A method and system for three-dimensional positioning of an array element

Abstract

The application relates to the technical field of computer vision, in particular to a three-dimensional positioning method and system of array elements, wherein the method comprises the following steps: acquiring a multi-view high-bit-depth gray image frame sequence of an array element to be measured; completing dynamic range conversion from high-bit-depth to low-bit-depth through adaptive bit-depth mapping enhancement; performing primary detection on the low-bit-depth image to obtain a candidate coded visual marker and a corresponding ROI region; performing local mapping enhancement on each ROI region to obtain an enhanced ROI image; performing secondary precision detection on the enhanced ROI image to output a target coded visual marker containing a target corner point and a single-frame pose; determining a corner point observation covariance according to the ROI dynamic range; constructing an incremental factor graph taking the single-frame pose as a state quantity and taking a re-projection error as an observation constraint, establishing an adaptive noise-weighted multi-frame joint optimization target; and performing optimization solving based on the optimization target to obtain three-dimensional coordinates of the array element in a world coordinate system. Through the application, the precision, consistency and real-time performance of three-dimensional positioning of the array element are improved.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, specifically to a method and system for three-dimensional positioning of array elements. Background Technology

[0002] In recent years, with the rapid development of computer vision and intelligent manufacturing technologies, vision-based positioning and measurement methods have been widely used in scenarios such as robot navigation, augmented reality, 3D reconstruction, and industrial measurement due to their advantages of low deployment cost, rich information content, and easy integration with automated processes. Among them, using cameras to acquire images and extract artificial markers or geometric features, and combining multi-frame image data to estimate pose and spatial structure, is an important technical route for achieving target localization and geometric calibration.

[0003] In manual labeling schemes, square-coded visual tags can provide tag IDs and corner positions in a single frame image, and further calculate tag poses, featuring fast detection speed and simple implementation. Furthermore, by deploying multiple tag boards, the robustness of the algorithm and the reliability of pose estimation in occluded scenarios can be improved. To further improve the consistency and robustness of multi-frame pose estimation, existing techniques often model pose estimation as a factor graph optimization problem, introducing observation constraints and prior constraints to achieve joint optimization of camera pose and target pose.

[0004] However, existing localization methods based on coded visual markers are still significantly constrained by image data quality and processing throughput. For example, under low-light, short-exposure, or high-resolution acquisition conditions, insufficient sensor light sensitivity can lead to increased image noise and decreased contrast, resulting in poor stability of marker edge and corner extraction. Simultaneously, when converting high-bit-depth images to general 8-bit images, simple linear compression can lead to insufficient utilization of the effective dynamic range, making it even harder to distinguish local low-contrast areas, thus reducing detection recall and corner localization accuracy. Furthermore, in large-field-of-view, high-resolution application scenarios, the large data volume of a single frame limits the overall system processing efficiency and real-time performance. Especially in environments with complex lighting and significant local contrast differences, problems such as large fluctuations in corner localization and difficulty in effectively suppressing false detections persist. Summary of the Invention

[0005] To address these issues, this invention provides a three-dimensional positioning method and system for array elements, aiming to solve the technical problems of existing technologies in high bit-depth grayscale image processing, such as insufficient separability of marked edges, poor stability of corner point extraction, unreasonable bit-depth conversion, insufficient real-time positioning, and poor robustness in complex scenarios.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] According to a first aspect of the present invention, the present invention provides a method for three-dimensional positioning of array elements, the method comprising: Acquire an image frame sequence for the array element to be tested; the image frame sequence includes high-bit-depth grayscale image frames from multiple perspectives. An adaptive bit-depth mapping enhancement technique is used to dynamically adjust the range of each high bit-depth grayscale image frame to generate a low bit-depth image. For each of the low-bit depth images, an initial candidate label detection is performed to obtain candidate encoded visual labels, and a corresponding ROI polygon region is generated for each of the candidate encoded visual labels. Local enhancement techniques based on candidate boundaries are used to perform local mapping enhancement processing on each of the ROI polygon regions to obtain enhanced ROI images; The enhanced ROI images are subjected to secondary refinement detection, and the target encoded visual label containing the target corner points and the target single-frame pose is output. The dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region is used as the observation quality index to determine the corner observation covariance of the target corner point. An incremental factor map is constructed with the target single-frame pose as the state variable and the reprojection error of the target corner point as the observation constraint, and an adaptive noise-weighted multi-frame joint optimization objective based on the corner point observation covariance is defined. Based on the multi-frame joint optimization objective, the incremental factor map is incrementally updated and solved to obtain the center pose of the array element and its three-dimensional coordinates in the world coordinate system.

[0008] Further, acquiring the image frame sequence for the element to be tested includes: The image acquisition device performs a preset trajectory movement relative to the array element under test to acquire multiple high-bit-depth grayscale image frames from multiple perspectives and in succession; the bit depth of the high-bit-depth grayscale image frames is 10 bits to 16 bits. Each of the high-bit deep grayscale image frames and its corresponding metadata is stored to generate an image frame sequence; The metadata includes at least one of frame number, timestamp, and image size.

[0009] Furthermore, the step of using adaptive bit-depth mapping enhancement technology to dynamically adjust the range of each of the high bit-depth grayscale image frames to generate a low bit-depth image includes: Gaussian denoising is performed on each high bit depth grayscale image frame in the image frame sequence to obtain a denoised image. The denoised image is then cropped to the effective bit depth range to obtain an effective bit depth image. The pixel intensity histograms of each effective bit depth image are statistically analyzed on the downsampling grid, and the intensity clipping interval is determined based on the pixel intensity distribution. The mathematical expression is as follows:

[0010]

[0011] in, Indicates the lower limit of strength cutting; Indicates the upper limit of strength reduction; Represents the quantile operator; Indicates the lower quantile parameter, For higher quantile parameters; The image gradient magnitude is calculated based on the intensity cropping interval, and the edge threshold is determined according to the gradient magnitude distribution. The mathematical expression is as follows:

[0012] in, Represents the set of image gradient magnitudes; This represents the gradient quantile parameter; Represents the gradient quantile operator; Indicates the edge threshold; The edge pixels whose pixel intensity is within the intensity clipping interval and whose gradient magnitude reaches the edge threshold are aggregated to generate an edge pixel set. The mapping intensity window is determined based on the pixel intensity distribution of the edge pixel set, and the mathematical expression is as follows:

[0013]

[0014] in, Indicates the lower bound of the mapping strength; Indicates the upper limit of the mapping strength; Represents the set of edge pixels; Represents the pixel intensity values ​​of the input valid bit-depth image; This represents the low-quantile parameter of the edge pixels; This represents the high-resolution parameter of the edge pixels; , Represents the edge pixel quantile operator; A linear mapping based on a mapping intensity window combined with saturation cropping is used to convert the effective bit-depth image into a low bit-depth image with a bit depth of 8 bits. The mathematical expression is as follows:

[0015] in, This represents the pixel intensity value of the output low-bit-depth image; Represents the pixel intensity values ​​of the input valid bit-depth image; Indicates the lower bound of the mapping strength; This indicates the upper limit of the mapping strength.

[0016] Further, the step of performing initial candidate label detection on each of the low-bit-depth images to obtain candidate encoded visual labels, and generating corresponding ROI polygon regions for each of the candidate encoded visual labels, includes: A square-coded visual marker detector is used to detect candidate markers in each of the low-bit-depth images and outputs the candidate coded visual markers; the candidate coded visual markers include at least a marker number, candidate corner points, candidate boundaries, and the geometric center of the candidate boundaries; Using the geometric center of the candidate boundary as a reference, the size of the candidate boundary of the low-bit depth image is expanded to obtain the corresponding ROI polygon region:

[0017] in, The corner points of the expanded ROI polygon region; The candidate corner points that constitute the candidate boundary; Indicates the expansion factor; Indicates the geometric center location of the candidate boundary.

[0018] Furthermore, the step of employing a local enhancement technique based on candidate boundaries to perform local mapping enhancement processing on each of the ROI polygon regions to obtain an enhanced ROI image includes: Within the ROI polygonal region, the pixel intensity distribution of the effective bit depth image is statistically analyzed to obtain local mapping parameters. The dynamic range of the ROI is then calculated based on these local mapping parameters, as expressed mathematically below:

[0019] in, Indicates the dynamic range of ROI; Indicates the upper limit of local mapping intensity; Indicates the lower bound of local mapping intensity; Represents the set of pixels within the polygonal region of the ROI; Represents the pixel intensity value within the polygonal region of the ROI; High-quantile parameters representing pixel intensity within the polygonal region of the ROI; The low quantile parameter represents the pixel intensity within the polygonal region of the ROI; , Represents the local pixel quantile operator; Based on the dynamic range of the ROI, local mapping enhancement is performed on the polygonal region of the ROI to obtain an enhanced ROI image, as expressed mathematically below:

[0020] in, This represents the pixel intensity value of the output enhanced ROI image; Represents the clipping function; This represents the pixel intensity value within the input ROI polygonal region; Indicates the lower bound of local mapping intensity; Indicates the dynamic range of ROI.

[0021] Further, the secondary refinement detection of each of the enhanced ROI images, outputting a target-encoded visual label containing target corner points and target single-frame pose, includes: In edge refinement mode, the enhanced ROI image is processed with a confirmation output for the candidate encoded visual markers, which is then used as the target encoded visual markers; or, In the edge refinement mode, for the candidate marker detection of the enhanced ROI image, the encoded visual markers that are detected by both the initial candidate marker detection and the secondary refinement detection are output as the target encoded visual markers. The target coded visual marker includes at least the target marker number, target corner point, target geometric center, and target single-frame pose; the target single-frame pose includes the coded visual marker pose and the image acquisition device pose.

[0022] Further, the step of using the dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region as an observation quality index to determine the corner observation covariance of the target corner point includes: The measurement noise parameters of the target corner point are determined using the following piecewise function:

[0023]

[0024] in, , They represent , Standard deviation of corner pixel noise in the direction; Represents the clipping function; , They represent , The linear intercept parameter of the direction; , They represent , The linear slope parameter of the direction; Indicates the dynamic range of ROI; This represents the minimum threshold for the noise standard deviation; This represents the maximum threshold value for the noise standard deviation; Based on the measured noise parameters, the corner observation covariance corresponding to each target coded visual marker is determined, and the mathematical expression is as follows:

[0025] in, Indicates the first The corner observation covariance corresponding to each target encoded visual marker; , They represent the first A target-coded visual marker in , Standard deviation of corner pixel noise in the direction.

[0026] Furthermore, the construction of an incremental factor map with the target single-frame pose as the state variable and the reprojection error of the target corner points as the observation constraint, and the definition of an adaptive noise-weighted multi-frame joint optimization objective based on the corner point observation covariance, includes: The reprojection error is defined mathematically as follows:

[0027] in, Indicates reprojection error; , representing the observed pixel coordinates of the target corner point; Indicates the intrinsic parameters of the image acquisition device; Indicates the pose of the image acquisition device; Represents the encoded visual marker pose; This indicates the three-dimensional coordinates of a corner point in the marked coordinate system; Construct an incremental factor map with the pose of the image acquisition device and the pose of the coded visual marker as state variables and the reprojection error as observation constraint; as well as, A multi-frame joint optimization objective is constructed by weighting the reprojection error using the corner observation covariance and introducing a robust kernel function. The mathematical expression is as follows:

[0028] in, Represents a robust kernel function; Indicates the first The pose of the image acquisition device corresponding to the frame image; Indicates the first The coded visual marker pose corresponding to each target coded visual marker; Indicates the first The first frame image corresponding to The first target encoding visual mark Reprojection error at each corner point; Indicates the first The corner observation covariance corresponding to each target encoded visual marker; This represents the constraint residual between poses of adjacent frames; For the corresponding covariance; , They represent the first The prior residuals and corresponding covariances of each target encoding visual tag during initialization.

[0029] Further, the incremental update solution of the incremental factor map based on the multi-frame joint optimization objective to obtain the element center pose and its three-dimensional coordinates in the world coordinate system includes: An incremental factor graph solver based on incremental smoothing and mapping algorithms is used. When the observation quality index of each acquired image frame meets the preset quality conditions, the corresponding target coded visual marker is added to the incremental factor graph, and an incremental update solution is performed on the incremental factor graph to obtain the optimized pose of each target coded visual marker in the world coordinate system. If the reference origin of the target encoded visual marker coincides with the center of the element to be tested, then the three-dimensional coordinates of the element to be tested in the world coordinate system are taken from the translation vector of the optimized pose. If there is a fixed rigid offset between the reference origin of the target encoded visual marker and the center of the array element to be tested, the array element center pose of the array element to be tested is calculated according to the pre-calibrated fixed rigid transformation, and the three-dimensional coordinates of the array element to be tested in the world coordinate system are calculated according to the rotation matrix and translation vector of the optimized pose.

[0030] According to a second aspect of the present invention, the present invention provides a three-dimensional positioning system for array elements, the system comprising: An image acquisition module is used to acquire an image frame sequence for the array element to be tested; the image frame sequence includes high-bit-depth grayscale image frames from multiple perspectives. The bit depth processing module is used to dynamically adjust the range of each of the high bit depth grayscale image frames using adaptive bit depth mapping enhancement technology to generate a low bit depth image. The initial detection module is used to perform initial candidate label detection on each of the low bit depth images to obtain candidate coded visual labels, and generate corresponding ROI polygon regions for each candidate coded visual label; The ROI enhancement module is used to perform local mapping enhancement processing on each of the ROI polygon regions using a candidate boundary-based local enhancement technique to obtain an enhanced ROI image. The secondary detection module is used to perform secondary refinement detection on each of the enhanced ROI images and output a target-coded visual label containing the target corner point and the target single-frame pose. The quality observation module is used to determine the corner observation covariance of the target corner point by using the dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region as the observation quality index. The fusion optimization module is used to construct an incremental factor map with the target single-frame pose as the state variable and the reprojection error of the target corner point as the observation constraint, and to define a multi-frame joint optimization objective based on the corner point observation covariance with adaptive noise weighting. The solution output module is used to incrementally update the incremental factor map based on the multi-frame joint optimization objective to obtain the center pose of the array element and its three-dimensional coordinates in the world coordinate system.

[0031] The present invention, by adopting the above technical solution, has at least the following beneficial effects: This invention provides a method and system for three-dimensional localization of array elements. The method involves acquiring an image frame sequence for the array element to be tested; the image frame sequence includes high-bit-depth grayscale image frames from multiple perspectives; dynamically adjusting the range of each high-bit-depth grayscale image frame using adaptive bit-depth mapping enhancement technology to generate a low-bit-depth image; performing initial candidate marker detection on each low-bit-depth image to obtain candidate coded visual markers, and generating corresponding ROI polygonal regions for each candidate coded visual marker; using a local enhancement technology based on candidate boundaries to perform local mapping enhancement processing on each ROI polygonal region to obtain an enhanced ROI image; and then performing local mapping enhancement processing on each enhanced ROI polygonal region. A secondary refinement detection is performed on a strong ROI image to output a target-encoded visual label containing the target corner points and the target single-frame pose. The dynamic range of the ROI, obtained by statistically analyzing the pixel intensity of the polygonal region of the ROI, is used as an observation quality index to determine the corner observation covariance of the target corner points. An incremental factor map is constructed with the target single-frame pose as the state variable and the reprojection error of the target corner points as the observation constraint. An adaptive noise-weighted multi-frame joint optimization objective based on the corner observation covariance is defined. The incremental factor map is incrementally updated based on the multi-frame joint optimization objective to obtain the element center pose and its 3D coordinates in the world coordinate system. This improves the separability of the label edges and the stability of the corner points in weak contrast and noise-perturbed scenarios, and more reasonably handles differences in observation quality during the multi-frame fusion optimization stage, thereby improving the accuracy, consistency, and real-time performance of the element 3D localization.

[0032] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0034] Figure 1 A flowchart illustrating a three-dimensional positioning method for array elements provided in an embodiment of the present invention is shown. Figure 2 A flowchart illustrating an embodiment of the adaptive bit-depth mapping enhancement method provided by the present invention is shown. Figure 3 A schematic diagram illustrating the principle of ROI polygon region expansion provided by an embodiment of the present invention is shown; Figure 4 A schematic diagram illustrating the principle of an incremental factor graph fusion optimization method provided in an embodiment of the present invention is shown. Figure 5 A schematic diagram of the structure of a three-dimensional positioning system for array elements provided in an embodiment of the present invention is shown. Detailed Implementation

[0035] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0036] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0037] This invention provides a method for three-dimensional positioning of array elements, applicable to flexible phased arrays, array measurement boards, or other scenarios requiring the acquisition of array element spatial coordinates. In general operation, coded visual markers can be set on the array elements to be measured. These coded visual markers can be set in a one-to-one correspondence with the array elements under test, or a fixed spatial transformation relationship can be established with the array elements under test through a rigid connection. Preferably, the coded visual markers are square coded visual markers; for example, AprilTag type markers are used as a specific example, but this invention does not limit this. Figure 1 As shown, the three-dimensional positioning method for array elements provided in this embodiment of the invention may include at least the following steps S101~S108: Step S101: Obtain the image frame sequence for the array element to be tested.

[0038] In this embodiment of the invention, for acquiring image frame sequences, an image acquisition device (such as a calibrated industrial camera) can output a high bit-depth grayscale image with a bit depth of 10 to 16 bits. For example, a grayscale image with a resolution of 5120×5120 pixels and a Mono12 pixel format can be used as the subsequent input image. The exposure time can be further set to 3000 microseconds, the gain to 16dB, and the frame rate to 10 to 20 frames per second to balance imaging brightness and acquisition stability.

[0039] It should be noted that the camera parameters described above are merely examples and do not constitute a limitation of the present invention. Before image acquisition, the camera's intrinsic parameters need to be calibrated to obtain the camera intrinsic parameter matrix used for subsequent single-frame pose estimation and reprojection error calculation, as shown in the following example:

[0040] in, and The camera in the image direction and Focal length in direction, and The coordinates of the main point.

[0041] To acquire multiple consecutive high-bit-depth grayscale image frames from multiple perspectives, in this embodiment of the invention, the image acquisition device can perform a preset trajectory movement relative to the array element under test. In practical applications, a robotic arm can drive the camera to move relative to the array element under test along a predetermined trajectory, and continuously acquire image frames during the movement. Preferably, the predetermined trajectory can be a spherical circular trajectory, a random trajectory, or other sampling trajectories covering different viewing angles.

[0042] Furthermore, each high-bit-depth grayscale image frame and its corresponding metadata are stored to generate an image frame sequence. In practical applications, each acquired original image frame is preferably saved as an original high-bit-depth array file in a lossless manner, while recording and storing corresponding metadata such as frame number, timestamp, and image size for subsequent processing and result traceability.

[0043] Step S102: Adaptive bit-depth mapping enhancement technology is used to dynamically adjust the range of each high bit-depth grayscale image frame to generate a low bit-depth image.

[0044] Since the original input image is a high-bit-depth grayscale image, directly converting it to an 8-bit grayscale image using a uniform linear compression method can easily lead to insufficient utilization of the effective dynamic range near the edges of the encoded visual markers, thereby reducing the recall rate and corner stability of candidate marker detection. Therefore, this embodiment of the invention employs an edge-driven adaptive bit-depth mapping enhancement method to map the high-bit-depth grayscale image into an 8-bit low-bit-depth image that is more suitable for detection. Figure 2 The diagram shown is a flowchart of the adaptive bit-depth mapping enhancement method, which includes at least steps S102-1 to S102-6: Step S102-1: Perform Gaussian denoising on each high bit depth grayscale image frame in the image frame sequence to obtain a denoised image. Crop the denoised image to the effective bit depth range to obtain an effective bit depth image.

[0045] Let the input high-bit-depth grayscale image be First, Gaussian denoising can be performed on the high-bit-depth grayscale image to obtain the denoised image. Preferably, the standard deviation of Gaussian denoising can be taken as... Furthermore, the denoising result can be cropped to the effective bit depth range of the high bit depth image to obtain the denoised effective bit depth image. This is to avoid outliers affecting subsequent quantile statistics.

[0046] Step S102-2: Calculate the pixel intensity histogram of each effective bit depth image on the downsampling grid, and determine the intensity clipping interval based on the pixel intensity distribution.

[0047] This step statistically analyzes the pixel intensity distribution of the effective bit-depth image on the downsampling grid. Let the downsampling factor be... Preferably, Furthermore, the intensity clipping interval is determined using the quantile operator. The mathematical expression is as follows:

[0048]

[0049] in, Indicates the lower limit of strength cutting; Indicates the upper limit of strength reduction; Represents the quantile operator; Indicates the lower quantile parameter, This is a high quantile parameter. Preferably, , .

[0050] Step S102-3: Calculate the image gradient magnitude based on the intensity cropping interval, and determine the edge threshold according to the gradient magnitude distribution.

[0051] After determining the intensity clipping interval, calculate the image gradient magnitude. The edge threshold is determined by the gradient magnitude distribution. Specifically, the gradient magnitude can be approximated by the sum of the absolute values ​​of the differences between adjacent pixels, and the edge threshold can be determined by the higher quantile of the gradient magnitude.

[0052] Among them, among them, Represents the set of image gradient magnitudes; This represents the gradient quantile parameter, with a preferred value of 95%. Represents the gradient quantile operator; This represents the edge threshold.

[0053] Step S102-4: Summarize edge pixels whose pixel intensity is within the intensity clipping interval and whose gradient magnitude reaches the edge threshold to generate an edge pixel set.

[0054] In other words, when the pixel intensity is located at [ The interval and the gradient magnitude is greater than or equal to Construct an edge pixel set from the pixels. :

[0055] in, Represents the coordinates in the effective depth image The corresponding pixel intensity value; Representing coordinates The corresponding image gradient magnitude.

[0056] Step S102-5: Determine the mapping intensity window based on the pixel intensity distribution of the edge pixel set.

[0057] The mathematical expression for this step is as follows:

[0058]

[0059] in, Indicates the lower bound of the mapping strength; Indicates the upper limit of the mapping strength; Represents the set of edge pixels; Represents the pixel intensity values ​​of the input valid bit-depth image; This represents the low-quantum parameter of edge pixels, with a preferred value of 5%. This represents the high quantile parameter of edge pixels, with a preferred value of 95%. , This represents the edge pixel quantile operator.

[0060] Step S102-6: Using linear mapping based on the mapping intensity window combined with saturation cropping technology, the effective bit depth image is converted into a low bit depth image with a bit depth of 8 bits.

[0061] The mathematical expression for this step is as follows:

[0062] in, This represents the pixel intensity value of the output low-bit-depth image; Represents the pixel intensity values ​​of the input valid bit-depth image; Indicates the lower bound of the mapping strength; This indicates the upper limit of the mapping strength.

[0063] Through such Figure 2 The adaptive bit-depth mapping enhancement method shown enables pixels associated with the edges of encoded visual markers to obtain a grayscale distribution more suitable for detection, thereby improving the stability of subsequent candidate marker detection.

[0064] Step S103: Perform initial candidate label detection for each low-bit depth image to obtain candidate coded visual labels, and generate corresponding ROI polygon regions for each candidate coded visual label.

[0065] This step can employ a square-coded visual marker detector to detect candidate markers in each low-bit-depth image, outputting candidate coded visual markers. These candidate coded visual markers can include marker numbers, candidate corner points, candidate boundaries, the geometric center of the candidate boundaries, and single-frame pose estimation results calculated based on camera intrinsics and marker side lengths. In practical applications, the initial candidate marker detection can be performed with edge refinement mode disabled to improve candidate marker recall.

[0066] Furthermore, such as Figure 3 The diagram illustrates the principle of generating corresponding ROI polygonal regions for candidate encoded visual markers. As can be seen, the specific implementation involves expanding the size of the candidate boundaries in the low-bit-depth image based on the geometric center of the candidate boundaries to obtain the corresponding ROI polygonal regions.

[0067] in, The corner points of the ROI after size expansion; The candidate corner points that constitute the candidate boundary; This represents the expansion factor, with a preferred value of 1.25. Indicates the geometric center location of the candidate boundary.

[0068] For the expanded ROI corner points, they can be further cropped to the effective range of the image to obtain the final ROI polygonal region. This step, by scaling the candidate boundaries, can preserve the grayscale information of the encoded visual marker edges and their neighborhoods during subsequent local enhancement, thereby improving the robustness of the refined detection.

[0069] Step S104: Local mapping enhancement processing is performed on each ROI polygon region using candidate boundary-based local enhancement technology to obtain an enhanced ROI image.

[0070] Within the ROI polygonal region, the effective bit depth image is statistically analyzed. The pixel intensity distribution is obtained to acquire local mapping parameters. The dynamic range of the Region of Interest (ROI) is calculated based on these parameters. Let the upper and lower limits of the local mapping for pixel intensity within the polygonal region of the ROI be respectively... and The mathematical expression for calculating the dynamic range of ROI is as follows:

[0071] in, Indicates the dynamic range of ROI; Indicates the upper limit of local mapping intensity; Indicates the lower bound of local mapping intensity; Represents the set of pixels within the polygonal region of the ROI; Represents the pixel intensity value within the polygonal region of the ROI; The high quantile parameter representing the pixel intensity within the polygonal region of the ROI, preferably 90%; The lower quantile parameter represents the pixel intensity within the polygonal region of the ROI, with a preferred value of 10%. , This represents the local pixel quantile operator.

[0072] It is understandable that the ROI dynamic range, as an indicator of observation quality, indicates that the larger the value, the more sufficient the information of the edges and corners around the candidate coded visual marker, and the higher the observation quality.

[0073] Furthermore, after obtaining the local mapping parameters and the dynamic range of the ROI, local mapping enhancement can be performed on the polygonal region of the ROI to obtain an enhanced ROI image. The mathematical expression is as follows:

[0074] in, This represents the pixel intensity value of the output enhanced ROI image; Represents the clipping function; This represents the pixel intensity value within the input ROI polygonal region; Indicates the lower bound of local mapping intensity; Indicates the dynamic range of ROI.

[0075] The local mapping method given in this step redistributes the grayscale dynamic range near each candidate coded visual marker, which enhances the edges and corners in the local weak contrast region, thereby improving the accuracy of corner extraction during secondary detection.

[0076] Step S105: Perform secondary refinement detection on each enhanced ROI image and output a target-coded visual label containing the target corner point and the target single-frame pose.

[0077] In practical applications, secondary refinement detection is performed in edge refinement mode to improve corner point localization accuracy. Specifically, in edge refinement mode, confirmation outputs for candidate coded visual markers can be performed on the enhanced ROI image as target coded visual markers; or, in edge refinement mode, candidate marker detection on the enhanced ROI image can output coded visual markers that were detected in both the initial candidate marker detection and secondary refinement detection as target coded visual markers to suppress false detections.

[0078] Thus, the refined coded visual marker detection results can be obtained, namely the target coded visual marker, including the target marker number, target corner point, target geometric center, and target single-frame pose (coded visual marker pose and camera pose) calculated based on camera intrinsic parameters and marker size.

[0079] Meanwhile, embodiments of the present invention can dynamically adjust the ROI dynamic range corresponding to the secondary refined detection results for each target encoded visual marker. (indicating the first) Each target-coded visual marker is included in the detection result and used as a quality basis for subsequent noise modeling and factor map weighting. Preferably, the side length of the target-coded visual marker is 8.5 mm, but it can also be adjusted according to the actual array element size and imaging field of view.

[0080] Step S106: The dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region is used as the observation quality index to determine the corner observation covariance of the target corner point.

[0081] Because contrast and noise levels vary between different image frames, different candidate coded visual markers, and different local regions, assigning the same weight to all corner observations in subsequent multi-frame fusion optimization can easily lead to low-quality observations negatively impacting the overall localization results. Therefore, this embodiment of the invention obtains the observation quality index, i.e., the ROI dynamic range, based on pixel intensity statistics within the ROI polygonal region, and adaptively determines the noise parameters of corner pixel noise standard deviation, corner observation covariance, and pose correlation factors based on the ROI dynamic range.

[0082] First, the measurement noise parameters of the target corner point are determined using the following piecewise function:

[0083]

[0084] in, , They represent , Standard deviation of corner pixel noise in the direction; Represents the clipping function; , They represent , The linear intercept parameter of the direction; , They represent , The linear slope parameter of the direction; Indicates the dynamic range of ROI; This represents the minimum threshold for the noise standard deviation; This represents the maximum threshold of the noise standard deviation.

[0085] In other words, the embodiments of the present invention are based on the dynamic range of ROI. Adaptively determine the standard deviation of corner pixel noise. Let the first pixel be... The standard deviations of the corner pixel noise of the candidate coded visual markers are respectively and Therefore, the following piecewise function can be used: when hour,

[0086] when hour,

[0087] in, Indicates the variable Crop to Within the interval. As can be seen from the above definition, when the dynamic range of the ROI is large, the standard deviation of corner pixels decreases accordingly; when the dynamic range of the ROI is small, the standard deviation of corner pixels increases accordingly, thus giving higher quality observations a greater weight in subsequent optimization.

[0088] Therefore, based on the measured noise parameters, the corner observation covariance corresponding to each target's coded visual marker is determined, and the mathematical expression is as follows:

[0089] in, Indicates the first The corner observation covariance corresponding to each target encoded visual marker; , They represent the first A target-coded visual marker in , Standard deviation of corner pixel noise in the direction.

[0090] Furthermore, embodiments of the present invention can also determine the noise standard deviation vector of pose-related factors based on observation quality indicators. This noise standard deviation vector serves as the basis for determining the pose-related noise standard deviation vector. The brightness index can be further defined, and its mathematical expression is as follows:

[0091] in, To avoid the smallest positive number with a denominator of zero; then the pose noise standard deviation vector It can then be defined as:

[0092]

[0093]

[0094] It should be noted that the pose noise standard deviation vectors mentioned above correspond to the noise parameters in the rotation and translation directions, and are used for the subsequent weighting of pose-related factors. This quality-driven noise modeling approach allows high-quality observations with a large ROI dynamic range to receive lower noise and higher weights, while low-quality observations receive higher noise and lower weights, thereby improving the stability of the overall optimization results.

[0095] Step S107: Construct an incremental factor map with the target single-frame pose as the state variable and the reprojection error of the target corner point as the observation constraint, and define a multi-frame joint optimization objective based on the corner point observation covariance with adaptive noise weighting.

[0096] As an optional embodiment of the present invention, for the observation results of the dynamic range of the ROI polygon region, before entering the factor graph fusion optimization, the observation results with the decoding Hamming distance exceeding the preset distance threshold can be removed to avoid erroneous observations from polluting the subsequent factor graph optimization.

[0097] Furthermore, for the target encoded visual tags after the removal process, a reprojection error is defined. Specifically, let the first... The camera pose corresponding to the frame image is , No. The pose of the encoded visual markers in the world coordinate system is: Mark the first coordinate system The three-dimensional coordinates of the corner points are Let the first The first frame The first target encoding visual mark The refined observation pixel coordinates of each corner point are The corresponding reprojection error can then be defined as:

[0098] in, Represents the pinhole projection function; Indicates the first The first frame image corresponding to The first target encoding visual mark Reprojection error at each corner point.

[0099] Therefore, an incremental factor graph is constructed, using camera pose and coded visual label pose as state variables and reprojection error as observation constraint. The embodiments of this invention employ the following... Figure 4 The incremental factor map fusion optimization method shown solves the incremental factor map. In the solution process, this embodiment of the invention constructs a multi-frame joint optimization objective that uses the corner observation covariance to weight the reprojection error and introduces a robust kernel function. The mathematical expression is as follows:

[0100] in, Represents a robust kernel function; Indicates the first The pose of the image acquisition device corresponding to the frame image; Indicates the first The coded visual marker pose corresponding to each target coded visual marker; Indicates the first The first frame image corresponding to The first target encoding visual mark Reprojection error at each corner point; Indicates the first The corner observation covariance corresponding to each target encoded visual marker; This represents the constraint residual between poses of adjacent frames; For the corresponding covariance; , They represent the first The prior residuals and corresponding covariances of each target encoding visual tag during initialization.

[0101] It should be noted that, for constructing the joint optimization objective across multiple frames, this embodiment of the invention can construct pose constraint factors based on the single-frame pose estimation results in the detection output. Specifically, the single-frame pose estimation results given by the secondary refined detection or the initial candidate label detection are used as the initial relative pose values ​​between adjacent frames, and a Between factor between the camera pose and the coded visual label pose is introduced into the incremental factor map. When there is a lack of shared coded visual labels between adjacent image frames and the camera pose of the previous frame is already available, a relaxed zero-pose Between factor can also be introduced between adjacent camera poses to improve the convergence and stability of the incremental solution process.

[0102] It should be noted that, to improve the real-time performance of high-resolution image processing, this embodiment of the invention, in the process of image frame acquisition and thread processing, preferably pre-allocates a fixed-size GPU memory slot pool based on the current program implementation, and transfers image data between stages such as uploading, Gaussian processing, bit-depth mapping enhancement, and ROI local enhancement through slot identifiers. In an optional embodiment of the invention, a combination of Compute Unified Device Architecture (CUDA) streams and queue-based inter-stage scheduling can be used to enable parallel overlapping execution of GPU processing, processor-side detection, and incremental optimization. That is, while the GPU is processing the nth frame of the image, the processor can execute the nth frame or the nth frame in parallel. The factor graph corresponding to each frame is updated incrementally, thus forming a pipelined execution mode in which acquisition, transmission, detection and optimization overlap.

[0103] Step S108: Based on the multi-frame joint optimization objective, the incremental factor map is incrementally updated and solved to obtain the center pose of the array element and its three-dimensional coordinates in the world coordinate system.

[0104] Specifically, embodiments of the present invention may employ an incremental factor graph solver based on the Incremental Smoothing and Mapping 2 (iSAM2) algorithm. When the observation quality index of each acquired image frame meets the preset quality conditions, the corresponding target coded visual marker is added to the incremental factor graph, and an incremental update solution is performed on the incremental factor graph to obtain the optimized pose of each target coded visual marker in the world coordinate system.

[0105] That is, for newly emerging target-coded visual markers, a quality control strategy can be executed before adding them to the factor map. Specifically, when the ROI dynamic range corresponding to a newly emerging target-coded visual marker is lower than a preset threshold and the cumulative number of observations has not reached a preset observation threshold, the marker is not immediately added to the incremental factor map. Instead, its observations are temporarily stored, and the marker is initialized and added to the incremental factor map only when its subsequent observations meet the preset quality conditions. Preferably, the ROI dynamic range threshold can be set to 200. When the same marker appears in a higher-quality observation in a subsequent frame, initialization is then performed, thereby avoiding the pollution of the overall graph optimization result by the low-quality first observation. During the solution process, the embodiments of the present invention do not need to completely resolve all state variables after each new observation is added, but only perform incremental optimization on the affected local variables. Therefore, it is particularly suitable for high-resolution image continuous processing scenarios.

[0106] After completing the incremental update solution, the optimized pose of each target encoded visual tag in the world coordinate system can be obtained. If the reference origin of the target encoded visual marker coincides with the center of the array element, then the first... The three-dimensional coordinates of each element to be tested can be directly taken from the translation vector of the encoded visual marker pose. If there is a fixed rigid offset between the target encoded visual marker and the center of the element to be tested, then a pre-calibrated fixed rigid transformation can be performed. The center pose of the array element to be tested is obtained by the following formula:

[0107] in, This indicates the pose of the center of the array element to be tested; This represents the center of the original array element to be tested.

[0108] Furthermore, the three-dimensional coordinates of the element under test in the world coordinate system are calculated based on the rotation matrix and translation vector of the optimized pose. The mathematical expression is as follows:

[0109] in, and These represent the rotation matrix and translation vector of the optimized pose of the target encoded visual marker, respectively. This indicates the fixed coordinates of the array element center in the marked coordinate system.

[0110] Finally, the three-dimensional coordinates of all array elements in a unified world coordinate system can be summarized and output.

[0111] As an optional embodiment of the present invention, error evaluation can also be used to verify the final output three-dimensional coordinate results.

[0112] Specifically, the optimized translation coordinates of each target-encoded visual marker in the last record can be extracted from the JSON file output by the factor graph optimization toolkit (Georgia Tech Smoothing And Mapping, GTSAM) as estimated coordinates; the corresponding reference coordinates can be high-precision calibration results, coordinates exported from CAD / STEP models, or other known standard coordinates.

[0113] Since there is usually a global rigid body deviation between the estimated coordinate system and the reference coordinate system, it is preferable to first use a rigid body registration method to align the two sets of coordinates. The process is as follows: Using a multi-starting-point rigid body registration method, iterative alignment is performed on multiple initial rotation candidates around the reference axis. In each iteration, firstly, the cost matrix between the estimated point set and the reference point set is calculated based on the current transformation; then, the Hungarian algorithm is used to perform one-to-one matching between the two sets of points; next, the matching results are clipped based on the matching distance; finally, based on the clipped matching point pairs, the Kabsch algorithm is used to estimate the new rigid body transformation, thus completing the iterative nearest-point alignment process with clipping.

[0114] In practical applications, let the first... The coordinates of the estimated points are: The corresponding reference coordinates are Let the aligned rotation matrix and translation vector be respectively and The rigid body registration process can then be expressed as the following optimization problem:

[0115] in, This represents the rotation matrix to be optimized; This represents the translation vector to be optimized. This represents the total number of points participating in the matching.

[0116] After alignment is completed, the first The residual vector of each matching point It can be represented as:

[0117] Furthermore, the corresponding Euclidean error It can be represented as:

[0118] Therefore, the average error of all array elements can be calculated separately. and root mean square error The mathematical expression is as follows:

[0119]

[0120] In addition, the median error and maximum error can be further statistically analyzed to comprehensively characterize the overall accuracy level of the positioning results.

[0121] This invention provides a method for three-dimensional localization of array elements. The method involves acquiring an image frame sequence for the array element to be tested; using adaptive bit-depth mapping enhancement technology to dynamically adjust the range of each high-bit-depth grayscale image frame to generate a low-bit-depth image; performing initial candidate marker detection on each low-bit-depth image to obtain candidate coded visual markers, and generating corresponding ROI polygonal regions for each candidate coded visual marker; using local enhancement technology based on candidate boundaries to perform local mapping enhancement processing on each ROI polygonal region to obtain an enhanced ROI image; performing secondary refinement detection on each enhanced ROI image to output target coded visual markers containing target corner points and target single-frame pose; using the ROI dynamic range obtained by pixel intensity statistics for the ROI polygonal region as an observation quality index to determine the corner observation covariance of the target corner points; constructing an incremental factor map with the target single-frame pose as the state variable and the reprojection error of the target corner points as the observation constraint, and defining an adaptive noise-weighted multi-frame joint optimization objective based on the corner observation covariance; and incrementally updating the incremental factor map based on the multi-frame joint optimization objective to obtain the array element center pose and its three-dimensional coordinates in the world coordinate system. This invention achieves stable and high-precision solving of the three-dimensional coordinates of array elements. Especially under conditions of high-resolution, low-contrast image acquisition with significant local brightness differences, this embodiment can balance detection reliability, optimized stability, and overall real-time performance. Compared with existing technologies, this invention has at least the following beneficial effects: 1) An edge-driven adaptive bit-depth mapping enhancement method is adopted to dynamically adjust the range of high bit-depth grayscale image frames, so that the pixels related to the edges of the encoded visual markers obtain a grayscale distribution that is more suitable for detection. This effectively improves the edge separability and corner extraction stability of the encoded visual markers, and improves the problem of insufficient dynamic range utilization when high bit-depth images are directly compressed into low bit-depth images.

[0122] 2) Based on candidate boundary ROI local enhancement and two-stage refined detection, the candidate region is locally mapped and enhanced. Combined with the confirmation mechanism of the first and second detection, the detection reliability and corner positioning accuracy of candidate encoded visual markers can be effectively improved, false detections can be suppressed, and the stability of detection results can be enhanced.

[0123] 3) By introducing an observation quality adaptive noise modeling method based on the dynamic range of ROI, the quality differences between different observations can be reasonably characterized, so that high-quality observations contribute more to the optimization results and low-quality observations cause less interference, thereby improving the consistency and stability of the coded visual marker pose estimation and the three-dimensional localization results of the array elements.

[0124] 4) Construct an incremental factor map with camera pose and coded visual marker pose as state variables and refined corner reprojection error as observation constraint. Combined with pose constraint factors, robust kernel function and quality control strategy for newly emerging coded visual markers, it can effectively suppress the adverse effects of abnormal and low-quality observations on optimization results and improve the convergence stability and overall positioning accuracy of the multi-frame fusion process.

[0125] 5) The bit depth mapping enhancement and ROI local enhancement are deployed on the GPU side and a DMA-based data transfer acceleration strategy is adopted to achieve the overlap of data transfer and GPU computing in time, which can effectively reduce the overhead of repeated copying and improve the real-time performance and overall processing efficiency of the array element three-dimensional positioning process.

[0126] Furthermore, as Figure 1 In specific implementation, embodiments of the present invention provide a three-dimensional positioning system for array elements, such as... Figure 5 As shown, the system may include: an image acquisition module 510, a bit depth processing module 520, an initial detection module 530, an ROI enhancement module 540, a secondary detection module 550, a quality observation module 560, a fusion optimization module 570, and a solution output module 580.

[0127] The image acquisition module 510 can be used to acquire an image frame sequence for the array element under test; the image frame sequence includes high bit-depth grayscale image frames from multiple perspectives. The bit depth processing module 520 can be used to dynamically adjust the range of each high bit depth grayscale image frame using adaptive bit depth mapping enhancement technology to generate a low bit depth image. The initial detection module 530 can be used to perform initial candidate label detection for each low bit depth image, obtain candidate coded visual labels, and generate corresponding ROI polygon regions for each candidate coded visual label. The ROI enhancement module 540 can be used to perform local mapping enhancement processing on each ROI polygon region using candidate boundary-based local enhancement technology to obtain an enhanced ROI image. The secondary detection module 550 can be used to perform secondary refinement detection on various enhanced ROI images and output target encoded visual labels containing target corner points and target single-frame pose. The quality observation module 560 can be used to determine the corner observation covariance of the target corner point by using the dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region as the observation quality index. The fusion optimization module 570 can be used to construct an incremental factor map with the target single-frame pose as the state variable and the reprojection error of the target corner point as the observation constraint, and define an adaptive noise-weighted multi-frame joint optimization objective based on the corner point observation covariance. The solution output module 580 can be used to incrementally update the incremental factor map based on the multi-frame joint optimization objective, and obtain the center pose of the array element and its three-dimensional coordinates in the world coordinate system.

[0128] It should be noted that other corresponding descriptions of the functional modules involved in the array element three-dimensional positioning system provided in this embodiment of the invention can be found in the following references. Figure 1 The corresponding description of the method shown will not be repeated here.

[0129] Those skilled in the art will clearly understand that the specific working process of the systems, devices, modules and units described above can be referred to the corresponding process in the foregoing method embodiments. For the sake of brevity, it will not be repeated here.

[0130] Furthermore, the functional units in the various embodiments of the present invention can be physically independent of each other, or two or more functional units can be integrated together, or all functional units can be integrated into one processing unit. The integrated functional units described above can be implemented in hardware, or in software or firmware.

[0131] Those skilled in the art will understand that if the integrated functional unit is implemented in software and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or all or part of it, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computing device (e.g., a personal computer, server, or network device) to execute all or part of the steps of the methods described in the embodiments of the present invention when running the instructions. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] Alternatively, all or part of the steps of the foregoing method embodiments can be implemented by hardware (such as a computing device, personal computer, server, or network device) related to program instructions. The program instructions can be stored in a computer-readable storage medium. When the program instructions are executed by the processor of the computing device, the computing device executes all or part of the steps of the methods described in the various embodiments of the present invention.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that within the spirit and principles of the present invention, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the corresponding technical solutions to depart from the protection scope of the present invention.

Claims

1. A three-dimensional positioning method for array elements, characterized in that, The method includes: Acquire an image frame sequence for the array element to be tested; the image frame sequence includes high-bit-depth grayscale image frames from multiple perspectives. An adaptive bit-depth mapping enhancement technique is used to dynamically adjust the range of each high bit-depth grayscale image frame to generate a low bit-depth image. For each of the low-bit depth images, an initial candidate label detection is performed to obtain candidate encoded visual labels, and a corresponding ROI polygon region is generated for each of the candidate encoded visual labels. Local enhancement techniques based on candidate boundaries are used to perform local mapping enhancement processing on each of the ROI polygon regions to obtain enhanced ROI images; The enhanced ROI images are subjected to secondary refinement detection, and the target encoded visual label containing the target corner points and the target single-frame pose is output. The dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region is used as the observation quality index to determine the corner observation covariance of the target corner point. An incremental factor map is constructed with the target single-frame pose as the state variable and the reprojection error of the target corner point as the observation constraint, and an adaptive noise-weighted multi-frame joint optimization objective based on the corner point observation covariance is defined. Based on the multi-frame joint optimization objective, the incremental factor map is incrementally updated and solved to obtain the center pose of the array element and its three-dimensional coordinates in the world coordinate system.

2. The method according to claim 1, characterized in that, The step of acquiring the image frame sequence for the array element to be tested includes: The image acquisition device performs a preset trajectory movement relative to the array element under test to acquire multiple high-bit-depth grayscale image frames from multiple perspectives and in succession; the bit depth of the high-bit-depth grayscale image frames is 10 bits to 16 bits. Each of the high-bit deep grayscale image frames and its corresponding metadata is stored to generate an image frame sequence; The metadata includes at least one of frame number, timestamp, and image size.

3. The method according to claim 1, characterized in that, The step of using adaptive bit-depth mapping enhancement technology to dynamically adjust the range of each of the high bit-depth grayscale image frames to generate a low bit-depth image includes: Gaussian denoising is performed on each high bit depth grayscale image frame in the image frame sequence to obtain a denoised image. The denoised image is then cropped to the effective bit depth range to obtain an effective bit depth image. The pixel intensity histograms of each effective bit depth image are statistically analyzed on the downsampling grid, and the intensity clipping interval is determined based on the pixel intensity distribution. The mathematical expression is as follows: in, Indicates the lower limit of strength cutting; Indicates the upper limit of strength reduction; Represents the quantile operator; Indicates the lower quantile parameter, For higher quantile parameters; The image gradient magnitude is calculated based on the intensity cropping interval, and the edge threshold is determined according to the gradient magnitude distribution. The mathematical expression is as follows: in, Represents the set of image gradient magnitudes; This represents the gradient quantile parameter; Represents the gradient quantile operator; Indicates the edge threshold; The edge pixels whose pixel intensity is within the intensity clipping interval and whose gradient magnitude reaches the edge threshold are aggregated to generate an edge pixel set. The mapping intensity window is determined based on the pixel intensity distribution of the edge pixel set, and the mathematical expression is as follows: in, Indicates the lower bound of the mapping strength; Indicates the upper limit of the mapping strength; Represents the set of edge pixels; Represents the pixel intensity values ​​of the input valid bit-depth image; This represents the low-quantile parameter of the edge pixels; This represents the high-resolution parameter of the edge pixels; , Represents the edge pixel quantile operator; A linear mapping based on a mapping intensity window combined with saturation cropping is used to convert the effective bit-depth image into a low bit-depth image with a bit depth of 8 bits. The mathematical expression is as follows: in, This represents the pixel intensity value of the output low-bit-depth image; Represents the pixel intensity values ​​of the input valid bit-depth image; Indicates the lower bound of the mapping strength; This indicates the upper limit of the mapping strength.

4. The method according to claim 1, characterized in that, The step of performing initial candidate label detection on each of the low-bit-depth images to obtain candidate encoded visual labels, and generating corresponding ROI polygon regions for each candidate encoded visual label, includes: A square-coded visual marker detector is used to detect candidate markers in each of the low-bit-depth images and outputs the candidate coded visual markers; the candidate coded visual markers include at least a marker number, candidate corner points, candidate boundaries, and the geometric center of the candidate boundaries; Using the geometric center of the candidate boundary as a reference, the size of the candidate boundary of the low-bit depth image is expanded to obtain the corresponding ROI polygon region: in, The corner points of the expanded ROI polygon region; The candidate corner points that constitute the candidate boundary; Indicates the expansion factor; Indicates the geometric center location of the candidate boundary.

5. The method according to claim 3, characterized in that, The process of using a candidate boundary-based local enhancement technique to perform local mapping enhancement on each of the ROI polygon regions to obtain an enhanced ROI image includes: Within the ROI polygonal region, the pixel intensity distribution of the effective bit depth image is statistically analyzed to obtain local mapping parameters. The dynamic range of the ROI is then calculated based on these local mapping parameters, as expressed mathematically below: in, Indicates the dynamic range of ROI; Indicates the upper limit of local mapping intensity; Indicates the lower bound of local mapping intensity; Represents the set of pixels within the polygonal region of the ROI; Represents the pixel intensity value within the polygonal region of the ROI; High-quantile parameters representing pixel intensity within the polygonal region of the ROI; The low quantile parameter represents the pixel intensity within the polygonal region of the ROI; , Represents the local pixel quantile operator; Based on the dynamic range of the ROI, local mapping enhancement is performed on the polygonal region of the ROI to obtain an enhanced ROI image, as expressed mathematically below: in, This represents the pixel intensity value of the output enhanced ROI image; Represents the clipping function; This represents the pixel intensity value within the input ROI polygonal region; Indicates the lower bound of local mapping intensity; Indicates the dynamic range of ROI.

6. The method according to claim 1, characterized in that, The secondary refinement detection of each enhanced ROI image, outputting a target-encoded visual label containing target corner points and target single-frame pose, includes: In edge refinement mode, the enhanced ROI image is processed with a confirmation output for the candidate encoded visual markers, which is then used as the target encoded visual markers; or, In the edge refinement mode, for the candidate marker detection of the enhanced ROI image, the encoded visual markers that are detected by both the initial candidate marker detection and the secondary refinement detection are output as the target encoded visual markers. The target coded visual marker includes at least the target marker number, target corner point, target geometric center, and target single-frame pose; the target single-frame pose includes the coded visual marker pose and the image acquisition device pose.

7. The method according to claim 1, characterized in that, The step of using the dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region as an observation quality index to determine the corner observation covariance of the target corner point includes: The measurement noise parameters of the target corner point are determined using the following piecewise function: in, , They represent , Standard deviation of corner pixel noise in the direction; Represents the clipping function; , They represent , The linear intercept parameter of the direction; , They represent , The linear slope parameter of the direction; Indicates the dynamic range of ROI; This represents the minimum threshold value for the noise standard deviation. This represents the maximum threshold value for the noise standard deviation; Based on the measured noise parameters, the corner observation covariance corresponding to each target coded visual marker is determined, and the mathematical expression is as follows: in, Indicates the first The corner observation covariance corresponding to each target encoded visual marker; , They represent the first A target-coded visual marker in , Standard deviation of corner pixel noise in the direction.

8. The method according to claim 6, characterized in that, The construction of the incremental factor map, with the target single-frame pose as the state variable and the reprojection error of the target corner points as the observation constraint, and the definition of an adaptive noise-weighted multi-frame joint optimization objective based on the corner point observation covariance, includes: The reprojection error is defined mathematically as follows: in, Indicates reprojection error; , representing the observed pixel coordinates of the target corner point; Indicates the intrinsic parameters of the image acquisition device; Indicates the pose of the image acquisition device; Represents the encoded visual marker pose; Indicates the three-dimensional coordinates of the corner point in the marked coordinate system; Construct an incremental factor map with the pose of the image acquisition device and the pose of the encoded visual marker as state variables and the reprojection error as observation constraint; as well as, A multi-frame joint optimization objective is constructed by weighting the reprojection error using the corner observation covariance and introducing a robust kernel function. The mathematical expression is as follows: in, Represents a robust kernel function; Indicates the first The pose of the image acquisition device corresponding to the frame image; Indicates the first The coded visual marker pose corresponding to each target coded visual marker; Indicates the first The first frame image corresponding to The first target encoding visual mark Reprojection error at each corner point; Indicates the first The corner observation covariance corresponding to each target encoded visual marker; This represents the constraint residual between poses of adjacent frames; For the corresponding covariance; , They represent the first The prior residuals and corresponding covariances of each target encoding visual tag during initialization.

9. The method according to any one of claims 1 to 8, characterized in that, The incremental update solution of the incremental factor map based on the multi-frame joint optimization objective, to obtain the element center pose and its three-dimensional coordinates in the world coordinate system, includes: An incremental factor graph solver based on incremental smoothing and mapping algorithms is used. When the observation quality index of each acquired image frame meets the preset quality conditions, the corresponding target coded visual marker is added to the incremental factor graph, and an incremental update solution is performed on the incremental factor graph to obtain the optimized pose of each target coded visual marker in the world coordinate system. If the reference origin of the target encoded visual marker coincides with the center of the element to be tested, then the three-dimensional coordinates of the element to be tested in the world coordinate system are taken from the translation vector of the optimized pose. If there is a fixed rigid offset between the reference origin of the target encoded visual marker and the center of the array element to be tested, the array element center pose of the array element to be tested is calculated according to the pre-calibrated fixed rigid transformation, and the three-dimensional coordinates of the array element to be tested in the world coordinate system are calculated according to the rotation matrix and translation vector of the optimized pose.

10. A three-dimensional positioning system for array elements, characterized in that, The system includes: An image acquisition module is used to acquire an image frame sequence for the array element to be tested; the image frame sequence includes high-bit-depth grayscale image frames from multiple perspectives. The bit depth processing module is used to dynamically adjust the range of each of the high bit depth grayscale image frames using adaptive bit depth mapping enhancement technology to generate a low bit depth image. The initial detection module is used to perform initial candidate label detection on each of the low bit depth images to obtain candidate coded visual labels, and generate corresponding ROI polygon regions for each candidate coded visual label; The ROI enhancement module is used to perform local mapping enhancement processing on each of the ROI polygon regions using a candidate boundary-based local enhancement technique to obtain an enhanced ROI image. The secondary detection module is used to perform secondary refinement detection on each of the enhanced ROI images and output a target-coded visual label containing the target corner point and the target single-frame pose. The quality observation module is used to determine the corner observation covariance of the target corner point by using the dynamic range of the ROI obtained by performing pixel intensity statistics on the ROI polygon region as the observation quality index. The fusion optimization module is used to construct an incremental factor map with the target single-frame pose as the state variable and the reprojection error of the target corner point as the observation constraint, and to define a multi-frame joint optimization objective based on the corner point observation covariance with adaptive noise weighting. The solution output module is used to incrementally update the incremental factor map based on the multi-frame joint optimization objective to obtain the center pose of the array element and its three-dimensional coordinates in the world coordinate system.

Images