4GHz and 5.8GHz frequency bands are selected to determine the number of available independent frequency bands; Based on the number and requirements of the multi-view system nodes, an independent frequency band is initially allocated to each system node. The system nodes include, but are not limited to: cameras, projectors, and processing units. The signal period of each system node is divided into N time slots; Create a system node time slot allocation table to determine the working status of each system node in different time slots. The working status includes: sending data, receiving data, and idle, to avoid time domain conflicts. The node time slot allocation table is stored in the system's storage unit and sent to each system node; According to the node time slot allocation table, data transmission is carried out in the allocated frequency band within the specified time slot; For example, the camera transmits image data using the allocated frequency band in one time slot, and the projector receives the data and performs projection processing in another time slot; A crystal oscillator clock source is used to provide a unified time reference for all system nodes; System nodes periodically send synchronization pulses to calibrate their time, ensuring that the system nodes transmit and receive data in the specified time slots within each signal cycle; Based on data transmission and reception, transmission interference is detected and compensated. If strong interference is detected, a dynamic reassembly mechanism is triggered. Specifically, the FPGA periodically samples the signals on each channel; Employing a parallel computing architecture, for each pair of channels i and j, the formula is used... Calculate the cross-correlation coefficient between channel i and channel j. ,in, and This represents the signal samples from channel i and channel j. and This represents the mean of the signal samples in channel i and channel j; Based on the comparison between the calculated cross-correlation coefficient and the preset correlation threshold, it is determined whether there is strong interference between channels i and j; It should be noted that the preset relationship threshold is obtained by collecting multiple sets of channel data in an interference-free scenario, calculating the Rayleigh distribution of the cross-relation coefficient, and using the upper limit of the 99% confidence interval. If they exist, record the channel pairs with strong interference and related information about them. Based on the obtained channel pairs with strong interference, a dynamic recombination mechanism is triggered; The system queries the list of currently available free channels via the wireless module; Based on the cross-correlation coefficient, the channel with the strongest signal strength and the least interference is selected, and system nodes on channel pairs with strong interference are switched to idle channels. For example, if the channel where node A is located is subject to strong interference, the system selects an idle channel for it and sends a switching command to system node A through the wireless module; After switching frequency bands, the system node time slot allocation table is updated to adapt to the new frequency band and channel conditions; Recalculate the time slot allocation for each system node in the new frequency band to ensure that no new time domain conflicts are generated; The updated system node time slot allocation table is stored in the system's storage unit and distributed to each system node; After receiving the new node time slot allocation table, each system node adjusts its working status according to the new time slot allocation. The system nodes perform synchronization operations again to ensure that each system node can transmit and receive data under the new frequency band and time slot allocation, thus completing the synchronization link; 3D Measurement Module: After the synchronous link is completed, the composite structure is used for adjustment and calibration. After calculating the LED grayscale value, the brightness is adjusted by dual modulation. The target cross-section data representation model is constructed for iterative optimization and solution. The absolute phase value is obtained by combining curvature adaptive phase unwrapping, and the energy function is constructed. The depth error is calculated based on the matching point pairs. After the synchronization link is completed, the LED array is mounted on the calibration board based on the composite structure, a coordinate system is established to determine the coordinates of each LED, a microprism reflection unit is embedded and infrared coded markers are set at the four corners for calibration, and the wavelength is preset to 850nm. It should be noted that the unique coding characteristics of infrared coded markers facilitate camera identification and positioning; By using reinforcement learning machine learning algorithms, the system learns and trains under different ambient light conditions to automatically adjust the reference light intensity, modulation coefficient, and trigger threshold. Based on octree block partitioning, a KNN graph is constructed for each sub-block, and local geometric features are calculated through EdgeConv. Local geometric features include, but are not limited to, curvature and normal vector. Embed the lightweight Transformer module and use the self-attention mechanism to capture cross-regional correlations; Specifically, based on PointNet, prior knowledge constraints are introduced to obtain material visual features, which include, but are not limited to, part symmetry and material distribution. Low-level details and high-level semantics are fused through residual connections. Local geometric features and material visual features are encoded into a 128-dimensional joint vector; A target cross-sectional data representation model is constructed by selecting the feature subset that contributes the most to the shape fitting using the random forest algorithm. Through formula Calculate the grayscale value of each LED. ,in, For structured light amplitude, The magnitude of the rectangular function. The spatial frequency is located along the x-axis. The spatial frequency is in the y-axis direction. The array spacing is along the x-axis. The array spacing is along the y-axis. This represents the coordinates of the LED array. This indicates the coordinates of the center of the LED array. This represents the standard deviation of the Gaussian kernel, used to control the range of Gaussian decay. If it is a rectangular function, then Less than 0.5 and If less than 0.5, then the rectangular function Otherwise, the rectangle function ; For example, based on an LED at coordinates (0.2, 0.2), the spatial frequency along the x-axis is 0.1, the spatial frequency along the y-axis is 0.06, and the amplitude of the rectangular function is obtained. The value is 100, the array spacing along the x-axis is 1, the array spacing along the y-axis is 1, and the standard deviation of the Gaussian kernel is 1. The value is 2. The grayscale value of the LED at coordinates (0.2, 0.2) is calculated using the formula. for ; The LED array alternately displays phase-shifted sinusoidal stripes according to the three-frequency combination, and the FPGA projects the stripes in a time-slice polling manner. Based on polling projection, pulse width-amplitude dual modulation is used through the formula Adjusting brightness stability, among which, This represents the LED grayscale value. This indicates the preset modulation frequency. The preset adjustment coefficient; It should be noted that the preset adjustment coefficient is based on the pulse width-amplitude dual modulation principle. The relationship between LED gray value and pulse width duty cycle, amplitude gain, modulation frequency and adjustment coefficient is modeled, brightness fluctuation rate is introduced, the relationship between adjustment coefficient and system noise is derived, and finally the preset adjustment coefficient is obtained. Set camera external parameters Projector position ,in, For the camera's rotation matrix, Let be the translation vector of the camera. Let m be the pose transformation matrix of the projector at the m-th position; Based on the camera's extrinsic parameters and the projector's pose, an absolute coordinate system is established using infrared coded points. Relative optimization of visible light feature points is then performed using the formula... Construct a target cross-sectional data representation model, in which, Constrain the reprojection error between the camera projection point and the actual observation point. Let be the projection function of the camera. For a point in three-dimensional space, For camera i to a spatial point The actual observation point For the projector in the first The pose transformation matrix for each position; Through adjacent pose transformation matrix Constrain the continuity of projector motion; Iterative optimization is performed based on the target cross-section data representation model to calculate the gradient and Hessian matrix of the model and update the camera extrinsic parameters. Projector position ; Image depth maps obtained from image data are analyzed using camera extrinsic parameters. Transform to the global coordinate system to obtain a 3D point cloud set. ; Based on perspective Points in From the perspective Searching for corresponding points on the polar line Generate matching point pairs ; Based on matching point pairs Calculate the distance in three-dimensional space to obtain the depth error; Based on the calculation of depth error, the depth map is divided into multiple sub-blocks using parallel computing and block processing. The error value of each sub-block is calculated in parallel on the GPU, and the depth error of the entire depth map is obtained by summing them. Multi-scale analysis is introduced to calculate the depth consistency error at different scales. During the iteration process, the depth error of the reprojection is continuously calculated. If the depth error of the reprojection is less than 0.5 pixels, the algorithm is considered to have converged and the iteration stops; otherwise, the iteration continues. Based on iterative optimization, the first 20 iterations are approximated by a sparse Jacobian matrix, and the main diagonal elements are updated. After 20 iterations, the trust region and the steepest descent method are combined, and the step size update formula is used to accelerate the convergence to 0.5 pixel error. Upon iteration termination, output camera extrinsic parameters. Projector position The calibration results; It integrates global coarse coding and local fine coding. For global coarse coding, a low-resolution coarse Gray code is used to cover the entire scene area; for local fine coding, high-frequency stripes are superimposed on the global coarse coding. Based on global coarse coding and local fine coding, a dual-channel DMA engine is integrated through FPGA to process global coarse coding and local fine coding respectively, and spatial block mapping technology is used to reduce computational latency. Based on the micromirror flipping characteristics of DLP projectors, a driving signal is generated by combining pulse width modulation and pre-excitation pulses to improve the stability of DMD micromirrors; Error correction is performed on the XOR operation results based on a lightweight CNN model embedded in FPGA. Combining curvature adaptive phase unwrapping to compensate for phase jump error and depth distortion, through the formula The absolute phase value was calculated. ,in, This represents the weights based on local contrast. This represents the curvature compensation term estimated using the Hessian matrix; It should be noted that the Hessian matrix is ​​a second-order partial derivative matrix of a multivariable function, describing the local curvature of the function; Based on the acquired image data, an energy function incorporating temporal smoothing and spatial gradient constraints is constructed. Specifically, the motion velocity is estimated using a dense optical flow field, the energy function parameters are dynamically adjusted, and motion sensitivity is adaptively smoothed. It should be noted that the dense optical flow field is a two-dimensional vector field, which assigns a motion vector to each pixel in the image. The motion vector represents the direction and magnitude of the displacement of the corresponding pixel between adjacent frames. Based on the FPGA parallel optical flow calculation engine, the algorithm uses row buffering and gradient covariance matrix to solve quickly; the variable update of the ADMM algorithm is split into 16-bit fixed-point main calculation and 32-bit floating-point correction using dual channels, which include: main channel and correction channel. For the main channel, multiplication and addition operations are performed based on FPGA logic units; For the correction channel, residual compensation is completed by performing a correction every 10 iterations through the coprocessor. Data architecture module: Combining a lightweight PointNet network, it divides the point cloud data, discretizes the integration region into a grid, calculates energy values ​​in parallel and summarizes them; based on an adaptive error control mechanism, it dynamically adjusts the order of the multipole expansion according to the density and distribution of the point cloud. Based on the obtained point cloud data, a lightweight PointNet network is embedded based on traditional K-nearest neighbor normal vector estimation. The input is a local point cloud block, and the output is the normal vector correction. Specifically, the point cloud data is divided into multiple sub-regions, and each local point cloud block is responsible for calculating the normal vector of points within a sub-region; Based on the constructed energy function, the integration region is discretized into a grid. Using a distributed computing platform, the discretized computing tasks are distributed to multiple computing nodes for parallel computing. Each node is responsible for calculating the energy value of a portion of the grid region and summarizing the results. Based on the traditional FMM algorithm, an adaptive error control mechanism is introduced to dynamically adjust the order of the multipole expansion according to the density and distribution of the point cloud. Specifically, in sparse regions of the point cloud, the unfolding order is reduced to decrease the computational load; In dense point cloud regions, increase the unfolding order; Enhanced anti-interference module: Performs anti-interference processing on image data, learns and trains under different ambient light conditions and adjusts measurement parameters, and adjusts the projected light intensity based on the predicted light intensity; Before inputting image data into the FPGA, image data is enhanced and made more resistant to interference. Specifically, by combining median filtering and Kalman filtering, median filtering removes obvious impulse noise, while Kalman filtering smooths the data. Deploy a lightweight LSTM network in the coprocessor, take the light intensity sequence of the past 5 frames as input, predict the light intensity of the next frame, and adjust the projected light intensity based on the predicted light intensity. The technical solution of this invention is as follows: a temperature-timing closed-loop compensation system is formed by thermocouples, FPGA built-in ADC, and PID algorithm to achieve dynamic binding of ambient temperature and timer parameters, breaking through the traditional static compensation mode; a low-collision synchronization network of heterogeneous nodes is constructed by combining a crystal oscillator clock source, wireless frequency band time slot allocation, and a synchronization pulse calibration triple-combination mechanism; strong interference detection based on cross-correlation coefficient and dynamic channel switching ensure data transmission stability; and brightness fluctuations in traditional structured light projection are solved by combining pulse width-amplitude dual modulation technology and curvature adaptive phase unwrapping algorithm. To address the phase error accumulation problem, a hybrid optimization approach combining sparse Jacobian matrix approximation and the trust region-steepest descent method is employed to resolve the non-convexity issue in camera-projector calibration. A lightweight CNN model is embedded with a parallel optical flow calculation engine based on row buffers and gradient covariance matrices to achieve hardware-based algorithm implementation. Based on octree block statistical point cloud density, the order of the multipole expansion is dynamically adjusted. Median filtering and Kalman filtering are cascaded, and combined with LSTM network prediction of light intensity sequences to achieve adaptive compensation for ambient light intensity. Based on the idle channel list obtained from wireless spectrum scanning, real-time switching of interference channels and reconstruction of time slot tables are achieved.

[0012] The above provides a detailed description of one embodiment of the present invention, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. The above formulas are all dimensionless numerical calculations, and the formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world situation. The preset parameters in the formulas are set by those skilled in the art based on actual conditions and historical experience, and can be adjusted according to actual conditions. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A three-dimensional vision measurement system based on multi-view structured light, characterized in that, Includes the following steps: Data collection module: Acquires image data, system data, and point cloud data through sensors and systems; Synchronous Link Module: Performs temperature drift compensation based on system data, synchronously outputs two pulse width modulation (PWM) signals, analyzes and determines strong interference, and triggers a dynamic reassembly mechanism if it is strong interference. 3D measurement module: It uses a composite structure for adjustment and calibration, adjusts brightness through dual modulation, constructs a target cross-section data representation model, obtains matching point pairs based on image data, calculates depth error and performs iterative optimization, analyzes and obtains absolute phase value, constructs energy function and summarizes the calculation; Data architecture module: Divide the point cloud data, discretize the integral region into a grid, calculate the energy value in parallel and summarize it; The order of the multipole expansion is dynamically adjusted based on the adaptive error control mechanism to compensate for phase jump error and depth distortion. Enhanced anti-interference module: Performs anti-interference processing on image data, adjusts measurement parameters, and adjusts the projected light intensity based on the predicted light intensity.

2. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The specific method for temperature drift compensation is as follows: Based on the acquired system data, the Field Programmable Gate Array (FPGA) determines the projection frame rate. The number of mode switching cycles k is determined by the formula. Generate synchronization pulse interval ,in, This represents the amount of temperature drift compensation.

3. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The method for analyzing and determining strong interference is as follows: A parallel computing architecture is adopted. For each pair of channels i and j, the cross-correlation coefficient between channels i and j is calculated by formula. Based on the comparison between the calculated cross-correlation coefficient and the preset correlation threshold, it is determined whether there is strong interference between channels i and j. If such a pair exists, the channel pair with strong interference and its related information are recorded, triggering a dynamic reassembly mechanism.

4. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The method for adjusting and calibrating is as follows: After the synchronization link is completed, the LED array is mounted on the calibration board based on the composite structure, a coordinate system is established to determine the coordinates of each LED, a microprism reflective unit is embedded, and infrared coded markers are set at the four corners for calibration.

5. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The method for adjusting brightness stability is as follows: The grayscale value of each LED is calculated by formula, and the LED array is used to alternately display sinusoidal stripes with phase shift according to the three-frequency combination. The FPGA projects the stripes in a time-slice polling manner. Based on polling projection, pulse width-amplitude dual modulation is used through the formula Adjusting brightness stability, among which, This represents the LED grayscale value. This indicates the preset modulation frequency. This is the preset adjustment coefficient.

6. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The method for constructing the target cross-sectional data representation model is as follows: Set camera external parameters and projector pose ,in, For the camera's rotation matrix, Let be the translation vector of the camera. Let m be the pose transformation matrix of the projector at the m-th position; Based on the camera's extrinsic parameters and the projector's pose, an absolute coordinate system is established using infrared coded points. Relative optimization of visible light feature points is then performed using the formula... Construct a target cross-sectional data representation model, in which, Let be the projection function of the camera. For a point in three-dimensional space, For camera i to a spatial point The actual observation point For the projector in the first The pose transformation matrix for each position.

7. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The method for obtaining the depth error is as follows: Image depth maps obtained from image data are analyzed using camera extrinsic parameters. Transform to the global coordinate system to obtain a 3D point cloud set. ; Based on perspective Points in From the perspective Searching for corresponding points on the polar line Generate matching point pairs ; Based on matching point pairs Calculate the distance in three-dimensional space to obtain the depth error.

8. The three-dimensional vision measurement system based on multi-view structured light according to claim 7, characterized in that, The method for iterative optimization and solution is as follows: Based on the calculation of depth error, the depth map is divided into multiple sub-blocks using parallel computing and block processing. The error value of each sub-block is calculated in parallel on the GPU, and the depth error of the entire depth map is obtained by summing them up. Multi-scale analysis is introduced to calculate the depth consistency error at different scales. During the iteration process, the depth error of the reprojection is continuously calculated. If the depth error of the reprojection is less than 0.5 pixels, the algorithm is considered to have converged and the iteration stops; otherwise, the iteration continues.

9. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The method for obtaining the absolute phase value is as follows: Combining curvature adaptive phase unwrapping to compensate for phase jump error and depth distortion, through the formula The absolute phase value was calculated. ,in, This represents the weights based on local contrast. This represents the curvature compensation term estimated using the Hessian matrix.

10. The three-dimensional vision measurement system based on multi-view structured light according to claim 1, characterized in that, The method for summarizing and calculating is as follows: The point cloud data is divided into multiple sub-regions, and each local point cloud block is responsible for calculating the normal vector of points within a sub-region. Based on the constructed energy function, the integration region is discretized into a grid. Using a distributed computing platform, the discretized computing tasks are distributed to multiple computing nodes for parallel computing. Each node is responsible for calculating the energy value of a portion of the grid region and summarizing the results.

Images