A computer vision-based three-dimensional reconstruction method and system

Abstract

The application belongs to the technical field of three-dimensional imaging, and discloses a three-dimensional reconstruction method and system based on computer vision, which comprises the following steps: using a single industrial camera instead of a right eye in a binocular camera system, shooting a vertical light bar image, and constructing a virtual reference left view by applying an image processing method. Then, the parallax between a real-time right view and the reference left view is solved, and the depth information of the target is obtained according to the "parallax-depth" formula derived by the application. The depth information of all frames is spliced, so that a three-dimensional point cloud model which can clearly show the shape and structural characteristics of the object can be generated. The application only needs to use a single industrial camera, without the need for additional auxiliary equipment, thereby reducing the cost of equipment.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional imaging technology, and in particular relates to a three-dimensional reconstruction method and system based on computer vision. Background Technology

[0002] The binocular stereo vision line laser scanning 3D imaging method uses a binocular camera to simultaneously capture two line laser images and extract the center line of the line laser to calculate depth information. Although this method can effectively acquire 3D information, it has high requirements for image acquisition and computing resources and relies on two high-precision cameras, which increases the overall cost of the system.

[0003] To address the aforementioned problems, this invention proposes a line laser scanning 3D imaging scheme based on virtual binocular stereo vision. Summary of the Invention

[0004] The purpose of this invention is to provide a computer vision-based three-dimensional reconstruction method and system to solve the problems existing in the prior art.

[0005] To achieve the above objectives, this invention provides a computer vision-based 3D reconstruction method, comprising:

[0006] S1: Fix the line laser generator above the camera's optical axis, with the laser plane coinciding with the camera's optical axis; mount the camera and laser generator together on the slide table;

[0007] S2: Place the object to be measured within the measurement range, use a line laser generator to project a line laser onto the object to be measured, start the slide table to move at a constant speed, and acquire the laser stripe image sequence through a camera;

[0008] S3: Take each image in the laser stripe image sequence as the right view, and construct a virtual left view based on the right view; calculate the disparity between each right view and the corresponding virtual left view;

[0009] S4: Input the parallax into the preset error correction formula and calculate the three-dimensional coordinates of each pixel;

[0010] S5: Stitch together the 3D points of all frames to obtain the final imaging data.

[0011] Optionally, in step S1, the laser plane corresponding to the line laser generator coincides with the optical axis of the camera, and the camera is an industrial camera with pre-calibrated intrinsic parameters.

[0012] Optionally, the construction of a virtual left view based on the right view includes two cases:

[0013] The first method, for cases where the background depth of field remains constant, extracts the center line of the laser stripes in the right view and calculates the horizontal coordinate D at the center line of the image.T ;

[0014] Create a new all-black image with the same size as the right view; insert a D-shaped element on the left side of the all-black image. T Column black pixels, delete D on the right side of the right view. T The portion following the column is copied to the blank area of ​​the all-black image to complete the construction of the virtual left view.

[0015] The second approach, which addresses changes in background depth, involves creating a vertical line passing through the center of the image in the right-eye camera's image and using it as a virtual left view.

[0016] Optionally, the error correction formula in step S4 is as follows:

[0017]

[0018] In the formula, It is a constant. For laser distance measurement, This represents the ranging error.

[0019] Optionally, the derivation process of the error correction formula is as follows:

[0020] Using a line laser generator, project a line laser along the plane normal onto a flat white wall. Move the slide to three preset positions and collect images of the line laser reflected from the wall at three positions at different distances from the wall. At the same time, use a laser rangefinder to record the wall distance measured when the images are collected, and calculate the parallax between each laser stripe image and the corresponding virtual left view.

[0021] The measurement distance is calculated by combining parallax and camera calibration parameters;

[0022] The difference between the distance to the wall recorded by the laser rangefinder and the measured distance is calculated to obtain the distance measurement error;

[0023] The relative error of ranging is calculated based on the ratio of ranging error to the measured distance;

[0024] Using the measured distance as the x-axis and the distance measurement error and relative distance measurement error as the y-axis, a distribution map of the error and relative error as a function of the measured distance is constructed.

[0025] Based on the constructed distribution map, an error correction formula is constructed using curve fitting.

[0026] On the other hand, to achieve the above objectives, the present invention provides a computer vision-based 3D reconstruction system, comprising:

[0027] The image acquisition module is used to fix the line laser generator above the camera optical axis, with the laser plane coinciding with the camera optical axis; to mount the camera and laser generator together on the slide table; to place the object to be measured within the measurement range, to use the line laser generator to project a line laser onto the object to be measured, to start the slide table to move at a constant speed, and to acquire the laser stripe image sequence through the camera.

[0028] The virtual left view construction module is used to take each image in the laser stripe image sequence as a right view, construct a virtual left view based on the right view, and calculate the disparity between each right view and the corresponding virtual left view.

[0029] The 3D reconstruction module is used to calculate the 3D coordinates of each pixel by inputting the parallax into a preset error correction formula; and stitches together the 3D points of all frames to obtain the final imaging data.

[0030] The technical effects of this invention are as follows:

[0031] This invention proposes a line laser 3D imaging method based on virtual binocular stereo vision. This method requires only one industrial camera to achieve measurement results equivalent to traditional binocular systems. It boasts advantages such as low equipment cost and low image processing load. Experimental results demonstrate that this invention can significantly improve point cloud density, achieving approximately a doubling of the density, thereby enabling more accurate reconstruction of object 3D features. 3D reconstruction using the point cloud data obtained by this invention generates 3D models with higher accuracy than traditional binocular stereo vision line laser scanning methods, validating the potential of this method in industrial applications. Attached Figure Description

[0032] Figure 1 This is a flowchart of the virtual binocular stereo vision line laser scanning three-dimensional imaging process in Embodiment 1 of the present invention;

[0033] Figure 2 shows the ideal imaging model of parallel-axis binocular stereo vision in Embodiment 1 of the present invention;

[0034] Figure 3 This is a diagram showing the composition and imaging of the binocular stereo vision line laser scanning system in Embodiment 1 of the present invention;

[0035] Figure 4 This is a schematic diagram of the virtual binocular stereo vision line laser scanning three-dimensional imaging system in Embodiment 1 of the present invention;

[0036] Figure 5 This is a schematic diagram illustrating the construction principle of the virtual left view in Embodiment 1 of the present invention;

[0037] Figure 6 This is a schematic diagram illustrating the construction and method of the left view in Embodiment 1 of the present invention;

[0038] Figure 7This refers to the virtual binocular stereo vision ranging structure error model in Embodiment 1 of the present invention;

[0039] Figure 8 This describes the parameter solving principle of the virtual binocular stereo vision line laser measurement system in Embodiment 1 of the present invention.

[0040] Figure 9 This is a flowchart of the virtual binocular stereo vision line laser distance measurement data acquisition process in Embodiment 1 of the present invention;

[0041] Figure 10 A schematic diagram is created for the left view in Embodiment 1 of the present invention;

[0042] Figure 11 This is the image acquisition experimental platform for the stereo vision line laser scanning imaging system in Embodiment 1 of the present invention;

[0043] Figure 12 The ranging error ∆Z of the image acquisition experimental platform of the stereo vision line laser scanning imaging system in Embodiment 1 of the present invention is... C1 and relative error R ∆ZC1 With measurement distance Z c1 Distribution map;

[0044] Figure 13 This refers to the virtual binocular stereo vision line laser scanning point cloud imaging method in Embodiment 1 of the present invention;

[0045] Figure 14 The distance measurement data within the range of 400~1900 mm in Embodiment 1 of the present invention;

[0046] Figure 15 This is a schematic diagram of the binocular stereo vision line laser scanning imaging process in Embodiment 1 of the present invention;

[0047] Figure 16 shows the imaging point cloud diagrams of the binocular stereo vision line laser scanning imaging method in Embodiment 1 of the present invention and the method described in this paper.

[0048] Figure 17 To draw a virtual left view light bar in the right view of Embodiment 2 of the present invention;

[0049] Figure 18 This is an assembly error imaging model of the laser generator in Embodiment 2 of the present invention;

[0050] Figure 19 This is a schematic diagram of the virtual binocular stereo vision line laser coal pile scanning principle in Embodiment 2 of the present invention;

[0051] Figure 20 This is an image of the virtual binocular stereo vision line laser scanning device and camera used in Embodiment 2 of the present invention for coal bunkers and coal piles.

[0052] Figure 21 This is a three-dimensional point cloud image of a coal pile based on line laser virtual binocular stereo vision scanning in Embodiment 2 of the present invention;

[0053] Figure 22 This is a diagram showing the distance and error of 10 points in the three-dimensional point cloud of a coal pile in Embodiment 2 of the present invention. Detailed Implementation

[0054] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0055] Example 1

[0056] like Figure 1 As shown in Figure 16, this embodiment provides a computer vision-based 3D reconstruction method for cases where the background depth of field remains unchanged. Specifically, it includes: fixing a line laser generator above the camera's optical axis, with the laser plane coinciding with the camera's optical axis; mounting the camera and laser generator together on a sliding table; placing the object to be measured within the measurement range, projecting a line laser beam onto the object using the line laser generator, starting the sliding table to move at a constant speed, and acquiring a sequence of laser stripe images through the camera; using each image in the laser stripe image sequence as a right view, and constructing a virtual left view based on the right view; calculating the disparity between each right view and the corresponding virtual left view; inputting the disparity into a preset error correction formula to calculate the 3D coordinates of each pixel; and stitching together the 3D points of all frames to obtain the final imaging data.

[0057] In practice, the method of this embodiment can also be applied to a cloud environment, which includes a terminal, a server, and a data storage system; the terminal communicates with the server via a network. The data storage system can store the data that the server needs to process. The data storage system can be set up separately or integrated into the server. The terminal can send image data, camera data, and other data to be processed to the server. After receiving the computational data to be processed, the server performs error correction and 3D imaging processing steps, and the server can feed back the final imaging data obtained to the terminal. In addition, in some embodiments, the computer vision-based 3D reconstruction method can also be implemented by the server or the terminal alone.

[0058] Among them, the terminal can be, but is not limited to, various desktop computers, laptops, smartphones, tablets and IoT devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc.

[0059] A server can be implemented using a standalone server, a server cluster consisting of multiple servers, or a cloud server.

[0060] In one exemplary embodiment, the method is executed by a computer device, specifically by a computer device such as a terminal or a server alone, or by both a terminal and a server.

[0061] This embodiment proposes a line laser 3D imaging method based on virtual binocular stereo vision. This method requires only one industrial camera to achieve measurement results equivalent to traditional binocular systems. It has advantages such as low equipment cost and low image processing load. Experimental results show that this method can significantly improve point cloud density, achieving approximately a doubling of the density, thus enabling more accurate reconstruction of the 3D features of objects. Using the point cloud data obtained in this embodiment for 3D reconstruction can generate a 3D model with higher accuracy than the traditional binocular stereo vision line laser scanning method. This verifies the potential of this method in industrial applications.

[0062] This embodiment analyzes the imaging characteristics at the point where the line laser and the camera's optical axis coincide, proposing the use of a single industrial camera to achieve the function of a binocular camera without the need for other auxiliary equipment. Furthermore, it proposes the principle of line laser ranging under virtual binocular stereo vision. Based on this principle, this embodiment further explores the error model of line laser ranging under virtual binocular stereo vision and proposes corresponding correction strategies. Finally, experiments verify that the line laser 3D imaging method using virtual binocular stereo vision has good 3D reconstruction efficiency and accuracy.

[0063] like Figure 1 The diagram illustrates the workflow of the virtual binocular stereo vision line laser scanning 3D imaging method proposed in this embodiment. This method mainly includes six key steps: camera calibration, constructing a virtual left view based on the right view, solving the virtual binocular stereo vision ranging formula after structural error correction, fitting the virtual binocular stereo vision ranging and imaging error correction formula, solving the binocular stereo vision ranging formula after error correction, and 3D point cloud imaging. The specific steps are as follows:

[0064] 1. Virtual binocular stereo vision imaging principle:

[0065] (1) Theoretical principle of binocular stereo vision: As shown in Figure 2(a), it is an ideal imaging model of parallel axis binocular stereo vision. As can be seen from Figure 2, the midpoint P in space forms two images in the left and right cameras respectively. and Imaging. In the left eye camera, the horizontal coordinates of point P. Greater than the horizontal coordinate of the corresponding point in the right eye camera Therefore, it can be deduced that the disparity D at the midpoint P of the binocular camera is equal to... minus As point P moves in space, its imaging horizontal coordinates are formed in the left and right cameras. and It will also change accordingly.

[0066] As shown in Figure 2(b), when spatial point P is located on the optical axis of the left eye camera and can only move back and forth along that optical axis, the image in the right camera... horizontal coordinates The image changes as the position of point P changes. However, the image in the left camera... horizontal coordinates It is not affected by changes in the position of point P before and after, and can be considered a constant. In a binocular camera, the parallax of a spatial point P... The parallax will change as its position changes, but this parallax change is mainly caused by the right view and is unrelated to the left view.

[0067] like Figure 3 As shown in (a), when the laser generator is positioned in the center of the binocular camera, the laser reflection regions on the object's surface can be observed in the laser reflection image acquired by the binocular camera. These regions change with the object's depth, thus affecting the horizontal coordinates of the imaged light stripes. and This will also change accordingly. In practical applications, these curved light stripes need to be processed to obtain accurate three-dimensional information.

[0068] Based on the situation in Figure 2(b), when point P moves back and forth along the optical axis of the left eye camera, the laser generator is installed above the optical axis of the left eye camera (e.g., Figure 3 (b) As shown in this diagram, in the laser imaging of the reflected lines on the object surface acquired by the binocular camera, it can be observed that in the left view, the light stripe is imaged as a vertical line, and its width changes only with the depth of the object, without any bending. In other words, in the left view, the horizontal coordinate of the light stripe is... It remains unchanged. However, in the right view, the imaging light stripes in the laser-reflected area on the object's surface exhibit a curved shape. This can be compared... Figure 3 (a) and Figure 3 As can be seen from the imaging results of the right view in (b), when imaging at the same depth, the line laser imaging is significantly more curved when the laser generator is deployed above the left eye camera than when it is deployed in the middle position.

[0069] (2) Virtual binocular stereo vision line laser scanning three-dimensional imaging principle: based on Figure 3 Based on the observations in (b), the following conclusions can be drawn: When the laser generator is positioned above the optical axis of the left eye camera, during the line laser scanning process, the light stripe in the left view remains straight and its width changes only with the depth of the scanned object. This indicates that regardless of the depth of the scanned object, the horizontal coordinate of the line laser centerline remains constant. Keep it constant.

[0070] Therefore, if we can obtain the horizontal coordinates of each point on the center line of the light strip that remains straight in the left view... This allows the left-eye camera to be removed without needing to capture the left view. Only the horizontal coordinates of each point are required. The disparity is calculated by comparing the points corresponding to the right view acquired by the right eye camera, thus realizing the function of a binocular camera. For example... Figure 4 The image acquisition module is shown.

[0071] 2. Construction of Virtual Left View Based on Virtual Binocular Stereo Vision Line Laser Scanning 3D Reconstruction

[0072] (1) Camera Calibration: Camera calibration refers to establishing the relationship between the pixel positions of the camera image and the positions of scene points. Based on the camera imaging model, the camera model parameters are solved by the correspondence between feature points in the image coordinate system and the world coordinate system. The model parameters that need to be calibrated for the camera include internal parameters and external parameters. In this embodiment, the StereoCamera Calibrator toolbox in MATLAB software is used to calibrate the stereo camera using Zhang's calibration method to obtain the camera's internal parameters and correct the image distortion caused by lens distortion.

[0073] (2) Virtual Left View Construction: In the method proposed in this embodiment, a single camera is used to construct a virtual binocular camera system to replace the actual binocular camera used in binocular stereo vision line laser scanning 3D reconstruction. Unlike existing monocular vision line laser scanning 3D reconstruction schemes, this embodiment needs to utilize the "parallax" in the binocular camera to calculate depth information. Since a monocular camera can only capture an image from one viewpoint at a specific time, and cannot solve for parallax using the left and right views as a binocular camera, this method uses the image captured by the monocular camera as the right view, and constructs a "left view" based on this right view to perform parallax calculation.

[0074] like Figure 5 The image shown is a line laser image acquired by the right eye camera. This method aims to create a virtual left view using the line laser image acquired by the right eye camera. Based on... Figure 3 (b) It can be seen that when the line laser propagates along the optical axis of the left eye camera, it forms a vertical imaging line in the left eye camera, and its width only changes with the depth of the object, without bending. Therefore, this embodiment proposes a method for creating a virtual left view based on the translation of the real imaging of the line laser from the right eye camera. The principle of this method is as follows: First, calculate the horizontal coordinate D from the center point of the line laser at a height of H / 2 in the right view to the optical center. T Then, add D in the virtual left view. TThe corresponding number of right-side column pixels are then deleted. This process completes the translation creation of the virtual left view based on the real-time line laser imaging from the right-eye camera. At this point, the parallax of the center point of each line laser line between the virtual left and right views is D. T Specific operations are as follows: Figure 6 As shown.

[0075] A line laser generator projects a line laser beam along the plane's normal to the plane, and a camera captures the image of the reflected light stripes. Figure 6 As shown in (a). Create a new blank image with the same resolution, and add a D to the left side of the new image. T Pixels with a column pixel value of (0, 0, 0) (see Figure 6 (b) The yellow area on the left). Then, from Figure 6 (a) Remove the shadow on the right side of the image at point D. T The column of pixels is used to copy the remaining portion to the blank area of ​​the new image, such as... Figure 6 As shown in (b). Figure 6 (a) as the left view, and Figure 6 (b) As the right view, there is a parallax D between the left and right visual perspectives. T .

[0076] 3. Solving the virtual binocular stereo vision line laser ranging formula after error correction

[0077] (1) Virtual binocular stereo vision line laser ranging model and ranging error analysis: In a virtual binocular ranging system, such as Figure 7 As shown, there is a virtual camera. and an industrial camera Assuming The coordinate system coincides with the world coordinate system, and Compared to The relative coordinates are Due to virtual cameras Through the camera They are obtained so that their optical axes are parallel and their focal length f remains unchanged.

[0078] Assuming a spatial point P moves back and forth along a line laser path, in an ideal model, the line laser path and the virtual camera... The optical axes should be perfectly aligned. However, in reality, due to equipment errors, the line laser and the virtual camera may misalign. There is an angle between the optical axes. .use Let point P be represented by the virtual camera. Actual imaging The coordinates of the location, and with Calculate the corresponding three-dimensional coordinates for the reference frame. Similarly, in the camera An imaging point will also be generated on it. And it in A reference frame has coordinate values .

[0079] From the coordinate transformation, we can obtain:

[0080] (1)

[0081] From the above formula, we can obtain:

[0082] (2)

[0083] In the formula: For the camera The actual imaging point of spatial point P in the coordinate system To the camera Distance from the center of the projection; Let point P in space be at the camera. Upper actual imaging point Horizontal coordinates; Let point P in space be at the camera. Upper actual imaging point The horizontal coordinates.

[0084] In virtual camera Above, spatial point P is projected to form a virtual imaging point. Its coordinates are Meanwhile, in In a coordinate system, a virtual point The coordinates are .according to Figure 7 The above conclusions can be drawn from the situation shown.

[0085] (3)

[0086] In the formula: For line lasers and virtual cameras The angle between the optical axis and the optical axis.

[0087] Combining formulas (2) and (3), we can obtain:

[0088] (4)

[0089] Assuming in a virtual camera Up, point Image is formed along the optical axis. It is in The imaging coordinates on are , and The coordinates in the coordinate system are .

[0090] (5)

[0091] In the formula: In the camera Virtual imaging point of spatial point P in coordinate system To the camera The distance from the center of the projection.

[0092] Assuming due to the camera The existence of leads to Directional deviation This caused world coordinate error Assuming that the spatial point P remains fixed in the virtual camera's imaging deviation, this assumption can be interpreted as originating from the camera. Distance measurement error caused by imaging deviation.

[0093] (6)

[0094] (7)

[0095] In the formula, For the reason The resulting actual imaging point distance; For the reason The distance of the virtual imaging point caused.

[0096] Combining equations (6) and (7), we can obtain the result considering imaging errors. Distance error at that time.

[0097] (8)

[0098] From formula (8), we can obtain:

[0099] (9)

[0100] According to formula (9), the distance error can be obtained. Angle with line laser generator Baseline distance T and virtual imaging point distance There is a connection between them.

[0101] (2) Error correction of virtual binocular stereo vision line laser ranging structure:

[0102] From formula (6), we can obtain:

[0103] (10)

[0104] From formulas (8) and (9), we can obtain:

[0105] (11)

[0106] Combining formulas (10) and (11), we get:

[0107] (12)

[0108] From formula (4), we can obtain:

[0109] (13)

[0110] Combining formulas (12) and (13), we get:

[0111] (14)

[0112] Parallax Combining formula (14), we get:

[0113] (15)

[0114] According to formula (15), in order to obtain the actual distance of point P in space... It is necessary to know the camera focal length f, the baseline distance T of the virtual stereo camera, and the angle between the line laser generator and the optical axis of the virtual camera. as well as camera Directional deviation Parallax D in virtual binocular camera imaging. Due to the camera focal length This can be obtained through the camera calibration process, and camera Directional deviation The image p is related to the spatial point P within the camera, and the image p is related to the position of the spatial point P and the physical size of the camera's CCD module. Therefore, This is a variable involving the location of a spatial point P and the physical size of the camera's CCD module. Once a specific camera is selected, its CCD module's physical size will not change. In this case, It will change as the position of point P in space changes.

[0115] From formula (15), we can obtain:

[0116] (16)

[0117] In the formula, for camera Directional deviation Caused by the camera The actual imaging point of spatial point P in the coordinate system To the camera Error distance of the projection center.

[0118] (3) Solving for virtual binocular stereo vision line laser ranging parameters: Based on Figure 8 As shown, a virtual binocular stereo vision line laser measurement system is placed on a small cart that can move back and forth. A line laser is projected onto a vertical wall using a line laser emitter, and an industrial camera captures images of the reflected line laser light. Simultaneously, a laser rangefinder records the position and distance of the captured images. The relationship between them can be calculated by analyzing the difference in horizontal position between the imaging point at half the height of the real line laser and the imaging point at half the height of the virtual line laser in the acquired image.

[0119] The distance between spatial point P and the camera projection center is obtained using a virtual binocular stereo vision line laser measurement system. There is a difference between the distance Z to spatial point P obtained by the laser rangefinder and the distance Z obtained by the laser rangefinder. According to formula (16), assuming hour The equation involves the baseline distance T of the virtual binocular camera and the angle between the line laser generator and the optical axis of the virtual camera. and distance difference These are the three unknowns. Therefore, at least the distance information of spatial point P at three locations is needed to solve the problem.

[0120] (4) Virtual binocular stereo vision line laser ranging imaging error correction: based on Figure 9 As shown, a virtual binocular stereo vision line laser measurement system is placed on a movable cart, and a line laser emitter is used to illuminate a vertical wall. Images of the reflected line laser light from the wall are captured using an industrial camera, and the equation (16) mentioned earlier is solved. The results were obtained by measuring within a given range and recording the distances measured by the virtual binocular stereo vision line laser measurement system. and the distance measured by the laser rangefinder Next, and Substituting into formula (17), the ranging error was calculated. .

[0121] (17)

[0122] Error Substituting into equation (18), the relative error of stereo ranging can be obtained. :

[0123] (18)

[0124] According to formula (16), we can know It is due to imaging errors during actual shooting. This causes distance measurement errors. And camera imaging errors. It refers to The camera is Deviation in direction This deviation is influenced by a combination of factors, including camera manufacturing process, mounting errors, calibration errors, shooting environment, centerline extraction errors, and the physical dimensions of the camera CCD module. To reduce the impact of imaging errors... Caused ranging error To improve the ranging accuracy of the method in this embodiment, an equation fitting method is used. along with Error curve fitting analysis was performed to analyze the changing trend.

[0125] (5) Virtual binocular stereo vision line laser scanning 3D point cloud imaging: Based on the distance measurement correction formula derived from the virtual binocular stereo vision line laser measurement error model and the distance measurement error correction formula caused by imaging error in the actual shooting process obtained by the equation fitting method, this embodiment proposes a virtual binocular stereo vision line laser ranging error correction method, which can be applied to 3D point cloud imaging of line laser virtual binocular stereo vision line laser scanning.

[0126] 4. Virtual binocular stereo vision line laser scanning 3D point cloud imaging experiment:

[0127] (1) Experimental Equipment Composition Table: To verify the effectiveness of the method in this embodiment, two experiments were conducted: a virtual binocular stereo vision line laser distance measurement experiment and a virtual binocular stereo vision line laser scanning 3D point cloud imaging experiment. For this purpose, a virtual binocular stereo vision line laser distance measurement image acquisition device and a stereo vision line laser scanning 3D point cloud imaging system image acquisition experimental platform were constructed. Specific equipment parameters are detailed in Table 1. The test environment was a Windows 10 operating system running Python 3.7, using OpenCV 3.4.2 and Open3D 0.17.0 libraries to support algorithm development and testing.

[0128] (2) Experimental design of virtual binocular stereo vision line laser scanning 3D point cloud imaging

[0129] 1) Camera Calibration: In the virtual binocular stereo vision line laser distance measurement experiment, the image acquisition device is composed as follows: Figure 5As shown in (a). This experiment used Zhang Zhengyou's camera calibration method to calibrate the industrial camera, and the calibration parameters are listed in Table 2.

[0130] 2) Create a baseline left view: Illuminate the plane using a line laser generator and acquire an image with a resolution of 1536×1024 using an image acquisition device, such as... Figure 10 As shown in (a). Subsequently, following the method for creating a virtual left view based on the real imaging translation of the right eye camera's line laser proposed in this embodiment, 370 columns of pixels are deleted on the right side while 370 columns of pixels with a value of (0, 0, 0) are added on the left side to obtain a new image, as shown. Figure 10 As shown in (b). [The text abruptly ends here.] Figure 10 (b) as the reference left view.

[0131] 3) Solving for virtual binocular stereo vision line laser ranging parameters: such as Figure 11 As shown, an image acquisition experimental platform for a stereo vision line laser scanning imaging system was constructed. This platform mainly consists of two parts: a CNC sliding guide linear module and an image acquisition module. The CNC sliding guide linear module comprises a programmable stepper motor pulse controller, a stepper motor driver, and a ball screw slide. Its main function is to drive the image acquisition system to move along the ball screw slide to achieve line laser scanning and image acquisition. It is worth noting that the image acquisition system is approximately 650 mm from the desktop, and the working distance is set between 350 and 650 mm.

[0132] according to Figure 8 The measurement procedure shown involves acquiring images of the reflected laser light from the wall at three locations at different distances, and simultaneously recording the measured wall distances at these locations using a laser rangefinder: 378.2 mm, 504.1 mm, and 634.2 mm. Using the method described in this embodiment, the parallax D of the images at these three locations was calculated to be 1080.10 px, 1000.01 px, and 949.03 px, respectively. (The last sentence appears to be incomplete and possibly refers to a separate measurement process.) =1714.9916, the limit distance T of the virtual stereo camera is calculated to be 81.7249 mm. Furthermore, the angle between the line laser generator and the optical axis of the virtual camera... =0.43078, while the distance difference =32.4402 mm.

[0133] (19)

[0134] 4) Imaging error correction: such as Figure 9As shown, the image acquisition device is moved back and forth along the normal direction of the vertical wall within a range of 350~650 mm from the vertical wall, and the distance of the virtual binocular stereo vision line laser measurement system measured by formula (19) is recorded. and the distance measured by a laser rangefinder Meanwhile, the ranging error is calculated using formulas (17) and (18). and relative ranging error .by The x-axis is... and Plot the distribution of error and relative error as a function of measurement distance on the ordinate axis, see details. Figure 12 .

[0135] according to Figure 12 The data shown indicates that, within the working range of 350~650 mm, the ranging error... The fluctuation ranges from -0.6 to 0.7 mm. The relative ranging error can be observed through the green line. The variation ranges from -0.09745% to 0.18995%. These results indicate that the ranging error... Mainly due to imaging errors during the actual imaging process of the camera. This is caused by camera imaging errors. Further analysis and processing are needed to eliminate the impact of camera imaging errors on ranging accuracy. This applies to the variation of measurement distance within the working range. Distance measurement error due to changes In this case, a camera imaging error formula (20) was established using curve fitting. The imaging error correction formula (20) obtained through curve fitting can more accurately correct the ranging error caused by camera imaging. This will help improve system accuracy and stability, and ensure reliable and accurate measurement results within different operating ranges.

[0136] (20)

[0137] 5) Virtual binocular stereo vision line laser scanning 3D point cloud imaging: In the experiment, an experimental platform was used for image acquisition, such as... Figure 13 As shown in (a), an industrial camera is used to capture and save images of the object illuminated by a linear laser within the operating range L of the CNC sliding guide linear module, as illustrated in the following diagram. Figure 13 As shown in (b).

[0138] To conduct further analysis and calculations, we will first... Figure 10 (b) The left view is used as the left view, and the 344 images acquired on the test bench are used as the right view. Through centerline extraction and stereo matching techniques, the line parallax between each group of images can be accurately calculated. Subsequently, the parallax of these rows is calculated using the parallax-depth formula obtained by solving for predetermined parameters. This formula allows us to obtain the depth coordinates of each pixel along the laser centerline in each image set. Considering factors such as camera movement speed and coordinate conversion, we can also calculate the horizontal and vertical coordinates of each pixel along the laser centerline in space. After these steps, we obtain 344 sets of 3D coordinate data for the laser centerline. By stitching together all 344 sets of depth information, we can generate a complete and clear 3D point cloud model that clearly shows the shape and structural features of the object, such as... Figure 12 As shown in (c).

[0139] 5. Analysis of Experimental Results:

[0140] (1) Experiment on correction of imaging error in virtual binocular stereo vision ranging: According to formulas (19) and (20), a ranging experiment was conducted within a distance range of 350~650 mm. Through this experiment, the distance of the virtual binocular stereo vision line laser measurement system can be obtained. and the distance obtained from the laser rangefinder Meanwhile, using formulas (18) and (19), the ranging error is calculated. and relative ranging error To more intuitively demonstrate the trends of error and relative error at different measurement distances, this embodiment uses... As the horizontal axis, and Plot the distribution on the ordinate. This clearly shows the error variations at each measurement point.

[0141] according to Figure 14 As shown, after correcting the imaging error using the correction formula (20), the ranging error... The error fluctuates between -0.1 and 0.3 mm. This indicates that the ranging results have been effectively adjusted and calibrated using the correction algorithm. The blue line shows the relative ranging error. The error fluctuated between -0.0283% and 0.046118%. Compared with the original error, the maximum ranging error decreased from 0.7 mm to only 0.3 mm, a reduction of 57.14%. At the same time, the maximum relative ranging error also decreased from as high as 0.18995% to 0.046118%, a reduction of 75.72%. Experimental data show that after correcting the imaging error using the correction formula (20), the measurement accuracy of the system was significantly improved. This correction enables the system to more accurately obtain the actual distance between the target object and the camera, while maintaining good stability.

[0142] (2) Binocular Stereo Vision 3D Point Cloud Imaging Experiment: To verify the speed and accuracy of virtual binocular stereo vision line laser scanning 3D point cloud imaging using the method of this embodiment, this embodiment specifically arranged a binocular stereo vision line laser scanning 3D point cloud imaging experiment for comparison. This experiment aims to evaluate the advantages and disadvantages of the method proposed in this embodiment in terms of speed and accuracy by comparing the imaging results under different technical means. In the experiment, the method of this embodiment was first used to perform 3D point cloud imaging, and the corresponding data was recorded. Then, under the same conditions, the binocular stereo vision line laser scanning method was used to perform 3D point cloud imaging, and the corresponding data was obtained. By comparing the results obtained by the two methods, their differences in speed and accuracy can be objectively evaluated.

[0143] 1) Camera calibration:

[0144] The image acquisition module is constructed using the following method. Figure 15 The combination shown in (a) was used to calibrate the binocular camera using Zhang Zhengyou's camera calibration method. The relevant calibration parameters are detailed in Table (3).

[0145] By comparing the camera calibration parameters in Tables (2) and (3), it can be seen that the method in this embodiment uses a single camera for calibration. Therefore, only the intrinsic parameters of the camera need to be solved, without needing to solve the rotation and translation matrices. As a result, since there are fewer calibration parameters, the impact of errors caused by camera calibration on the imaging results is relatively limited.

[0146] 2) Image Acquisition and Point Cloud Imaging: According to experimental requirements, high-precision image acquisition was performed using a binocular stereo vision line laser scanning method experimental platform. For example... Figure 15 As shown in (a), a binocular stereo vision line laser scanning operation was performed on a standard sphere with a diameter of 198.8 mm. By applying a CNC sliding guide linear module, high-quality image data acquired by the binocular camera was successfully recorded and stored within the operating range L, and is displayed in detail on [the image]. Figure 15 (c)

[0147] Within a 400 mm range of the CNC sliding guide linear module's operation, this experiment successfully acquired 172 pairs of images with a resolution of 1536×1024, totaling 344 images. For each pair of images, centerline extraction and stereo matching were performed using the left eye camera as the reference. The line parallax D of each image pair was calculated. iSubstituting these values ​​into the "parallax-depth" formula of the binocular camera, the coordinate information of each pixel in the depth direction can be obtained. Simultaneously, by combining coordinate transformation and camera movement speed calculation, the horizontal and vertical coordinates of each pixel along the line laser centerline in space are obtained. These steps successfully acquire a total of 172 sets of 3D coordinate information for the line laser centerline, which is of significant importance for further research and application. Finally, after stitching together these 172 sets of depth information, a 3D point cloud model (such as...) is generated. Figure 15 (d) is shown.

[0148] (3) Comparison of point cloud between binocular stereo vision line laser scanning imaging method and the method in this embodiment

[0149] 1) Comparison of Point Cloud Density and Imaging Time: As shown in Figure 16, Figure 16(a) shows the point cloud image of the model sphere obtained by binocular stereo vision line laser scanning imaging. In contrast, Figure 16(b) shows the point cloud image generated by the method proposed in this embodiment. As shown in Figure 16, this embodiment and the binocular stereo vision line laser scanning imaging method obtained the same number of image samples (344 in total) within the same time. However, due to limiting factors such as image acquisition frequency and data transmission, the image data obtained by the binocular stereo vision line laser scanning imaging method includes 174 left views and 174 right views. Depth information is obtained by calculating the disparity information between the corresponding left and right views, and finally a point cloud model containing 174 3D line segments is constructed. In contrast, in the method proposed in this embodiment, when using a single camera for image acquisition, all 344 images are treated as right-view images. By comparing with the constructed left view and calculating the disparity value, more accurate depth information is obtained. Therefore, a total of 344 three-dimensional lines are created in the final construction of the point cloud model. This is twice the point cloud density of binocular stereo vision line laser scanning imaging. This increased point cloud density allows for better reconstruction of the three-dimensional features of objects.

[0150] In this experiment, the binocular stereo vision line laser scanning imaging method requires extracting center lines, stereo matching, solving disparity, and calculating pixel depth information from 344 images, taking approximately 19.2794 seconds in total. The method proposed in this embodiment requires the same processing steps for 345 images, taking approximately 19.3407 seconds in total. Experimental data shows that the time consumption of the method in this embodiment is comparable to that of the binocular stereo vision line laser scanning imaging method during image processing. However, if a point cloud image of similar density is to be acquired, the method in this embodiment can complete the processing in only about half the time.

[0151] 2) Comparison of 3D Point Cloud Reconstruction Accuracy: To preprocess the point cloud data obtained by the binocular stereo vision line laser scanning imaging method and the method of this embodiment, the Alpha Shapes algorithm was used. This algorithm can effectively extract feature points and filter the point cloud to reduce noise interference. Then, the Delaunay triangulation algorithm was used to reconstruct the surface of the filtered point cloud. After 3D reconstruction using the binocular stereo vision line laser scanning imaging method and the method of this embodiment, the measurement radius of the calibration sphere can be calculated respectively, as shown in Table 4.

[0152] According to the data in Table 4, the standard sphere radius error measured using the binocular stereo vision line laser scanning imaging method is 0.27 mm, while the standard sphere radius error measured using the method of this embodiment is 0.18 mm. This result indicates that the method proposed in this embodiment is superior to the traditional binocular stereo vision line laser scanning imaging method in terms of reconstruction accuracy.

[0153] In summary, this embodiment addresses the issues of high computational load and low efficiency in binocular stereo vision line laser scanning algorithms. It proposes a ranging method based on virtual binocular stereo vision line lasers, utilizing the imaging characteristics of line lasers at the camera's optical axis. Compared to traditional binocular camera systems, this embodiment exhibits significant advantages, particularly in reducing equipment costs. This method requires only a single industrial camera, eliminating the need for additional auxiliary equipment and thus reducing costs. Secondly, it significantly reduces image processing workload. Traditional binocular camera systems require matching and correction of images acquired from two cameras, demanding sophisticated algorithms; while the method proposed in this embodiment achieves the same effect by processing only the image acquired from a single camera.

[0154] The core principle of this method lies in using a single industrial camera to replace the right eye in a binocular camera system. By capturing images of vertical light stripes, a virtual reference left view is constructed using image processing methods. Subsequently, the disparity between the real-time right view and the reference left view is calculated, and the depth information of the target is obtained based on the "disparity-depth" formula derived in this embodiment.

[0155] Based on the method proposed in this embodiment, 3D point cloud imaging and 3D point cloud reconstruction experiments were conducted. The 3D point cloud imaging experiments show that, despite similar image processing times, the method proposed in this embodiment achieves almost double the point cloud density compared to traditional binocular stereo vision 3D point cloud imaging, thus reconstructing the 3D characteristics of objects more accurately. Furthermore, if similar density 3D point cloud data is required, the method in this embodiment requires only half the time of the ordinary binocular stereo vision 3D point cloud imaging method. The 3D reconstruction experiment using a standard sphere shows that 3D reconstruction using the point cloud data acquired using the method in this embodiment yields a 3D model with higher accuracy than that obtained using binocular stereo vision line laser scanning imaging. Based on these results, it is foreseeable that the 3D imaging method proposed in this embodiment will have broad industrial application potential and can meet the needs of modern industrial measurement.

[0156] Example 2

[0157] like Figures 17-22 As shown, this embodiment provides a 3D reconstruction method for background depth variations. The left view is created by drawing a vertical line through the center of the right-eye camera image and using it as the basis for the virtual left view. This method solves for parallax by comparing the real-image light stripe from the right-eye camera with the vertical line through the image center. It can achieve the effect of acquiring both left and right views using a single camera, thus creating a virtual binocular camera effect. Figure 17 As shown.

[0158] To obtain the depth information of each pixel on the center line of the line laser, it is necessary to calculate the disparity by matching the pixels on the actual line laser center line imaged by the camera with the pixels on the center line of the virtual left view, and then use the "disparity-depth" formula derived in this embodiment to calculate the depth of each pixel. However, the "disparity-depth" formula uses an image-based... The derivation is performed on two pixels. Due to installation errors during the assembly of the laser generator, the pixels at the center of the actual line laser image do not perfectly conform to the derived "parallax-depth" formula.

[0159] like Figure 18 As shown, the angle generated when considering the actual imaging caused by laser generator installation errors. After addressing the issue, this embodiment allows us to observe that the parallax of the top pixel in the image is... The parallax of the bottom pixel is And the image The parallax at that location is ,and However, since the actual line laser images are obtained by illuminating the same plane with a line laser, the "parallax-depth" formula only applies to pattern height ( The result at the given location cannot be generalized to pixels at other locations on the center line of the linear laser stripe. Based on this fact, this embodiment must adjust the horizontal coordinates of other pixels on the actual imaging line laser center line accordingly to ensure that each pixel located on the line laser center line conforms to the "parallax-depth" formula proposed in this embodiment.

[0160] Depend on Figure 18 It can be seen that the angle between the center lines of the linear laser is:

[0161] (twenty one)

[0162] Assuming the coordinates of the pixel on the center line of the actual imaging line laser are... ,but:

[0163] (twenty two)

[0164] Therefore, after substituting the pixels of the actual line laser center line in the camera into Formula 2 for horizontal coordinate correction, all pixels on the actual line laser center line in the camera can satisfy the "parallax-depth" formula proposed in this embodiment.

[0165] Figure 19 This embodiment demonstrates a virtual binocular stereo vision line laser scanning system, primarily composed of a double-girder bridge crane and a vision acquisition device. The vision acquisition device consists of an industrial camera and a laser generator, both fixed to the crossbeam of the double-girder bridge crane via a bracket. The crane utilizes advanced control technologies such as a programmable logic controller (PLC), frequency converter, and position encoder to achieve intelligent operation. During the operation of the virtual binocular stereo vision line laser scanning system, the target coal pile is first irradiated by the laser generator, and then the industrial camera captures the reflected laser image. During this process, the bridge crane drives the vision acquisition device to move along the material area to complete the scanning of the entire area. To accurately record the position of the vision acquisition device during the scanning process, the double-girder bridge crane is equipped with a laser rangefinder to accurately record the position coordinates when the camera captures the image.

[0166] This 3D reconstruction experiment of a coal pile was conducted in a coal storage silo at a heat source plant. The basic parameters of the silo are shown in Table 5. According to the data in Table 5, the dimensions of the coal storage silo in this experiment were 100 meters long, 30 meters wide, and 25 meters high. The grab crane was 20 meters above the ground, and the maximum coal stacking height was 8 meters. The bridge crane was installed at a height of 20 meters, and the maximum reconstruction error remained within 15 centimeters when generating the 3D point cloud data.

[0167] This embodiment uses a HIKVISION MV-CE060-10UC camera with an MVLHF1628M-6MP [12mm] lens, combined with a LYUE 520MA1000YLV line laser generator with a wavelength of 520nm and a power of 1W, to construct a virtual binocular stereo vision line laser scanning image acquisition system. Specific parameters are shown in Table 6. Simultaneously, the center line of the line laser stripes is extracted using an adaptive Gaussian sliding window method, and the "parallax-depth" formula is derived through an imaging error correction function. The horizontal coordinates of the pixels on the center line of the line laser are then corrected, thus forming a complete virtual binocular stereo vision line laser ranging system. Furthermore, the acquisition of line laser scanning images is achieved based on the forward and backward movement and image acquisition system of a double-girder bridge crane in a coal storage bunker.

[0168] like Figure 20 As shown in (a), this is a virtual binocular stereo vision line laser scanning image acquisition system. Figure 20 (b) The yellow area represents the scanning range of the 3D point cloud reconstruction experiment of the coal pile. This experiment used a single camera for image acquisition and did not use multiple cameras for image stitching across the coal bunker. Due to the 31.0° field of view of the MVL-HF1628M-6MP [12mm] lens, the camera's field of view could not cover the lateral width of the coal bunker. The double-girder bridge crane trolley allows the crane body to move back and forth along the Y-axis of the coal storage bunker, thereby driving the line laser virtual binocular stereo vision scanning image acquisition system to photograph and scan the coal pile. Figure 20 (c) shows a laptop computer used to acquire images and record their position values ​​on the Y-axis. The industrial camera captures a reflective laser image of a coal pile, as shown... Figure 20 As shown in (d).

[0169] A double-girder bridge crane was used to drive a line laser virtual binocular stereo vision scanning image acquisition system to acquire line laser images of the yellow area along the Y-axis of the coal storage bunker. The center lines of the line laser stripes were extracted using an adaptive Gaussian sliding window method, and their horizontal coordinates were corrected. Subsequently, the depth information of all line lasers in the scanned area was calculated according to the "parallax-depth" formula derived in this embodiment, generating a three-dimensional point cloud map of the coal pile distribution, as shown below. Figure 21 As shown.

[0170] Based on the data shown in Table 7, this study conducted a three-dimensional reconstruction experiment of a coal pile, taking into account the limitations of the bridge crane width and the safe distance to the coal storage bunker. In the experiment, the average moving speed of the crane was set to 1.35 m / s, achieving a scanning process of 90 meters in length. By setting the image acquisition frequency to 0.1 frames per second, a total of 1215 images were acquired, and the entire scanning process took 121.5 seconds. Applying the image processing algorithm proposed in this embodiment to perform depth calculations on the acquired images took a total of 113.2216 seconds. The experimental results show that, under the condition of an average crane speed of 1.35 m / s, the line laser virtual binocular stereo vision scanning technology developed in this embodiment has a significantly lower image processing time than the entire scanning time, fully demonstrating the significant advantage of this technology in processing speed. Therefore, the three-dimensional imaging technology proposed in this embodiment not only demonstrates excellent timeliness but also effectively achieves efficient three-dimensional reconstruction of coal piles through real-time adjustment of crane speed and image acquisition frequency. The successful application of this technology provides a high-efficiency and high-accuracy three-dimensional imaging solution for related fields.

[0171] To verify the accuracy of the proposed method in generating a 3D coal pile model, this embodiment uses a laser rangefinder mounted on a double-bridge crane trolley for experimentation. This equipment allows for the acquisition of height information between the coal pile and the laser rangefinder, further evaluating the method's accuracy. During the experiment, 10 evenly distributed locations within the scanning area were selected as sample points. These locations covered the entire coal pile surface and were strategically planned to ensure broad and non-repetitive coverage. For each selected location, the laser rangefinder was used to measure the height and record the corresponding data. Subsequently, the actual measured values ​​were compared with the point cloud height information obtained by the proposed method. The difference between the two is used to evaluate the accuracy and reliability of the proposed method in generating the 3D coal pile model. This comparative evaluation effectively verifies whether the proposed method possesses sufficient accuracy in handling coal pile modeling problems in real-world scenarios.

[0172] Height data was measured at 10 locations within a specified scanning area. For error calculation, this embodiment uses the distance measured by the laser rangefinder as the baseline distance and the distance between the measurement points measured by the 3D point cloud as the actual measurement distance. By comparing these two sets of data, measurement errors present during the measurement process can be assessed. Table 8 provides detailed measurement data.

[0173] according to Figure 22The results show that the measurement range in this experiment was 9977.1~13037.1 mm, with the distance measurement error fluctuating between -12.5 and 13.1 mm. Detailed analysis and statistics show that the average measurement error was approximately 1.99 mm, and the average relative measurement error was approximately 0.022%. Based on the above data analysis results, it can be concluded that the 3D point cloud data of the coal pile generated by the virtual binocular stereo vision line laser scanning 3D imaging method proposed in this embodiment has a highly accurate reconstruction capability. This means that the method can accurately capture and reconstruct information such as the shape, size, and surface features of the coal pile. Furthermore, it demonstrates good adaptability in meeting the requirements of intelligent digital coal storage management systems.

[0174] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A 3D reconstruction method based on computer vision, characterized in that, include: S1: Fix the line laser generator above the camera's optical axis, with the laser plane coinciding with the camera's optical axis; The camera and laser are mounted together on the slide. S2: Place the object to be measured within the measurement range, use a line laser generator to project a line laser onto the object to be measured, start the slide table to move at a constant speed, and acquire the laser stripe image sequence through a camera; S3: Take each image in the laser stripe image sequence as the right view, and construct a virtual left view based on the right view; calculate the disparity between each right view and the corresponding virtual left view; S4: Input the parallax into the preset error correction formula and calculate the three-dimensional coordinates of each pixel; S5: Stitch together the 3D points of all frames to obtain the final imaging data.

2. The computer vision-based 3D reconstruction method according to claim 1, characterized in that, In step S1, the laser plane corresponding to the line laser generator coincides with the optical axis of the camera, which is an industrial camera with pre-calibrated intrinsic parameters.

3. The computer vision-based 3D reconstruction method according to claim 1, characterized in that, The construction of a virtual left view based on the right view specifically includes: For cases where the background depth of field remains constant: extract the center line of the laser stripes in the right view and calculate the horizontal coordinate D at the center line of the image. T Create a new all-black image of the same size as the right view; insert a D-shaped element on the left side of the all-black image. T Column of black pixels, delete D on the right side of the right view. T The portion following the column is copied to the blank area of ​​the all-black image to complete the construction of the virtual left view; For situations involving changes in background depth: Create a vertical line passing through the center of the image in the right view, and use the image with the vertical line as the virtual left view.

4. The computer vision-based 3D reconstruction method according to claim 3, characterized in that, The error correction formula in step S4 is as follows: In the formula, It is a constant. For laser distance measurement, This represents the ranging error.

5. The computer vision-based 3D reconstruction method according to claim 4, characterized in that, The derivation process of the error correction formula is as follows: Using a line laser generator, project a line laser along the plane normal onto a flat white wall. Move the slide to three preset positions and collect images of the line laser reflected from the wall at three positions at different distances from the wall. At the same time, use a laser rangefinder to record the wall distance measured when the images are collected, and calculate the parallax between each laser stripe image and the corresponding virtual left view. The measurement distance is calculated by combining parallax and camera calibration parameters; The difference between the distance to the wall recorded by the laser rangefinder and the measured distance is calculated to obtain the distance measurement error; The relative error of ranging is calculated based on the ratio of ranging error to the measured distance; Using the measured distance as the x-axis and the distance measurement error and relative distance measurement error as the y-axis, a distribution map of the error and relative error as a function of the measured distance is constructed. Based on the constructed distribution map, an error correction formula is constructed using curve fitting.

6. A computer vision-based 3D reconstruction system, characterized in that, include: The image acquisition module is used to fix the line laser generator above the camera optical axis, with the laser plane coinciding with the camera optical axis. The camera and laser are mounted together on the slide table; the object to be measured is placed within the measurement range, a line laser generator is used to project a line laser onto the object to be measured, the slide table is started to move at a constant speed, and the laser stripe image sequence is acquired by the camera. The virtual left view construction module is used to take each image in the laser stripe image sequence as a right view, construct a virtual left view based on the right view, and calculate the disparity between each right view and the corresponding virtual left view. The 3D reconstruction module is used to calculate the 3D coordinates of each pixel by inputting the parallax into a preset error correction formula; and stitches together the 3D points of all frames to obtain the final imaging data.

Images