Vision-based target contour reconstruction and in-plane track measurement method and system

Abstract

The invention provides a target contour reconstruction and in-plane track measurement method and system based on vision. The method comprises the steps that a polygon target is fixedly installed on a target to be measured, a camera is fixed for imaging, and a camera coordinate system is established; performing coarse precision matching on the target through a cross-correlation algorithm to obtain an ROI (Region of Interest), and iteratively updating a matching position in a time sequence video frame; sub-pixel-level resolution improvement and binarization segmentation are carried out on the ROI, and a target edge contour is extracted; performing linear fitting on the edge point set based on a gradient space point set clustering algorithm, and reconstructing a target contour; and converting the pixel displacement into a physical displacement amount, and calculating an attitude angle in the target plane based on the angular point position. According to the invention, the dynamic contour of the target is reconstructed by using the camera, and the in-plane displacement and attitude high-precision measurement is carried out; while high precision and high environmental adaptability are ensured, miniaturization, low power consumption and high efficiency of measurement equipment are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of in-plane trajectory measurement of a target, and specifically, to a method and system for target contour reconstruction and in-plane trajectory measurement based on vision. Background Art

[0002] Dynamic target tracking technology has a wide range of applications in the fields of structural health monitoring (SHM), manufacturing process monitoring, aerospace, etc. The time-series displacement information of each target point plays an important role in structural safety assessment, long-term status verification, etc. In contact measurement schemes, when using a triaxial accelerometer to measure specific points, wiring and networking are required, which is inconvenient to operate, and attaching it to the surface of the object to be measured is likely to change its mechanical properties. Non-contact measurement schemes for large structures mainly include: active measurement schemes and vision-based schemes. Taking a laser Doppler vibrometer as an example, it obtains displacement and velocity data through the Doppler frequency modulation generated by the vibration of the target. However, such single-point laser instruments are usually large in size and high in energy consumption, and it is difficult to meet the requirements of multi-target synchronous tracking in complex system monitoring.

[0003] In-plane trajectory measurement of a target based on vision is a new type of measurement technology. In existing vision-based in-plane trajectory measurement methods, the optical flow method and digital image correlation method are easily affected by light changes, and the template matching method is likely to fail when the visual form of the target changes. When using a biaxial accelerometer to measure specific points, wiring and networking are required, which is inconvenient to operate, and attaching it to the surface of the object to be measured is likely to change its mechanical properties; the laser Doppler vibrometer is large in size and high in energy consumption, and multiple devices need to cooperate for measurement. In existing vision-based trajectory measurement methods, the optical flow method and digital image correlation method are easily affected by light changes, and the template matching method is likely to fail when the visual form of the target changes. Therefore, the stability of in-plane trajectory detection of the target in a complex environment is weak. Therefore, there is an urgent need for a method and system for target contour reconstruction and trajectory measurement based on vision.

[0004] Patent application document CN116007505A discloses a method, device and computer equipment for correcting the displacement of a target by monocular vision measurement. The method includes: acquiring an image of a target area; binarizing the image of the target area to obtain a binarized image; extracting the target contour from the binarized image to obtain the target contour coordinates; calculating the pixel coordinates of the four corner points of the target according to the target contour coordinates; calculating the angle between the target and the camera plane according to the pixel coordinates of the four corner points of the target; and correcting the displacement of the target according to the angle between the target and the camera plane to obtain a correction result. However, this patent cannot completely solve the existing technical problems. It is difficult to correctly identify when the target contour is unclear and cannot meet the requirements of the present invention. Summary of the Invention

[0005] Aiming at the defects in the prior art, the purpose of the present invention is to provide a vision-based target contour reconstruction and in-plane trajectory measurement method and system.

[0006] According to the vision-based target contour reconstruction and in-plane trajectory measurement method provided by the present invention, it includes:

[0007] Step 1: Install a fixed polygon target on the target to be measured, use a camera to image the target, and establish a camera coordinate system;

[0008] Step 2: Conduct rough matching and resolution improvement of the target in each frame of the image;

[0009] Step 3: Extract the rough contour of the target edge from the matched image area;

[0010] Step 4: Extract multi-dimensional gradient information from the edge point set, cluster the point set of the formed gradient coordinate system, perform line fitting, and reconstruct the target contour;

[0011] Step 5: Convert the pixel displacement of this frame into a physical displacement, and calculate the in-plane attitude angle of the target based on the corner position;

[0012] Step 6: Repeat Steps 2 to 5 for each frame of the video to reconstruct the dynamic contour change, displacement, and attitude time series information of the target.

[0013] Preferably, the rough matching of the target in Step 2 includes methods such as template matching based on cross-correlation and normalized cross-correlation.

[0014] Preferably, the extraction of the rough contour of the target edge in Step 3 includes methods such as separating the target shape using binary segmentation and judging pixel attribution using multi-connected regions to extract the contour.

[0015] Preferably, the extraction of gradient information from the edge point set in Step 4 specifically includes:

[0016] Using methods including a filter kernel and an edge detection operator to calculate the gradient direction component of the contour point position, such as G x and G y etc., taking them as coordinate values, constructing a gradient space point coordinate system G(I,j), which can be expressed as:

[0017] G(i,j) = G x (i,j) + i·G y (i,j)

[0018] The clustering of the formed gradient coordinate system point set in Step 4 specifically includes, but is not limited to, the following methods:

[0019] Search radius r and iteration step d for dense areas rUpdate the clustering center until ‖d r -d r-1 ‖ < t1, where t1 is a preset threshold;

[0020] Remove noise points through the outlier filtering threshold, and fit each edge line using the least squares method.

[0021] Preferably, the expression for converting the pixel displacement of this frame to the physical displacement in step 5 is:

[0022] The actual displacement of the target is converted through the scale factor:

[0023]

[0024] where D t represents the actual side length of the target, and I t represents the pixel side length of the target.

[0025] According to the vision-based target contour reconstruction and in-plane trajectory measurement system provided by the present invention, it includes:

[0026] A video signal sensing module for collecting dynamic sequential images of the target;

[0027] An image preprocessing module, including a rough target position localization unit, a resolution enhancement unit, and a rough edge contour extraction unit;

[0028] A contour reconstruction and trajectory calculation module for gradient point clustering, line fitting, and conversion of pixel displacement to physical displacement;

[0029] A data display and storage module for outputting the target displacement, attitude angle, and error analysis results.

[0030] Preferably, the contour reconstruction and trajectory calculation module includes:

[0031] A video stream data processing unit for actually calculating the contour of each target and the current trajectory of each frame;

[0032] A multi-target parallel processing unit that supports contour reconstruction and trajectory measurement of at least 4 independent targets simultaneously.

[0033] Preferably, the system is further integrated into an embedded device, including:

[0034] A low-power image processor for real-time processing of video stream data;

[0035] A wireless transmission module for sending the measurement results to a remote monitoring terminal;

[0036] An anti-interference housing, encapsulated with electromagnetic shielding materials, suitable for industrial complex environments.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. The present invention extracts the original contour by binarizing and segmenting the target image, maps the contour point set to the gradient coordinate system in combination with the edge gradient information, and obtains the parameters of each edge line by clustering to fit the target contour information, solving the problem of target tracking in complex working conditions. On the basis of measuring the in-plane displacement, it provides the dynamic contour change information of the target, solving the problem that a single camera cannot analyze the change of the target state, including but not limited to the limitations of the visual detection method being prone to failure under working conditions such as the attitude change of the multi-point target, the decrease in clarity, and the interference of environmental changes.

[0039] 2. The present invention extracts effective information through a multi-process optimization method, filters out noise interference, realizes high-precision and high-stability contour extraction. Through steps such as binarization segmentation, gradient point set clustering, and outlier filtering, it gradually extracts effective information to exclude interference and establishes a new method for coordinate mapping of time-series information, solving the limitation of the existing visual measurement technology being prone to environmental interference when relying on the original pixel information for calculation.

[0040] 3. The present invention establishes a method and system for target contour reconstruction and in-plane trajectory measurement based on vision, dynamically tracks the contour shape of the target according to the time-series video information, proposes a high-precision pixel-physical displacement conversion method and an angle attitude measurement method, effectively solves the problem of high-precision measurement of the target state by a single camera under complex working conditions, has strong adaptability under the influence of noise, provides a multi-dimensional, high-precision, and anti-interference solution method in the field of visual measurement, and realizes the miniaturization, low power consumption, and high efficiency of the measurement device while ensuring high precision and high environmental adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] By reading the following detailed description of the non-limiting embodiments with reference to the accompanying drawings, other features, objects, and advantages of the present invention will become more apparent:

[0042] Figure 1 It is a flowchart of the method for target contour reconstruction and in-plane trajectory measurement based on vision of the present invention;

[0043] Figure 2 It is a schematic diagram of the target contour reconstruction method of the embodiment of the present invention;

[0044] Figure 3 It is a schematic diagram of the experimental test scenario of the embodiment of the present invention;

[0045] Figure 4 (a) The target is fixed on the sliding table and reciprocates 10 mm in the X direction; Figure 4 (b) The target is fixed on the sliding table and reciprocates 6 mm in the Z direction; Figure 4 (c) The target is fixed on the sliding table and performs two-dimensional movement in the X-Z plane;Figure 4 (d) After setting the rotation angle, extract the change in its rotation angle according to the target trajectory;

[0046] Figure 5 This is the structural block diagram of the vision-based target contour reconstruction and in-plane trajectory measurement system of the present invention. Detailed implementation manners

[0047] The present invention will be described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several changes and improvements can still be made. These all belong to the protection scope of the present invention.

[0048] Embodiment

[0049] The present invention provides a vision-based target contour reconstruction and trajectory measurement method, including:

[0050] As Figure 1 shown, first, install a fixed polygon target on the object to be measured or the measurement point; then, install a fixed camera within a distance that ensures a high resolution of the object to be measured in the pixel space and align it with the object to be measured, and keep it stationary to establish a camera coordinate system; then control the camera to record the sequential motion sequence of the target; select the region of interest (ROI) from the first frame of the video, and traverse and find the matching position at the beginning of each subsequent frame. For the region of interest of each frame, perform binary segmentation operation to extract the original edge contour of the target, extract the gradient information on the original image for each point of the edge contour and perform gradient coordinate system point clustering to obtain a point set that can be mapped to the edge straight line; finally, reconstruct the target contour in this frame according to the parameters of each side of the target; according to the positions of each corner point and the center point, combined with the previous displacement information, convert the pixel displacement to the actual displacement value; use the inverse trigonometric function to convert the relative position of the corner point to the rotation angle around the axis.

[0051] The specific process is as follows:

[0052] Step 1, single-pixel accuracy matching of the target position;

[0053] Step 1.1, install a polygon target on the object to be measured or the measurement point;

[0054] Step 1.2, install a fixed camera at a distance where the target is clearly imaged to form a camera coordinate system;

[0055] Step 1.3, perform rough accuracy target position matching based on the cross-correlation algorithm: S(x, y) represents the image to be searched with size [m, n], i and j are the indices of m and n respectively, g(x, y) represents the template image, S x,y and g respectively represent the average values of the currently selected region and the template;

[0056] Taking the minimum bounding box of the target in the first frame as a reference, calculate the cross-correlation coefficient matrix R(x, y) of the target template in the search area of each frame, where each pixel corresponds to a unique correlation coefficient value; the maximum value in the coefficient matrix is the rough-precision matching point in that frame, and this bounding box is the region of interest;

[0057] The expression of R(x, y) is:

[0058]

[0059] Due to the attitude change of the target, it is necessary to set the mis-matching discrimination condition during the matching process. Preferably, the matching result of the current frame after misjudgment can be determined by superimposing the displacement of the target pixels in the previous frame and the matching result of the previous frame.

[0060] Step 2, preprocessing the image of the region of interest;

[0061] Step 2.1, sub-pixel resolution improvement;

[0062] Use the interpolation algorithm to improve the resolution of the target area after rough positioning. The value of each new sub-pixel point is the weighted average of the original pixel points in the n×n neighborhood according to the weight;

[0063] Bilinear interpolation processing: Use the bilinear interpolation algorithm to improve the resolution of the target area after rough positioning. The value of each new pixel point is the weighted value of the linear distance of the four original pixel points in the neighborhood. In the following weight formula, f(Q ij ) is the coordinate number of the original reference point of the academician, and x and y are the coordinates of the mapping point:

[0064]

[0065] Bicubic interpolation processing: Use the bicubic interpolation algorithm to improve the resolution of the target area after rough positioning. The value of each new pixel point is the weighted average of the original pixel points in the n×n neighborhood. In the following weight formula, x is the distance between the mapping point and the original point, and a is the shape adjustment factor:

[0066]

[0067] Step 2.2, binary segmentation of the target;

[0068] Perform binary segmentation on the ROI with improved resolution, and distinguish the target and background regions through a certain threshold;

[0069] For the ROI with enhanced resolution, global OTSU algorithm is used for binary segmentation, and the segmentation threshold is determined by maximizing the color difference variance σ between the target and background regions. The calculation of σ is as follows, where M is the global mean, p1 is the proportion of foreground pixels, and m1 and m2 are the mean values of foreground and background pixels respectively. Preferably, the threshold traversal range is 0 - 255 gray levels, and the value that maximizes the foreground-background variance is taken as the segmentation boundary:

[0070]

[0071] p1*m1+(1-p1)*m2=1

[0072]

[0073] In addition to the above specific embodiments, for the segmentation of the target, when there are light changes within the ROI, an adaptive threshold segmentation method can be used. The image is smoothed using a Gaussian filter of size r, scaled by a factor s, and then the smoothed image is compared with the original image. Pixels with brightness values lower than the original image are classified as foreground, while other pixels are assigned as background.

[0074]

[0075] Step 2.3, Binary image preprocessing and target edge contour extraction;

[0076] Connected component processing: Define that pixel points (I) connected by a 4-connected path (P) belong to the same object, where the Manhattan distance d between adjacent pixels M =1:

[0077] d M ((x i ,y i ),(x i+1 ,y i+1 ))=∣x i -x i+1 ∣+∣y i -y i+1 ∣=1

[0078]

[0079] The target sub-patterns segmented as background and the background interference items segmented as targets are respectively fused with and suppressed by the target main body, and the contour edge pixels are extracted using the 4-connected discrimination criterion N4. All points where all 4-connected domain points are 1 are used as internal pixel points with a value of 0.

[0080] Step 3, Edge point set gradient information clustering and line fitting;

[0081] Step 3.1, Original image preprocessing and directional gradient convolution calculation;

[0082] Perform directional convolution operation on the extracted, interpolated, and enlarged image to extract the gradient values of each contour point. Among them, I(I, j) represents the image after resolution enhancement, and Ga represents a smoothing filter with a size of r, which is used to suppress noise. The directional gradient operator S with a size of n is used x and S y to perform two-dimensional convolution operations respectively to obtain the gradient components G x and G y ;

[0083] In the standard test scenario, a Sobel operator with a size of 3 can be selected to effectively capture fine edge features:

[0084]

[0085]

[0086] X = (x1, x2)

[0087]

[0088] In addition to the above specific embodiments, the extraction of the gradient values of each edge point can be performed using methods such as the Roberts operator and the Prewitt operator:

[0089] Roberts operator:

[0090] Prewitt operator:

[0091] Step 3.2, construction of the gradient space point coordinate system;

[0092] Take the gradient components of each contour point, including G x and G y etc., as coordinate values to construct the gradient space point coordinate system G(I, j), and the expression is:

[0093] G(i, j) = G x (i, j) + i * G y (i, j)

[0094] The coordinates of each point represent the position distribution of a specific point on the target contour in the gradient space.

[0095] Step 3.3, gradient point aggregation clustering based on mean shift;

[0096] Perform density center (clustering) search on the gradient space point set G using the mean shift algorithm. The n-sided polygon target needs to finally search for n independent clusters. Each cluster represents the mapping of the point set at that place to an edge line. In this step, C idenotes the current cluster center, and S represents the set of points whose distance from C i is less than r; r is the search radius for dense regions, and d r is the search step size, and t1 and t2 are the iteration termination thresholds;

[0097] Step 3.3.1, traverse all points to make them visited. Select an unmarked data point as the starting center C i representing the i-th category;

[0098] Step 3.3.2, when ∥d r -d r-1 ∥>t1, select an unmarked data point as the starting center C i representing the i-th category; the search points satisfy: S = {x: ∥(x - C i )∥ < r 2 |x ∈ G}; calculate the new center C by the mean shift formula i = C i + d r update C i ; r = r + 1; repeat Step 3.3.2;

[0099] Step 3.3.3, if min ∥C i - C k ∥>t2, for k = 1, 2,..., i - 1, increase the visit count of the circled points under the new category i;

[0100] Otherwise, execute Add the visit counts of the circled points involved back to category k;

[0101] Repeat starting from 3.3.1 until all points are visited;

[0102] In addition to the above embodiments, clustering can be performed by means of density center search. For each C i point, calculate the number of points in the set whose distance is less than r as the point density within this range. Among all the points, the top n points in the density descending order are regarded as the cluster centers.

[0103] Step 3.3.4, sort each point into the corresponding category with the most visit counts;

[0104] Step 3.3.5, select n categories with the most points;

[0105] Step 3.4, noise point filtering and line fitting;

[0106] During the clustering process of the gradient space point set coordinate system, there are interference points that do not belong to the point set mapped by the line segment, which need to be verified and filtered; this step describes the data refinement and sorting process of the selected clustering points and the mapped edge lines, and filters out the noise points in each point clustering; preferentially select the point set in the flattest and least noisy section of each edge line as the initial fitting data; r 01 to r 04 As the threshold for outlier filtering, it is affected by target characteristics (such as pixel size and current direction); these parameters are crucial for the stability of target tracking and should be set to appropriate values, which can be determined through iterative adjustment and preliminary test data analysis;

[0107] Step 3.4.1, if the median of the formula ||(I,j)-median(I,j)|| is greater than r 01 :

[0108] Remove the outliers from the following datasets:

[0109] Retain the data points that satisfy {I k (I,j)|||(I,j)-median(I,j)||<r 02 condition;

[0110] Step 3.4.2, if the mapped edge line is horizontal or vertical, remove the outliers from the following datasets: ||j-median(j)||>r 03 or ||i-median(i)||>r 04 ;

[0111] Step 3.4.3, sort the phase angle ∠C represented by the clustering center k from 0 to 2π;

[0112] Step 3.4.4, select k = 1, 2, …, N as the top horizontal edge lines;

[0113] Step 3.4.5, sequentially select the clustering point sets mapped to each adjacent side in the order of phase angle.

[0114] Step 3.5, edge straight line fitting;

[0115] For the set of original pixel coordinate points represented by each gradient clustering point set, use the least squares method to fit the original edge line.

[0116] Step 4, target contour restoration and high-precision in-plane trajectory measurement;

[0117] Step 4.1, target contour restoration;

[0118] Based on the edge parameters calculated in step 3.5, calculate that the intersection points of the straight lines are the target corner points, and the intersection points obtained from the connection lines of each group of diagonal points are the target center points;

[0119] Step 4.2, in-plane displacement measurement based on the homography matrix;

[0120] Introduce a new homography transformation, use the relationship between two spatial planes to convert the pixel displacement to the physical displacement amount. This method corrects the visual distortion of the target and restores it from an inclined angle to an upright angle. Therefore, the displacement of the target center in the camera coordinate system can be accurately converted into a vector in the target plane, P c and P w represent the corner coordinates in the camera coordinate system and the world coordinate system (defined on the target plane). The homography matrix A(t) maps these two coordinate systems. Although the matrix has 9 parameters, it has only 8 degrees of freedom, meaning that it can be determined using the four corner coordinates of the polygonal target;

[0121] P w (t) = A(t)·P c (t)

[0122]

[0123] For each frame, the difference between the projected center coordinate P w (t) and the value of the previous frame provides its pixel displacement vector in the front view. Define a scale factor to convert this discrete displacement vector into a physical displacement, where D t represents the actual side length of the target, and I t represents the pixel side length of the target:

[0124]

[0125] Calculate the in-plane displacement time series of the target center point along the coordinate system constructed in step 1.2:

[0126] D(t i ) - D(t i-1 ) = (P w,center (t i ) - P w,center (t i-1 ))·SF

[0127] In addition to the above specific embodiments, use SF to complete the conversion from pixel displacement to physical displacement amount according to the target side length in each frame combined with the balance parameters manually selected in the starting frame. D and Δd respectively represent the target displacement and the displacement increment vector. It should be noted that L i (t1) represents the target contour manually selected in the first frame. Ω represents the line segment used for measurement and its corresponding corner coordinates:

[0128] D(t i+1 ) = D(t i ) + Δd(t i+1 ) * SF * Bal

[0129]

[0130] Ω = {(i, j) | i ∈ {a, b, …}, j = i + 1}

[0131] Here, the lines in Ω should maintain the same length in the current test scenario, without visual compression caused by target movement, such as vertical edges during horizontal movement, etc.

[0132] Step 4.3, in-plane attitude measurement based on corner positions;

[0133] The measurement scene diagram is as Figure 3 shown. Using the corner points P identified in each frame, the pitch angle θ of the target is determined through inverse trigonometric functions p and the rotation angles θ around the Y and Z axes r , and the pitch angle θ p is derived from the slope of the top horizontal edge; L1, P left , P right respectively represent the first top edge in the clockwise direction and its two corners, expressed as:

[0134]

[0135] For a square target, the rotation angles θ around the Y and Z axes r are calculated using the ratio of the distances between parallel side lines. L i refers to each edge line counted clockwise, and the distance d(L I , L j ) between parallel lines is defined as the shortest distance between line segment points and is expressed as:

[0136]

[0137] d(L i , L j ) = min ∥P(L i ) - P(L j ) ∥

[0138] Based on the target contour reconstruction method and in-plane trajectory measurement technology provided by the present invention, Figure 3 a two-degree-of-freedom displacement system around the X and Z axes and a rotation system around the Y axis are shown. Using a square target and a background designed as a highly reflective copper sheet, random shadows and bright spots can be generated around the target, forming a changing bright and dark block background; Figure 4Show the analysis results that the system can still accurately measure the trajectory of the target under displacements and attitude changes in various modes. In the experiment, reciprocating displacements with periodic amplitudes of 10 mm and 6 mm in the X and Z directions, in-plane displacements, and rotational motions with an inclination angle of 20° in the two-dimensional plane were respectively set for the target; using the above method, the target contour was reconstructed, and the displacements of the center point and the changes in the attitude angles were measured; in the X, Z, and two-dimensional displacement experiments, the root mean square errors of the three motion modes were 0.0555, 0.0762, and 0.1264 (mm), respectively. For the rotational motion of the target, this method accurately restored the curve of the rotational angle change.

[0139] The vision-based target contour reconstruction and in-plane trajectory measurement system, its structural block diagram is as Figure 5 shown, including:

[0140] A video signal sensing head, used to collect the dynamic changes of the target and transmit the signal to the computer;

[0141] A video signal processing module, including an image signal preprocessing unit and a contour reconstruction and trajectory measurement unit;

[0142] The processing content of the image signal preprocessing unit includes receiving the sequential video signal transmitted by the sensing head, and separating the target based on the coarse positioning technology of the correlation coefficient matrix and the target-background separation binary technology for anti-interference;

[0143] The contour reconstruction and trajectory measurement unit, including a target contour reconstruction, in-plane displacement measurement, and attitude measurement unit, uses the processing results of the image signal preprocessing unit to realize the target contour reconstruction and trajectory measurement of the sequential sequence through the above steps 3 to 4;

[0144] A data display and storage module, used to display and store intermediate processing and final measurement data as needed;

[0145] A reference target module, used to install or arrange the points to be measured and form a world coordinate system based on the plane.

[0146] The vision-based target contour reconstruction and trajectory measurement method and system proposed by the present invention have the following advantages: The present invention breaks through the limitations of the existing vision-based measurement methods that are sensitive to light and prone to failure under noise interference, and introduces a binary segmentation and gradient point clustering algorithm to realize the extraction of high-noise-resistant target contours. The present invention overcomes the defect that the existing methods provide less detection information dimensions, and proposes a complete target contour reconstruction method. The present invention realizes a miniaturized, low-power, and high-precision planar trajectory detection system through the combination of vision and the pixel-physical displacement conversion method.

[0147] Those skilled in the art know that, in addition to implementing the systems, devices and their respective modules provided by the present invention in the form of pure computer-readable program code, it is entirely possible to logically program the method steps so that the systems, devices and their respective modules provided by the present invention are implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same program. Therefore, the systems, devices and their respective modules provided by the present invention can be considered as a kind of hardware components, and the modules included therein for implementing various programs can also be regarded as the structures within the hardware components; the modules for implementing various functions can also be regarded as either software programs for implementing methods or structures within hardware components.

[0148] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A vision-based target contour reconstruction and in-plane trajectory measurement method, characterized in that Including: Step 1: Install a fixed polygonal target on the target to be measured, use a camera to image the target, and establish a camera coordinate system; Step 2: Conduct rough matching and resolution improvement of the target in each frame of the image; Step 3: Extract the rough contour of the target edge from the matched image area; Step 4: Extract multi-dimensional gradient information from the edge point set, cluster the point set of the composed gradient coordinate system, conduct linear fitting, and reconstruct the target contour; Step 5: Convert the pixel displacement of this frame into a physical displacement, and calculate the in-plane attitude angle of the target based on the corner position; Step 6: Repeat Steps 2 to 5 for each frame of the video to reconstruct the dynamic contour change, displacement, and attitude time series information of the target.

2. The vision-based target contour reconstruction and in-plane trajectory measurement method according to claim 1, wherein The rough matching of the target in Step 2 includes using template matching methods based on cross-correlation and normalized cross-correlation.

3. The vision-based target contour reconstruction and in-plane trajectory measurement method according to claim 2, wherein Extracting the rough contour of the target edge in Step 3 includes separating the target shape using binary segmentation and determining pixel attribution using multi-connected domain to extract the contour.

4. The vision-based target contour reconstruction and in-plane trajectory measurement method according to claim 3, wherein Extracting multi-dimensional gradient information from the edge point set in Step 4 specifically includes: Calculate the gradient direction component G of the contour point positions using a method including a filtering kernel and an edge detection operator x and G y , take them as coordinate values, and construct a gradient space point coordinate system G(I, j), and the expression is: G(i,j) = G x (i,j) + i·G y (i,j) Clustering the point set of the composed gradient coordinate system in Step 4 includes: With a dense area search radius r and an iteration step size d r Update the cluster center until ‖d r -d r-1 ‖ < t1, where t1 is a preset threshold; Removing noise points through an outlier filtering threshold and using the least squares method to fit each edge line.

5. The method for reconstructing the target contour and measuring the in-plane trajectory based on vision according to claim 4, characterized in that, In the process of converting the pixel displacement of this frame into a physical displacement in Step 5, the actual displacement of the target is converted through a scale factor: Among them, D t represents the actual side length of the target, and I t represents the pixel side length of the target.

6. A vision-based target contour reconstruction and in-plane trajectory measurement system, characterized in that, Including: A video signal sensing module for collecting dynamic time series images of the target; An image preprocessing module, including a rough target position localization unit, a resolution improvement unit, and a rough edge contour extraction unit; A contour reconstruction and trajectory calculation module for clustering gradient point sets, linear fitting, and converting pixel displacement to physical displacement; A data display and storage module for outputting the target displacement, attitude angle, and error analysis results.

7. The vision-based target contour reconstruction and in-plane trajectory measurement system according to claim 6, wherein The contour reconstruction and trajectory calculation module includes: A video stream data processing unit that actually calculates the contour and current trajectory of each target in each frame; A multi-target parallel processing unit that supports contour reconstruction and trajectory measurement of at least 4 independent targets simultaneously.

8. The vision-based target contour reconstruction and in-plane trajectory measurement system according to claim 7, characterized in that, The system is further integrated into an embedded device, including: A low-power image processor for real-time processing of video stream data; A wireless transmission module for sending the measurement results to a remote monitoring terminal; An anti-interference housing, encapsulated with electromagnetic shielding materials, suitable for industrial complex environments.

Images