Modeling device and method for inner wall of wine jar

Abstract

The application discloses a wine jar inner wall modeling device and method, and relates to the technical field of wine jar inner wall analysis. The device comprises a driving mechanism, a shell and a scanning modeling mechanism. In use, the driving mechanism is first fixed on the mouth of the wine jar, and then the driving mechanism is started. The lower end of the shell enters the interior of the wine jar and is positioned near the central axis. Then, the driving mechanism drives the shell to ascend and descend along the central axis to realize the scanning operation function. The device avoids the operation of the operator entering the interior of the wine jar to paste auxiliary mark points or manually adjusting the position of the equipment, thereby completely eliminating the influence of glue residues on the cleanliness of the wine jar and meeting the strict requirements of the brewing site on the health level.

Description

Technical Field

[0001] This invention relates to the field of wine jar inner wall analysis technology, and in particular to a wine jar inner wall modeling device and method. Background Technology

[0002] Wine jars are the main equipment used in winemaking and for storing wine. Wine jar inner wall modeling technology is primarily used for accurately calculating the volume of the jar's inner wall, optimizing brewing processes, and monitoring the aging process. Current technologies are mainly divided into two categories: contact modeling and non-contact modeling, but both have certain limitations.

[0003] Contact measurement technologies mainly include mechanical probe scanning and flexible sensor array methods. These methods primarily use robotic arms to manually scan and record three-dimensional coordinate point cloud data along the inner wall of the wine jar using a probe, or attach flexible pressure / deformation sensors to the inner wall of the jar and infer the shape of the inner wall from the data. These methods are slow, unsuitable for large-scale applications, complex to install, costly, and difficult to popularize; furthermore, the probes may contaminate the wine, making them unsuitable for on-site operating environments.

[0004] Non-contact modeling technologies mainly include laser scanning, ultrasonic 3D imaging, and optical photogrammetry. Laser scanning (LiDAR / structured light) uses a laser scanner or structured light projector to acquire 3D point cloud data of the inner wall. It can achieve high accuracy (±0.1mm) and model complex inner wall structures, but the equipment is expensive (tens of thousands to millions of yuan) and the operation is complex. It mainly requires the pasting of marker points to complete the scanning of the entire inner wall of the jar, which takes a long time to collect data for the inner wall of a single jar. In addition, the glue left by these auxiliary marker points has a significant impact on the cleanliness of the jar, affecting subsequent brewing. Ultrasonic 3D imaging reconstructs the 3D model of the inner wall by reflecting signals from multiple arrays of ultrasonic sensors. It can penetrate the wine liquid and adapt to non-transparent environments, but its resolution is low (±1cm), making it difficult to accurately model tiny bumps and depressions. Optical photogrammetry (multi-view 3D reconstruction) takes photos of the inner wall of the jar from multiple angles and uses the SFM (Structure of Motion) algorithm to model it. It is low-cost (only requires a camera) and suitable for static modeling, but it depends on lighting conditions. Reflections or shadows on the inner wall of the jar can lead to data loss, making it difficult to form accurate inner wall point cloud data. Without affixing markers, there is currently no fast scanning method on the market that can achieve an accuracy of 0.1mm for the inner wall of wine jars. Therefore, solving this problem has significant practical implications.

[0005] Without attaching markers, existing technology cannot scan the inner wall with high precision in one go. Moreover, during scanning, it is necessary for people to climb into the wine jar to attach markers. When removing the markers after scanning, it is easy to leave adhesive residue, which makes non-contact scanning impossible. In addition, the reconstruction accuracy will also be affected if there are water stains or no shape features on the inner wall. Summary of the Invention

[0006] The main objective of this invention is to propose a modeling device and method for the inner wall of a wine jar, aiming to solve the technical problems that existing technologies cannot scan the inner wall with high precision in one go without attaching markers. Furthermore, the scanning process requires a person to climb into the wine jar to attach markers, and removing the markers after scanning can easily leave adhesive residue, making non-contact scanning impossible. Additionally, the reconstruction accuracy is also affected by the presence of water stains or the lack of morphological features on the inner wall.

[0007] To achieve the above objectives, in a first aspect, the present invention provides a modeling device for the inner wall of a wine jar, comprising: A driving mechanism is installed at the mouth of the wine jar and extends from the mouth of the jar into the interior of the wine jar. A casing, wherein the casing is mounted at one end of the drive mechanism located inside the wine jar, the casing having a mounting position formed thereon, and the drive mechanism capable of driving the casing to rotate within the wine jar; and... A scanning modeling mechanism is installed at the installation position, facing the inner wall of the wine jar. The driving mechanism can drive the scanning modeling mechanism to rotate inside the wine jar and scan the inner wall of the wine jar to model the wine jar.

[0008] In one embodiment, the scanning modeling mechanism includes: A scanning component, the scanning component being mounted at the mounting position, the scanning component being disposed facing the inner wall of the wine jar; and, A modeling terminal is communicatively connected to the scanning component. The modeling terminal is used to receive point cloud data inside the wine jar obtained by the scanning component and to model it.

[0009] In one embodiment, a plurality of mounting grooves are formed at circumferential intervals along the outer casing at the mounting location; The scanning component includes multiple first scanning modules and multiple second scanning modules. All first scanning modules and all second scanning modules are installed circumferentially and spaced apart from each other in the mounting slot. All first scanning modules and all second scanning modules are arranged facing the inner wall of the wine jar, and all first scanning modules and all second scanning modules are communicatively connected to the modeling terminal.

[0010] In one embodiment, the first scanning module includes: A first mounting base, which is installed in a corresponding mounting slot; and... A camera is mounted on the first mounting base, and the lens of the camera is positioned facing the inner wall of the wine jar.

[0011] In one embodiment, the second scanning module includes: A second mounting base is installed in a corresponding mounting slot, and the second mounting base is spaced apart from the first mounting base; and, A line laser is mounted on the second mounting base, with the laser output end of the line laser facing the inner wall of the wine jar.

[0012] In one embodiment, the first scanning module and the second scanning module are distributed at intervals along the height direction of the wine jar.

[0013] In one embodiment, the drive mechanism includes: A support frame is installed at the mouth of the wine jar, and a rotating through hole is formed at the center of the support frame in a vertical direction; A rotary drive member, mounted on the support leg, the output end of the rotary drive member extending downward and rotatably passing through the rotary through-hole; and, A rotating rod is installed at the output end of the rotating drive component. The rotating rod is located inside the wine jar and connected to the outer shell. The rotating drive component can drive the rotating rod to rotate the outer shell so that the scanning modeling mechanism can scan the inner wall of the wine jar to model the wine jar.

[0014] Based on the same technical concept, in a second aspect, the present invention also proposes a method for modeling the inner wall of a wine jar, using the wine jar inner wall modeling device described in the first aspect. The scanning modeling mechanism includes at least one first scanning module and at least one second scanning module. The first scanning module includes a camera, and the second scanning module includes a line laser. The method includes the following steps: The device is installed at the mouth of the wine jar, with the first scanning module facing the inner side of the wine jar and the second scanning module facing the bottom of the inner wall of the wine jar. When the drive mechanism drives the outer shell to rotate, it controls the line laser to project laser stripes onto the inner wall of the wine jar, and at the same time controls the camera to acquire the laser stripe image; Based on the acquired laser stripe images, the side point clouds and bottom point clouds were reconstructed respectively. Determine the relative pose between the first scanning module and the second scanning module; The side point cloud and the bottom point cloud are fused together according to the relative pose to obtain a complete point cloud of the inner wall of the wine jar. A model of the inner wall of the wine jar is constructed based on the complete point cloud of the inner wall of the wine jar.

[0015] In one embodiment, the step of reconstructing the side point cloud and bottom point cloud from the acquired laser stripe image includes: For each laser stripe image acquired by the camera, the pixel coordinates of the center point of the laser stripe are extracted; Based on the camera intrinsic parameters and the laser plane equation, the three-dimensional point corresponding to each pixel is calculated by triangulation to obtain a single frame point cloud. Based on the rotation angle recorded during the rotation process, the point clouds of each single frame are stitched together to obtain the side point cloud and the bottom point cloud respectively.

[0016] In one embodiment, prior to the step of installing the device at the mouth of the wine jar and positioning the first scanning module facing the inner wall of the wine jar and the second scanning module facing the bottom of the inner wall of the wine jar, the method further includes: The camera's intrinsic parameters are calibrated using a calibration board to obtain the camera's intrinsic parameter matrix and distortion coefficients; The calibration plate is placed in different positions, laser stripes are projected, the center point of the stripes is extracted, and the laser plane equation is fitted by the least squares method.

[0017] The technical solution of this invention, through the setting of a drive mechanism, a shell, and a scanning and modeling mechanism, allows for the modeling of the inner wall of the wine jar. During use, the drive mechanism is first securely fixed to the mouth of the wine jar, then activated. The shell, along with the lower end of the extension arm, enters the interior of the wine jar and is positioned near the central axis. Subsequently, the drive mechanism drives the shell to rotate continuously or in segments around the central axis inside the wine jar. During rotation, the scanning and modeling mechanism works synchronously, its scanning direction always remaining radially outward. As the shell rotates, the scanning path forms a spiral or multiple parallel circular trajectory, gradually covering the entire area of ​​the inner wall of the wine jar. By continuously collecting the three-dimensional coordinate data of each point on the inner wall, a complete point cloud dataset is formed. Subsequent processing, including denoising, registration, and meshing, allows for the reconstruction of a high-precision three-dimensional model of the inner wall of the wine jar. Furthermore, this invention enables non-contact, marker-free, high-precision inner wall scanning and modeling. The drive mechanism extends from the mouth of the jar to the interior and drives the shell to rotate, avoiding the need for operators to enter the wine jar to attach auxiliary markers or manually adjust the equipment position, thus completely eliminating the impact of glue residue on the cleanliness of the wine jar and meeting the stringent hygiene requirements of brewing sites. Meanwhile, since the scanning and modeling mechanism rotates synchronously with the outer shell, it can complete the full coverage scan of the inner wall in one go without the need for external multi-station setup or multiple equipment relocations, significantly shortening the acquisition time. Typically, the scanning process of a single wine jar can be completed within a few minutes, which significantly improves efficiency compared to the traditional method of climbing into the jar or taking multiple stations. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0019] Figure 1 A schematic diagram of the structure of the wine jar inner wall modeling device provided by the present invention; Figure 2 for Figure 1 Another structural schematic diagram of the modeling device for the inner wall of a wine jar, as shown in the example; Figure 3 This is a flowchart illustrating the method for modeling the inner wall of a wine jar, as exemplified by the present invention.

[0020] Explanation of icon numbers: 100. Drive mechanism; 200. Wine jar; 300. Outer shell; 400. Scanning and modeling mechanism; 410. Scanning component; 420. Modeling terminal; 310. Mounting slot; 411. First scanning module; 412. Second scanning module; 413. First mounting base; 414. Camera; 415. Second mounting base; 416. Line laser; 110. Support leg; 120. Rotary drive component; 130. Rotating rod.

[0021] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0024] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0025] Without attaching markers, existing technology cannot scan the inner wall with high precision in one go. Moreover, during scanning, it is necessary for people to climb into the wine jar to attach markers. When removing the markers after scanning, it is easy to leave adhesive residue, which makes non-contact scanning impossible. In addition, the reconstruction accuracy will also be affected if there are water stains or no shape features on the inner wall.

[0026] This invention proposes a device and method for modeling the inner wall of a wine jar.

[0027] Please see Figures 1 to 3 For ease of understanding, this wine jar inner wall modeling device includes: A drive mechanism 100 is installed at the mouth of the wine jar 200 and extends from the mouth into the wine jar 200. The outer casing 300 is mounted at one end of the drive mechanism 100 located inside the wine jar 200. A mounting position is formed on the outer casing 300, and the drive mechanism 100 can drive the outer casing 300 to rotate within the wine jar 200; and... The scanning modeling mechanism 400 is installed at the installation position and is positioned facing the inner wall of the wine jar 200. The drive mechanism 100 can drive the scanning modeling mechanism 400 to rotate inside the wine jar 200 and scan the inner wall of the wine jar 200 to model the wine jar 200.

[0028] Specifically, the wine jar 200 is placed vertically with its opening facing upwards. The drive mechanism 100 is fixedly installed at the opening of the wine jar 200, for example, by means of a flange or clamping structure, to reliably connect to the edge of the opening, ensuring that the entire device does not shift relative to the opening during operation. The main body of the drive mechanism 100 is located above the opening, and its output end extends downwards from the opening into the internal cavity of the wine jar 200, forming a slender extension arm. The length of this extension arm is preset according to the height of the wine jar 200, allowing one end of the drive mechanism 100 inside the wine jar 200 to reach near the bottom of the wine jar 200.

[0029] The outer casing 300 is installed at the end of the drive mechanism 100 that extends into the wine jar 200, i.e., the lower end of the extension arm. The outer casing 300 has a cylindrical or nearly spherical structure, with its outer diameter smaller than the inner diameter of the wine jar 200, to ensure sufficient rotation space and clearance within the wine jar 200. Mounting positions are provided on the side walls of the outer casing 300; these positions can be recesses, flat platforms, or fixed brackets, for supporting the scanning and modeling mechanism 400. The mounting positions are designed to face radially outward, i.e., towards the inner wall of the wine jar 200. The drive mechanism 100, through its internal motor or reduction gear, can drive the outer casing 300 to rotate circumferentially around a vertical axis within the wine jar 200. The rotation angle range covers 360°, and the rotation speed is controllable, typically set in the range of 5~60 rpm to balance scanning efficiency and data acquisition stability.

[0030] The scanning modeling mechanism 400 is fixedly installed at the mounting position of the housing 300, with its scanning probe or transmitting and receiving components facing the inner wall of the wine jar 200. Specifically, the scanning modeling mechanism 400 can employ a laser rangefinder, a structured light projection module, or a high-precision industrial camera 414 assembly. In one embodiment, a line laser scanner using the laser triangulation principle is used as the scanning modeling mechanism 400. This scanner emits a linear laser beam to the inner wall of the wine jar 200, and the receiver captures the reflected light spot. The three-dimensional coordinates of each point relative to the scanner's coordinate system are calculated in real time using trigonometric relationships.

[0031] In this embodiment, during the use of the wine jar 200 inner wall modeling device, the drive mechanism 100 is first securely fixed to the mouth of the wine jar 200. The drive mechanism 100 is then activated, and the outer shell 300, along with the lower end of the extension arm, enters the interior of the wine jar 200 and is positioned near the central axis. Subsequently, the drive mechanism 100 drives the outer shell 300 to rotate continuously or in segments around the central axis inside the wine jar 200. During rotation, the scanning modeling mechanism 400 works synchronously, its scanning direction always remaining radially outward. As the outer shell 300 rotates, the scanning path forms a spiral or multiple parallel circular trajectory, gradually covering the entire area of ​​the wine jar 200 inner wall. By continuously collecting the three-dimensional coordinate data of each point on the inner wall, a complete point cloud dataset is formed. After subsequent processing such as denoising, registration, and meshing, a high-precision three-dimensional model of the wine jar 200 inner wall can be reconstructed. Furthermore, this invention enables non-contact, marker-free, high-precision inner wall scanning modeling. The drive mechanism 100 extends from the mouth of the jar into the interior and drives the outer casing 300 to rotate. This eliminates the need for operators to enter the jar 200 to attach auxiliary markers or manually adjust the equipment position, thus completely eliminating the impact of glue residue on the cleanliness of the jar 200 and meeting the stringent hygiene requirements of brewing sites. Simultaneously, because the scanning and modeling mechanism 400 rotates synchronously with the outer casing 300, it eliminates the need for multiple external setups or repeated equipment relocations, allowing for a one-time comprehensive scan of the entire inner wall. This significantly shortens the acquisition time; typically, scanning a single jar 200 can be completed within minutes, significantly improving efficiency compared to traditional methods that require climbing into the jar or multiple setups.

[0032] In one embodiment, the scanning modeling mechanism 400 includes: Scanning component 410, the scanning component 410 is installed at the mounting position, and the scanning component 410 is positioned facing the inner wall of the wine jar 200; and, The modeling terminal 420 is communicatively connected to the scanning component 410. The modeling terminal 420 is used to receive the point cloud data inside the wine jar 200 obtained by scanning by the scanning component 410 and to model it.

[0033] Specifically, the scanning component 410 is installed at the mounting position of the housing 300, with its detection end facing the inner wall of the wine jar 200. As the driving mechanism 100 rotates the housing 300 around the central axis of the wine jar 200, the scanning component 410 continuously emits structured beams or laser pulses towards the inner wall, and captures the echo signals after diffuse reflection or scattering by the inner wall through a high-sensitivity receiving array. The received signals are amplified, filtered, and converted from analog to digital by the signal conditioning circuit inside the scanning component 410, forming the original distance-angle corresponding sampling sequence, i.e., point cloud data fragments in the local coordinate system. The modeling terminal 420 and the scanning component 410 achieve real-time communication connection through shielded twisted pair cable, Ethernet cable or industrial-grade wireless transmission module. After receiving the continuously transmitted point cloud data stream, the modeling terminal 420 caches it according to the timestamp and attitude label order, and then performs data cleaning (removing outliers and motion blur points), coordinate transformation (converting each local point cloud to a global coordinate system with the central axis of the wine jar 200 as the Z-axis), fine registration of overlapping areas by ICP, point cloud downsampling and normal vector estimation, surface reconstruction (using shielded Poisson reconstruction or moving least squares method to generate a smooth mesh), and model closure and texture mapping, etc. Finally, it outputs a three-dimensional solid model of the inner wall of the wine jar 200 that can be used for volume integral calculation or finite element analysis.

[0034] The drive mechanism 100 is fixedly installed at the mouth of the jar and extends downwards from the mouth. The outer shell 300 is installed at one end of the drive mechanism 100 inside the wine jar 200 and rotates circumferentially with it. The scanning component 410, as the core acquisition unit of the scanning modeling mechanism 400, is directly fixed to the installation position of the outer shell 300. Therefore, the rotation trajectory of the scanning component 410 is strictly controlled by the drive mechanism 100, forming a spiral scanning path with a constant angular velocity or segmented uniform speed. This path achieves 360° continuous coverage in the circumferential direction, and in the vertical direction, through the axial feed or layered pause positioning of the drive mechanism 100, it achieves a complete layer-by-layer scan of the wine jar 200 from the bottom to the shoulder and even the neck area, avoiding the cumulative errors and data loss caused by traditional multi-station setups or manual adjustment of the viewing angle.

[0035] More specifically, the scanning component 410 employs a line structured light scanning module. This module internally includes a semiconductor line laser, a high-frame-rate CMOS camera 414, and a dedicated triangulation processor. The line laser projects a laser stripe approximately 0.5mm to 2mm wide onto the inner wall of the wine jar 200. The CMOS camera 414 captures the stripe deformation image at a fixed baseline distance and tilt angle. The processor calculates the three-dimensional coordinates of each position on the stripe point-by-point according to factory calibration parameters and a real-time image grayscale center extraction algorithm, and packages these coordinates into point cloud frames at a rate of tens of thousands of points per second, transmitting them to the modeling terminal 420 via a gigabit Ethernet interface. The modeling terminal 420 uses an industrial computer equipped with a high-performance GPU, running dedicated point cloud processing pipeline software. After receiving the continuous frames generated by rotation, it first performs coarse registration using the angle increment provided by the rotary encoder built into the drive mechanism 100, and then eliminates accumulated drift using a feature-based NDT algorithm or a globally optimized Bundle Adjustment method, ultimately generating a dense point cloud model with a point spacing of less than 0.2mm and an accuracy better than ±0.1mm. This implementation method is particularly suitable for wine jars 200 with minor glaze undulations or localized corrosion pits on the inner wall. It can clearly capture sub-millimeter-level geometric details, providing a reliable data basis for volume calculation and monitoring of inner wall deformation during the aging process.

[0036] The scanning component 410 uses a ToF laser depth camera 414. This camera 414 integrates a VCSEL laser array and a high-density SPAD pixel array, simultaneously outputting depth values ​​from tens of thousands to hundreds of thousands of sampling points within a single modulation exposure cycle, forming a dense depth image. As the scanning component 410 rotates with the outer casing 300, it triggers an array acquisition every preset angle (e.g., 0.2°–0.5°), simultaneously recording the corresponding rotation angle and axial height information. After receiving the depth image sequence, the modeling terminal 420 projects each frame onto a unified coordinate system based on attitude labels. After removing noise using voxel filtering, it accumulates multi-view depth information using a fusion algorithm based on the truncated signed distance function (TSDF) to generate a volume representation with a voxel resolution of 0.1 mm–0.3 mm. Finally, it extracts the zero-level set to obtain a closed triangular mesh model. This implementation exhibits high robustness to ambient light interference and changes in the diffuse reflectance of the inner wall surface. Even if the lighting inside the wine jar 200 is uneven or there is a small amount of residual wine mist, it can maintain stable depth measurement consistency and complete point cloud coverage.

[0037] In this embodiment, by focusing the scanning component 410 on high-speed, high signal-to-noise ratio raw point cloud acquisition, while placing the modeling terminal 420 outside the wine jar 200 to handle computationally intensive post-processing and model reconstruction, this application effectively avoids the heat dissipation, power consumption, and reliability problems caused by integrating complex computing units inside a small rotating shell 300. Simultaneously, since the modeling terminal 420 can utilize a high-frequency multi-core processor and a professional graphics accelerator card, the entire point cloud processing to model output cycle can be controlled to be completed within tens of seconds to several minutes after the scanning process ends, significantly superior to traditional solutions relying on on-site manual intervention or expensive offline workstation post-processing. This further refinement, while maintaining the core advantages of marker-free, non-contact, and full coverage in a single rotation, achieves efficient collaboration between data acquisition and modeling processes through a reasonable division of hardware and software functions. This provides a complete and reproducible technical path for subsequent applications such as precise metering of the wine jar 200, optimization of ingredient ratios, and traceability of aging quality, demonstrating high engineering practicality and promotional value.

[0038] In one embodiment, a plurality of mounting grooves 310 are formed at the mounting position and are spaced apart circumferentially along the housing 300; The scanning component 410 includes multiple first scanning modules 411 and multiple second scanning modules 412. All first scanning modules 411 and all second scanning modules 412 are installed circumferentially and spaced apart from each other in the mounting groove 310. All first scanning modules 411 and all second scanning modules 412 are arranged facing the inner wall of the wine jar 200, and all first scanning modules 411 and all second scanning modules 412 are communicatively connected to the modeling terminal 420.

[0039] Specifically, the outer surface of the outer casing 300 is machined with four or six mounting slots 310 evenly distributed circumferentially. Each mounting slot 310 has a rectangular groove or dovetail groove structure, and a positioning pin hole and a fastening thread hole are preset in the slot. The first scanning module 411 and the second scanning module 412 are alternately installed in adjacent mounting slots 310, for example, arranged in the order of "first-second-first-second...". The effective detection surface of all modules is radially outward aligned with the inner wall of the wine jar 200. After the drive mechanism 100 is started, the outer casing 300 drives all scanning modules to rotate synchronously. Each module independently collects point cloud data of a local area of ​​the inner wall in its own orientation and uploads the data to the modeling terminal 420 in real time through its own communication interface (e.g., shielded cable or industrial wireless module). After receiving multi-channel point cloud streams from different circumferential orientations, the modeling terminal 420 projects the multi-channel data onto the central coordinate system of the wine jar 200 based on the fixed installation angle of each module relative to the shell 300 (which can be obtained through factory calibration or on-site laser alignment) and the global angle information fed back by the rotary encoder of the drive mechanism 100. Then, it uses a multi-source fusion algorithm (such as graph-optimized global registration or weighted TSDF voxel fusion) to complete the data integration and model reconstruction.

[0040] In this embodiment, by forming multiple circumferentially spaced mounting slots 310 at the installation location and alternately installing the first scanning module 411 and the second scanning module 412 therein, this application achieves parallel acquisition of multiple sensors and complementary heterogeneous data. In actual operation, the operator only needs to place the device into the wine jar 200 from the mouth of the jar, fix the drive mechanism 100, and start the rotational scanning. The entire process does not require manual adjustment of the module angle or multiple entries and exits from the wine jar 200, and can complete the high-density coverage of the entire inner wall in one go. Compared with the traditional single-sensor or external multi-station setup, this structure effectively eliminates the point cloud loss caused by single-view occlusion (such as the large arc area on the shoulder of the wine jar 200 or the bottom corner), while significantly shortening the scanning time (a complete scan of a typical wine jar 200 with a height of 1.2m to 1.8m can be completed within 3 to 8 minutes), and improves the overall modeling accuracy and data integrity to a high level of existing marker-free technology without increasing the size and complexity of the device.

[0041] More specifically, the installation angles of all first scanning modules 411 and second scanning modules 412 are ensured to be consistent through a high-precision mechanical positioning reference surface. During the rotation of the outer shell 300, the sampling triggering of each module is controlled by a unified synchronous pulse from the drive mechanism 100, ensuring that the point clouds collected by adjacent modules have a fixed overlap rate in the circumferential direction. The modeling terminal 420 adopts a time-synchronized multi-channel data stream processing pipeline. First, each point cloud is preprocessed independently (noise filtering, motion distortion correction). Then, fine registration is performed using common-view feature points in the overlapping area. Finally, the final model is generated through a weighted fusion strategy (assigning higher geometric weights to line structured light data and higher density weights to ToF data). This implementation further improves the robustness and consistency of the point cloud under complex internal wall geometry conditions. It is particularly suitable for aged jars with non-uniform corrosion, local glaze loss, or severe curvature changes in the shoulder, and can reliably support subsequent applications such as volume integration, microscopic deformation analysis during the aging process, and optimization of feeding processes.

[0042] In one embodiment, the first scanning module includes: The first mounting base 413 is installed in the corresponding mounting slot 310; and... Camera 414 is mounted on the first mounting base 413, and the lens of camera 414 is set facing the inner wall of wine jar 200.

[0043] Specifically, the first mounting base 413 is fixedly installed in the corresponding mounting groove 310. The first mounting base 413 is typically precision-machined from aluminum alloy or stainless steel, and its bottom surface is flush with the positioning reference surface of the mounting groove 310. High-repeatability positioning and fastening are achieved through countersunk screws or positioning pins. The first mounting base 413 is provided with an inclined mounting platform, which has a preset elevation angle (typically 15° to 35°) relative to the radial direction of the outer shell 300. This allows the lens of the camera 414 mounted on it to be better aimed at the mid-to-long distance area of ​​the inner wall of the wine jar 200, avoiding near-distance blind spots or excessive distortion. The camera 414 is fixedly mounted on the mounting platform of the first mounting base 413, with the optical axis of the lens facing the inner wall of the wine jar 200, ensuring that the field of view of the camera 414 always covers the annular area corresponding to the radial gap between the outer shell 300 and the inner wall during the entire rotational scanning process.

[0044] The outer casing 300 rotates inside the wine jar 200 along with the drive mechanism 100. The camera 414, as the core photoelectric acquisition element of the first scanning module 411, completes circumferential motion together with the outer casing 300. During the actual scanning process, the drive mechanism 100 drives the outer casing 300 to rotate at a constant or segmented controllable angular velocity. Each camera 414 continuously captures a sequence of images of the inner wall of the wine jar 200 at a fixed frame rate (e.g., 20fps to 120fps), while simultaneously recording the corresponding rotation angle (provided by the encoder of the drive mechanism 100) and axial height information (obtained by the axial displacement sensor or layered positioning control of the drive mechanism 100).

[0045] In this embodiment, by mounting the camera 414 on the first mounting base 413 with angle positioning function and distributing all the first scanning modules 411 evenly along the 300° circumference of the outer shell, this application further enhances the image-based multi-view geometric reconstruction capability based on multi-sensor collaborative acquisition. This arrangement effectively compensates for the sparse or hollow point cloud defects that are easily encountered by a single depth sensor in areas with weak texture or high reflectivity. At the same time, without the need to attach any highly reflective markers or coded points to the inner wall, reliable pose estimation and dense matching can be completed by relying on the weak features such as the slight undulations of the natural glaze surface, seam lines, and local color differences of the inner wall, which significantly improves the robustness and integrity of modeling under the condition of no auxiliary markers.

[0046] More specifically, the intrinsic parameters (focal length, principal point, distortion coefficient) and extrinsic parameters (mounting pose relative to the first mounting base 413) of all cameras 414 are uniformly calibrated using a high-precision calibration plate before leaving the factory, and the calibration results are stored in the configuration file of the modeling terminal 420. Before scanning begins, the modeling terminal 420 can automatically select the most suitable operating parameters (gain, exposure time, white balance, etc.) of the camera 414 based on the rough input value of the diameter of the wine jar 200. During the scanning process, the brightness histogram and feature point count of the images of each camera 414 are monitored in real time. If the field of view of a certain camera 414 is overexposed or the feature points decrease sharply, the brightness of the corresponding auxiliary lighting or the exposure parameters of the camera 414 are automatically adjusted to achieve adaptive image quality optimization. This closed-loop control further ensures the consistency and usability of image data for different sizes of wine jars 200 and different inner wall conditions.

[0047] In one embodiment, the second scanning module 412 includes: The second mounting base 415 is installed in a corresponding mounting slot 310, and the second mounting base 415 is spaced apart from the first mounting base 413; and, A line laser device 416 is mounted on a second mounting base 415, with the laser output end of the line laser device 416 facing the inner wall of the wine jar 200.

[0048] Specifically, the second mounting base 415 is fixedly installed in the corresponding mounting groove 310. Its material is typically the same aluminum alloy or stainless steel as the first mounting base 413 to ensure consistent thermal expansion coefficients and mechanical rigidity. The mounting reference surface of the second mounting base 415 is precisely fitted with the mounting groove 310, and reliable fixing with a repeatability accuracy better than 0.02mm is achieved through locating pins and multiple fastening screws. The second mounting base 415 has a dedicated inclined mounting surface with a preset pitch angle (typically 10° to 30°) relative to the radial direction of the outer casing 300. This allows the laser output end of the line laser instrument 416 mounted on it to project onto the inner wall of the wine jar 200 at a suitable incident angle, avoiding excessive compression or stretching deformation of the laser stripes in areas of drastic curvature changes. The line laser instrument 416 is fixedly mounted on the mounting surface of the second mounting base 415, with its laser output end strictly facing the inner wall of the wine jar 200, ensuring that the output line laser stripes always fall within the effective annular detection zone between the outer casing 300 and the inner wall throughout the entire rotational scanning process.

[0049] The line laser instrument 416 employs a semiconductor line laser (typically with a wavelength between 405 nm and 660 nm) combined with a cylindrical lens group to generate a uniform line beam with a width of approximately 0.4 mm to 1.2 mm and a controlled divergence angle. A temperature control module (TEC) is integrated at the laser's rear end to stabilize output power and wavelength drift. The receiving end uses a high-resolution industrial camera 414 (typically with a pixel size of 3.45 μm to 5.5 μm) matched to the laser wavelength. A fixed baseline distance (typically 15 mm to 50 mm, optimized based on the inner diameter of the wine jar 200 and the expected measurement range) is maintained between the camera 414 and the laser. A narrow-bandpass filter is installed in front of the lens of the camera 414, allowing only laser wavelength light to pass through, significantly suppressing interference from diffused light and residual natural light from the internal environment of the wine jar 200. The internal processor of the line laser instrument 416 performs operations such as gray-scale center extraction (with sub-pixel accuracy typically reaching 0.05 to 0.1 pixels), stripe skeletonization, and triangulation on each frame of captured laser stripe image based on the intrinsic parameter matrix, baseline vector, and distortion coefficients obtained from factory calibration. It outputs a local point cloud segment corresponding to the frame in real time (typically 1,000 to 4,000 points per frame) and transmits it to the modeling terminal 420 at high bandwidth via gigabit Ethernet or fiber optic interface.

[0050] The line laser instrument 416 additionally integrates an automatic power adjustment unit, which adjusts the laser output power (typically ranging from 0.5mW to 50mW) based on the average grayscale or saturation pixel ratio of the real-time received image to adapt to changes in the reflectivity of the glaze in different areas of the inner wall of the wine jar 200 (e.g., the smooth glaze of a new jar results in strong reflection, while localized wine stains or corrosion in an aged jar lead to reduced diffuse reflectivity). Simultaneously, the camera 414 dynamically adjusts the gain and exposure time, forming a closed-loop brightness control circuit. This adaptive adjustment ensures that the laser stripes maintain a clear and extractable centerline shape throughout the entire scanning height range, even in the high-curvature transition area at the shoulder or at the bottom corner, avoiding overexposure fusion or underexposure and breakage of the stripes, thus guaranteeing the continuity and geometric accuracy of the point cloud data (typical single-point measurement accuracy ±0.05mm to ±0.10mm).

[0051] In this embodiment, by mounting the line laser 416 on a second mounting base 415 with angle and positioning functions, and by distributing all second scanning modules 412 alternately with the first scanning module 411 along the circumference of the outer shell 300, this application achieves the organic synergy of high-precision structured light measurement and multi-view image reconstruction. In actual operation, the dense, high signal-to-noise ratio geometric point cloud provided by the line laser 416 is mainly responsible for capturing the absolute three-dimensional contour and subtle surface undulations of the inner wall, while the image from the camera 414 of the first scanning module 411 supplements the texture details and pose constraints of weak feature areas. The two are complementary and fused at the modeling terminal 420 through the common viewing features and geometric consistency constraints of the overlapping areas, which can significantly reduce the point cloud holes or matching failures that occur in high reflectivity or low texture areas with single structured light, and finally obtain a complete three-dimensional model of the inner wall with uniform point spacing, continuous surface normal vectors, and geometric accuracy within ±0.08mm.

[0052] In one embodiment, the first scanning module 411 and the second scanning module 412 are distributed at intervals along the height direction of the wine jar 200.

[0053] Specifically, the outer casing 300 has at least two sets of mounting slots 310 arranged in layers along the height direction, with each set of slots maintaining a circumferential alternating arrangement. For example, the first scanning module 411 (with camera 414 as its core) is mainly arranged in one layer of mounting slots 310 in the lower part of the outer casing 300 (near the bottom of the wine jar 200); the second scanning module 412 (with line laser 416 as its core) is mainly arranged in one or more layers of mounting slots 310 in the middle part (corresponding to the main cylindrical section of the wine jar 200); and the first scanning module 411 is rearranged in one layer of mounting slots 310 in the upper part (near the shoulder and neck transition area of ​​the wine jar 200). The axial position of each scanning module is accurately positioned by a precision-machined stepped surface or positioning ring on the outer casing 300. The axial spacing between layers is set according to the typical height and curvature variation characteristics of the wine jar 200 (usually 80mm to 250mm) to ensure that there is a moderate overlap between adjacent scanning layers (typical overlap rate 15% to 35%). When the drive mechanism 100 drives the outer shell 300 to rotate upward / down at a constant speed or in a segmented feeding manner, each layer module synchronously collects the inner wall data of the corresponding height segment and uploads its respective point cloud fragment (with height label) to the modeling terminal 420 in real time.

[0054] The first scanning module 411 (camera type 414) is preferentially deployed in the upper layer of the outer shell 300 corresponding to the large curvature change area on the shoulder of the wine jar 200 and the neck constriction area, as well as the lower layer of the outer shell 300 corresponding to the bottom corner transition area. These areas have drastic changes in curvature radius and complex surface normal vector distribution. The geometric reconstruction method based on multi-view images can utilize weak features such as natural texture, small undulations of the glaze surface, and seam lines, and more reliably recover the local three-dimensional structure through parallax and optical flow constraints between adjacent frames. This avoids measurement blind spots or sharp decreases in accuracy caused by severe compression or overstretching of laser stripes at extreme curvatures by single structured light. The second scanning module 412 (line laser type) is mainly concentrated in the middle layer of the outer shell 300 corresponding to the approximately cylindrical section in the middle of the wine jar 200. The curvature change in this area is relatively gentle and the surface is relatively smooth. Line structured light triangulation can give full play to its sub-millimeter-level lateral and depth resolution advantages and quickly generate high-density point cloud data with extremely high geometric accuracy. After receiving the multi-source point cloud with layered annotations, the modeling terminal 420 first performs coarse stitching based on the height labels of each module and the global Z coordinates provided by the axial displacement sensor of the drive mechanism 100 (resolution is usually ≤0.1mm). Then, it performs fine axial registration using the common-view point cloud features in the overlapping area (e.g., based on FPFH feature descriptor matching or NDT point-to-plane distance minimization). Finally, it uses a layered weighted fusion strategy (giving higher weight to the camera 414 reconstruction data for the shoulder / bottom area and higher weight to the line laser data for the middle area) to generate a continuous and dense overall inner wall model.

[0055] In this embodiment, by arranging the first scanning module 411 and the second scanning module 412 in a layered manner along the height direction, this application achieves adaptive matching for the differences in geometry and surface characteristics of different axial regions of the wine jar 200. In actual operation, after the device is lowered from the mouth of the jar to the bottom, the drive mechanism 100 controls the outer shell 300 to spirally rise from bottom to top (or from top to bottom). The entire scanning process does not require segmented disassembly or repositioning. It can utilize the complementary characteristics of each layer of modules to complete the complete coverage from the bottom rounded corners, the main cylinder wall, to the shoulder and neck of the jar in one continuous movement. Compared with the traditional arrangement where all sensors are located on the same height plane, this layered structure significantly reduces the problems of local point cloud sparseness, voids, or geometric distortion caused by the mismatch of a single sensor type to a specific area. Especially in the large arc transition area of ​​the shoulder and the concave corner of the bottom, the integrity of the point cloud and the continuity of the normal vector are significantly improved. The overall geometric accuracy of the final model can be stably controlled within ±0.10mm, and the surface detail restoration capability is significantly better than that of similar devices without layered arrangement.

[0056] In one embodiment, the drive mechanism 100 includes: Support leg 110 is installed at the mouth of wine jar 200, and a rotating through hole is formed at the center of support leg 110 in a vertical direction. A rotary drive 120 is mounted on a support bracket, and its output end extends downward and rotatably passes through a rotary through-hole; and, A rotating rod 130 is installed at the output end of a rotating drive 120. The rotating rod 130 is located inside the wine jar 200 and is connected to the outer shell 300. The rotating drive 120 can drive the rotating rod 130 to rotate the outer shell 300 so that the scanning modeling mechanism 400 can scan the inner wall of the wine jar 200 to model the wine jar 200.

[0057] Specifically, the support bracket 110 is installed at the mouth of the wine jar 200. It is typically made of stainless steel or high-strength aluminum alloy through welding / casting, and has a three- or four-legged radial structure. The bottom of the bracket is equipped with height-adjustable support feet or rubber shock-absorbing pads to accommodate different mouth diameters of the wine jar 200 (typically ranging from 500mm to 1200mm) and irregular deformations at the mouth edge. A vertically penetrating rotating through hole is machined at the center of the support bracket 110. The inner wall of this through hole is inlaid with a high-precision self-lubricating bushing or rolling bearing seat. The diameter of the through hole is slightly larger than the outer diameter of the rotating rod 130, ensuring that the radial runout of the rotating rod 130 is controlled within 0.05mm when it rotates, while avoiding significant wear during long-term operation.

[0058] The rotary drive component 120 is fixedly mounted on the central platform of the support bracket 110. It typically uses a servo motor or stepper motor paired with a high-precision planetary reducer. Its output end extends vertically downwards and rotatably passes through the rotary through-hole. The housing of the rotary drive component 120 is securely connected to the central platform of the support bracket 110 via multi-point bolts. The output shaft and the rotating rod 130 are connected by a key or tension sleeve to ensure gapless torque transmission and coaxiality better than 0.02mm. The rotating rod 130 is a slender, rigid rod, typically made of stainless steel or carbon steel plated with hard chrome. Its upper end connects to the output end of the rotary drive component 120, and its lower end extends into the wine jar 200 and is fixedly connected to the upper end of the outer casing 300 (using flange bolts, quick clamps, or threaded tightening).

[0059] An additional axial feed guide or lead screw pair can be installed in the middle or lower part of the rotating rod 130. This, in conjunction with the lifting servo motor coaxially mounted with the rotating drive component 120, enables the outer casing 300 to rise or fall uniformly along the axial direction while rotating, forming a spiral scanning trajectory. The pitch (axial displacement per revolution) of this spiral lifting is set according to the height of the wine jar 200 and the vertical field of view height of the scanning module (typical pitch 80mm~200mm) to ensure a reliable overlap area (overlap rate typically 20%~40%) between the scanning bands of adjacent rotation cycles. The modeling terminal 420, based on the real-time attitude information provided by the rotary encoder (resolution ≤0.01°) and the axial displacement sensor (resolution ≤0.1mm), performs time-space tag binding on the point clouds or image frames acquired by each module. Subsequently, it performs global registration and fusion based on the overlapping area features to generate a continuous, stepless, complete 3D model of the inner wall.

[0060] A sealing cover or dust cover is installed above the rotating through hole of the support leg 110. The passage of the rotating rod 130 adopts a lip seal or labyrinth seal structure to effectively prevent the odor of the wine lees, volatile acid mist, or small amounts of wine splashed into the drive component and bearing parts inside the wine jar 200. The surface of the housing 300 of the rotating drive component 120 is coated with a corrosion-resistant coating and equipped with an independent air-cooled or water-cooled heat dissipation channel to ensure that the motor temperature rise is controlled within a reasonable range during long-term continuous operation (typically 4-12 minutes per scan). An electrical rotary joint (slip ring) is provided at the connection between the lower end of the rotating rod 130 and the housing 300 to reliably transmit power, signal, and data cables from the stationary support leg 110 part to the rotating housing 300 part, avoiding cable tangling or breakage.

[0061] In this embodiment, by stably installing the support bracket 110 at the mouth of the jar, fixing the rotary drive component 120 at the center of the bracket and outputting power downwards, and having the rotating rod 130 pass through the through hole and drive the outer shell 300 to rotate inside the wine jar 200, this application achieves stable and controllable circumferential sweeping motion of the scanning and modeling mechanism 400 inside the wine jar 200. In actual operation, the operator only needs to securely set up the support bracket 110 at the mouth of the jar, quickly connect the rotating rod 130 to the outer shell 300, and start the rotary drive component 120 to complete the automatic scanning of the inner wall from the bottom to the neck of the jar. There is no need for personnel to enter the wine jar 200 to adjust the angle or perform multiple positioning operations, which greatly reduces the operational risks and labor intensity. Compared with the traditional solution that requires the use of hoisting equipment or multiple stations to manually set up the scanner, this drive mechanism 100 has a compact structure, is easy to install, and moves smoothly. It can achieve high-density, full-coverage inner wall data acquisition in one continuous movement while maintaining the overall lightweight of the device and non-contact measurement. The final modeling accuracy and data integrity are significantly improved.

[0062] Based on the same technical concept, in a second aspect, the present invention also proposes a method for modeling the inner wall of a wine jar, using the wine jar inner wall modeling device described in the first aspect. The scanning modeling mechanism includes at least one first scanning module and at least one second scanning module. The first scanning module includes a camera, and the second scanning module includes a line laser. The method includes the following steps: S100. Install the device at the mouth of the wine jar, with the first scanning module facing the inner wall side of the wine jar and the second scanning module facing the bottom of the inner wall of the wine jar. S200. When the driving mechanism drives the outer shell to rotate, it controls the line laser to project laser stripes onto the inner wall of the wine jar, and at the same time controls the camera to acquire the laser stripe image. S300. Based on the acquired laser stripe images, the side point cloud and bottom point cloud are reconstructed respectively. S400: Determine the relative pose between the first scanning module and the second scanning module; S500: Based on the relative pose, the side point cloud and the bottom point cloud are fused to obtain a complete point cloud of the inner wall of the wine jar. S600. Establish a model of the inner wall of the wine jar based on the complete point cloud of the inner wall of the wine jar.

[0063] Specifically, in this embodiment, firstly, the three or four support legs of the support bracket 110 are adjusted to be firmly fitted against the edge of the jar mouth, and the deviation between the axis of the central rotating through hole and the geometric symmetry axis of the wine jar 200 is controlled within ±1.5 mm. Then, the lower end of the rotating rod 130 is connected to the top of the outer shell 300 via a high-precision flange or expansion sleeve, so that the outer shell 300 is suspended entirely within the jar space. Through the real-time image preview window of the modeling terminal 420, the initial circumferential angle and axial suspension height of the outer shell 300 are finely adjusted to ensure that the optical axes of all first scanning modules 411 (camera 414 lens) are basically radially pointed towards the barrel wall and shoulder area of ​​the wine jar 200, and that the laser projection surfaces of all second scanning modules 412 (line laser instrument 416) mainly cover the bottom plane of the jar and the lower corner area near the bottom side wall. After installation and positioning, the zero-position height of the device relative to the jar mouth and the inherent assembly external parameters of each module relative to the coordinate system of the outer shell 300 are recorded.

[0064] The rotary drive 120 drives the outer shell 300 to rotate uniformly around the central axis of the jar at a set angular velocity (preferably within the range of 0.6 to 2.8 rpm). Simultaneously, the axial lifting servo controls the outer shell 300 to continuously spiral upward from the bottom of the jar with a matching pitch (preferably within the range of 85 to 170 mm / revolution). During this process, the line lasers 416 of all the second scanning modules 412 project single or double parallel laser lines at a synchronous trigger frequency (preferably 80 to 160 Hz). The laser stripes mainly fall on the bottom of the jar and the area near the bottom sidewall. The high-speed area array camera 414 built into each line laser 416 uses a global shutter and an exposure time matched with the projection frequency to continuously capture stripe deformation images, and immediately completes sub-pixel extraction and distortion correction of the stripe center line, outputting raw measurement data with tags for acquisition time, rotation angle, and axial displacement. At the same time, the cameras 414 of all the first scanning modules 411 continuously capture images at the same or higher frame rate (preferably 70 to 140 fps) to acquire an image sequence that simultaneously includes the strong structural features of the laser stripes and the natural texture of the glaze (microcracks, glaze bubbles, seams, color differences, etc.).

[0065] For the laser stripe image sequence acquired by the second scanning module 412, the image is first distorted using the pre-done camera 414 intrinsic parameter matrix K and distortion coefficients. Then, the center sub-pixel coordinates (uj, vj) of each frame are extracted. Combined with the calibrated laser plane equation (expressed as ax+by+cz+d=0 in the camera 414 coordinate system), the three-dimensional point coordinates (xc, yc, zc) in the camera 414 coordinate system are obtained by solving the ray-plane intersection point. Then, based on the circumferential angle θk and axial displacement sensor data fed back in real time by the rotary encoder, the local point cloud of each frame is transformed to a unified device coordinate system through the rotation matrix R(θk) and corresponding translation transformation, completing the continuous stitching of circumferential and axial directions, generating a high-density bottom point cloud mainly composed of the bottom of the jar and the near-bottom sidewall (typical point spacing 0.07~0.16 mm, depth direction accuracy better than ±0.06 mm).

[0066] For the image sequence acquired by the first scanning module 411, a structured light multi-view reconstruction process assisted by laser stripes is adopted: First, the strong edge features of the laser stripes in each frame are used for initial matching and coarse pose estimation. Then, after removing the stripe area, ORB or SIFT feature extraction, descriptor matching, and RANSAC coarse screening are performed on the natural texture features of the glaze surface. Subsequently, the poses of all cameras 414 and the sparse point cloud are globally optimized through bundle adjustment to obtain the side point cloud (typical point spacing 0.10–0.30 mm) covering the main body of the side wall, the shoulder, and the neck area of ​​the jar. The two types of point clouds have a large spatial overlap area near the bottom side wall and the lower edge of the shoulder, providing a reliable co-view correspondence for subsequent pose unification and fine fusion.

[0067] Using the high geometric accuracy of the bottom point cloud as a benchmark, a rigid transformation is performed on the side point cloud. A calibration sphere or polyhedral calibration target is placed within the common field of view of both modules, and all line lasers (416) are simultaneously triggered to scan the calibration object, extracting the coordinates of the sphere's center or planar feature points. Using at least six pairs of non-collinear corresponding points, the initial rotation matrix R and translation vector T are solved using singular value decomposition (SVD). Subsequently, the Levenberg-Marquardt algorithm is used to nonlinearly optimize the reprojection error of all corresponding point pairs, resulting in a fixed transformation of each camera's (414) coordinate system relative to the second scanning module's (412) coordinate system. This relative pose remains stable after factory calibration, requiring only online verification or fine-tuning upon initial field use or after a certain cumulative operating time threshold.

[0068] First, the side and bottom point clouds are transformed to the same coordinate system using the aforementioned relative poses, achieving preliminary spatial alignment. Then, adaptive weight fusion is applied to the overlapping areas: the bottom point cloud (from the line laser source) is assigned a weight of 0.78–0.93 to the bottom and near-bottom sidewall regions, while the side point cloud (from camera 414) is assigned a weight of 0.68–0.89 to the shoulder and neck regions. The weight function continuously varies with axial height and local curvature gradient. The fused point cloud is then subjected to radius filtering to remove outliers, voxel downsampling (voxel side length 0.09–0.18 mm), and total variation regularization surface smoothing based on anisotropic diffusion filtering. Finally, a complete inner wall point cloud of the wine jar (200mm) with uniform point spacing, continuous normal vectors, and virtually no voids is obtained.

[0069] A shielded Poisson surface reconstruction algorithm, combined with normal vector direction constraints and closed topological priors for the inner wall, generates a smooth, watertight triangular mesh surface model. After reconstruction, color projection texture mapping based on a 414-image sequence from the camera, as well as automatic extraction of key geometric parameters (maximum inner diameter, minimum inner diameter, maximum radius of curvature of the shoulder, roundness deviation of the bottom of the jar, overall volume integral estimation, etc.) can be further performed. The final model is output in standard formats (PLY / OBJ / STL), supporting subsequent capacity verification, crack / glaze peeling detection, digital analysis of brewing processes, or archiving for cultural relic preservation.

[0070] In this embodiment, the second scanning module 412 (line laser instrument 416) is mainly used to acquire high-precision geometric data of the bottom and near-bottom corner areas of the wine jar, while the first scanning module 411 (camera 414) is mainly used to capture rich multi-view information of the texture of the side walls and shoulder / neck areas. The strong structured light feature constraint of the overlapping area of ​​the two modules is used to achieve precise relative pose unification and adaptive weighted fusion. This method effectively solves technical problems such as the blind spot in measuring the concave corner of the bottom of the wine jar 200, the sparse or missing point cloud in the weak texture area of ​​the highly reflective glaze, and the insufficient adaptability of a single module to the differences in geometric characteristics across the entire height. A high-quality 3D model of the inner wall of the wine jar 200 with geometric accuracy better than ±0.10 mm, high surface integrity, and no need for manual marking can be obtained in a single continuous spiral scanning motion. This method has significant automation, measurement consistency, and practical engineering application value.

[0071] In one embodiment, step S300 includes: S310. For each laser stripe image acquired by the camera, extract the pixel coordinates of the center point of the laser stripe; S320. Based on the camera intrinsic parameters and the laser plane equation, the three-dimensional point corresponding to each pixel is calculated by triangulation to obtain a single frame point cloud. S330. Based on the rotation angle recorded during the rotation process, the point clouds of each single frame are stitched together to obtain the side point cloud and the bottom point cloud respectively.

[0072] Specifically, for each frame of laser stripe image acquired by the first scanning module 411 and the second scanning module 412, distortion correction is first performed. The pixel coordinates of the original image are inversely mapped using the pre-calibrated intrinsic parameter matrix K of the camera 414 and the distortion coefficient vectors (radial distortion k1, k2, k3 and tangential distortion p1, p2) to obtain a distortion-free image. Then, Gaussian filtering (σ is typically 1.0–1.8) is used to smooth the image to suppress noise. Next, along the theoretical direction perpendicular to the laser stripe (usually the image v direction), the gray-level centroid method or Steiger Gaussian fitting method is performed on each column of pixels to extract the center pixel coordinates (uj, vj). For multiple laser lines, each stripe can be separated first using gray-level thresholding or morphological operations, and then the center line can be calculated separately. This extraction accuracy is typically 0.02–0.05 pixels, effectively reducing the quantization error in subsequent 3D reconstruction.

[0073] For each extracted fringe center pixel coordinate (uj, vj), it is first back-projected into a normalized ray direction vector d = K in the camera 414 coordinate system using the camera 414 intrinsic parameter matrix K. -¹·[uj,vj,1]^T. Subsequently, the intersection of this ray equation and the pre-calibrated laser plane equation (represented as ax + by + cz + d = 0 in the same camera 414 coordinate system) is calculated, and the three-dimensional coordinates of the intersection point (xc,yc,zc) are determined. The solution process involves parametrically deriving the ray equation Pc = t·d (t>0), substituting it into the plane equation to obtain t = -d / (a·dx + b·dy + c·dz), and then Pc = t·d. This calculation is performed independently within each frame, forming a single-frame local point cloud fragment. Specifically, the single-frame point cloud of the second scanning module 412 (bottom facing) mainly covers the bottom plane and the corner area near the bottom sidewall, with high point density and depth accuracy better than ±0.06 mm; the single-frame point cloud of the first scanning module 411 (side facing) covers the cylinder wall and shoulder / neck area, with moderate point density but containing more surface detail information.

[0074] The rotary drive unit 120 is equipped with a high-resolution incremental encoder (preferably with a resolution ≤0.01°), which synchronously records the current circumferential rotation angle θk for each acquired image frame. Simultaneously, the displacement sensor of the axial lifting servo provides the current axial coordinate zk in real time. Each point Pc in the single-frame point cloud is transformed to a unified device coordinate system through a rigid transformation: Pdevice = R(θk)·Pc + T(zk), where R(θk) is the rotation matrix around the Z-axis, and T(zk) is the translation vector along the Z-axis. Sufficient overlap exists between adjacent frames (circumferential overlap typically 25%–45%, axial overlap 20%–40%). Accumulated errors are eliminated through coarse registration based on collinearity features or ICP, followed by voxel filtering and statistical filtering to remove noise points, ultimately completing continuous stitching. For the data from the second scanning module 412, a high-precision bottom point cloud (typical global point spacing 0.08–0.17 mm) is obtained by stitching together the data, mainly covering the bottom and near-bottom sidewalls. For the data from the first scanning module 411, a side point cloud (typical global point spacing 0.12–0.32 mm) is obtained by stitching together the data, covering the main body of the sidewalls, the shoulder, and the neck area. The two types of point clouds form a reliable spatial overlap in the near-bottom sidewall region, providing sufficient corresponding constraints for subsequent pose unification and fusion.

[0075] In this embodiment, by performing sub-pixel-level center extraction on the laser stripe image, triangulation reconstruction based on calibration parameters, and continuous coordinate transformation stitching driven by rotation angle, this step generates side point clouds and bottom point clouds with high geometric accuracy and complete detail preservation. The bottom point cloud effectively overcomes the measurement difficulties of the concave corners and highly reflective areas of the bottom of the jar, while the side point cloud makes full use of the natural texture of the glaze to supplement the geometric constraints of the weak structured light areas. The spatial overlap between the two lays the foundation for subsequent global fusion. This processing flow can be completed in a single continuous spiral scan, avoiding the cumulative error and efficiency loss caused by traditional multi-region instrument setups. This significantly improves the integrity, consistency, and reconstruction accuracy of the point cloud of the entire inner surface of the 200mm inner wall of the wine jar, effectively solving the technical problems of missing bottom blind areas, sparse side wall weak texture areas, and difficulty in unifying multi-view data in the prior art, providing reliable data support for the subsequent establishment of a high-fidelity 3D model.

[0076] In one embodiment, prior to the step of installing the device at the mouth of the wine jar 200 and positioning the first scanning module 411 facing the inner wall side of the wine jar 200 and the second scanning module 412 facing the bottom of the inner wall of the wine jar 200, the method further includes: The intrinsic parameters of camera 414 are calibrated using a calibration board to obtain the intrinsic parameter matrix and distortion coefficients of camera 414. The calibration plate is placed in different positions, laser stripes are projected, the center point of the stripes is extracted, and the laser plane equation is fitted by the least squares method.

[0077] Specifically, before the device leaves the factory or is used in the field, the classic Zhang Zhengyou calibration method is first used to calibrate the intrinsic parameters of all cameras 414 in the first scanning module 411 and the built-in cameras 414 in the second scanning module 412. A high-precision checkerboard calibration board (the grid side length is usually 20-30 mm) is placed in different postures and distances, and 15-30 images are acquired. Subpixel-level corner coordinates are extracted using a corner detection algorithm. The intrinsic parameter matrix K of the camera 414 (including focal lengths fx and fy, principal points u0 and v0) and distortion coefficients (radial distortion k1-k3, tangential distortion p1 and p2) are solved by least squares optimization. This set of parameters is then fixed in the configuration file of the modeling terminal 420.

[0078] Next, laser plane calibration is performed. The same calibration plate is placed at different positions within the laser projection range (at least 6-10 poses, covering near, medium, and far distances). The control line laser 416 projects one or more laser lines, and corresponding images are acquired. The coordinates of the center pixels of the stripes in each image are extracted using the gray-scale centroid method or Gaussian fitting. These two-dimensional points are back-projected into the ray directions in the camera 414 coordinate system using known intrinsic parameters. Then, utilizing the coplanar constraint between all rays and the corresponding laser plane, the geometric error is minimized using SVD or Levenberg-Marquardt algorithms to fit the laser plane equation ax + by + cz + d = 0 in the camera 414 coordinate system. This calibration result provides accurate structured light constraints for subsequent triangulation measurements.

[0079] The step of determining the relative pose between the first scanning module 411 and the second scanning module 412 includes placing a calibration ball within the common field of view of the first scanning module 411 and the second scanning module 412; controlling the first scanning module 411 and the second scanning module 412 to scan the calibration ball synchronously, obtaining a first point cloud and a second point cloud containing the calibration ball respectively; fitting the center coordinates of the calibration ball from the first point cloud and the second point cloud respectively; and using singular value decomposition to solve the rotation matrix and translation vector from the coordinate system of the second scanning module 412 to the coordinate system of the first scanning module 411 based on at least three sets of center coordinate pairs, as the relative pose.

[0080] Specifically, after the device is assembled or during routine maintenance, a high-precision calibration sphere with a diameter of 50–120 mm is placed within the common field of view of the two types of scanning modules (usually located in the transition area near the bottom sidewall), ensuring that the surface of the sphere is fully covered by laser stripes. All first scanning modules 411 and second scanning modules 412 are simultaneously triggered to perform at least 3–5 scans at different heights or angles, generating a first point cloud (mainly reconstructed by the side camera 414) and a second point cloud (mainly reconstructed by the bottom line laser 416) containing the calibration sphere. For each set of point clouds, a RANSAC algorithm combined with a least-squares spherical fitting algorithm is used to accurately extract the three-dimensional coordinates of the sphere center (with an accuracy typically better than ±0.05 mm). Multiple sets of corresponding sphere center coordinate pairs (at least 3 pairs, non-collinear) are collected, and the initial rotation matrix R and translation vector T of the rigid transformation are directly solved using singular value decomposition (SVD), i.e., the relative pose from the coordinate system of the second scanning module 412 to the coordinate system of the first scanning module 411.

[0081] It also includes a step of nonlinearly optimizing the relative pose: using multiple sets of sphere center coordinate pairs, the Levenberg-Marquardt algorithm is used to minimize the projection error and refine the rotation matrix and translation vector.

[0082] Specifically, after obtaining the initial SVD solution, all corresponding sphere center pairs are substituted into the reprojection error function, and the Levenberg-Marquardt algorithm is used for nonlinear iterative optimization. The error term is defined as minimizing the sum of squared Euclidean distances between the projection of the second point cloud sphere center in the 411 coordinate system of the first scanning module and the actual first point cloud sphere center after the current R and T transformations. After 10 to 30 iterations, the average reprojection error can usually be converged to within 0.03 to 0.08 mm, obtaining higher-precision relative pose parameters, which are then saved to the configuration file.

[0083] The step of fusing the side point cloud and the bottom point cloud according to the relative pose includes: transforming the bottom point cloud to the coordinate system of the side point cloud according to the relative pose to achieve coarse fusion; and using the iterative nearest point algorithm to perform fine registration on the overlapping part of the coarsely fused point cloud to obtain the complete inner wall point cloud of the wine jar 200.

[0084] Specifically, after the actual scanning of the wine jar 200 is completed, the bottom point cloud (from the second scanning module 412) is first transformed to the coordinate system of the first scanning module 411 using the relative pose obtained from the above calibration, achieving preliminary spatial alignment of the two types of point clouds and completing coarse fusion. Subsequently, for the overlapping area (mainly located in the range of approximately 80-250 mm in height near the bottom sidewall), a fine registration is performed using the iterative nearest point (ICP) algorithm, which is either point-to-point or point-to-plane. During the ICP iteration process, reasonable distance thresholds (initially 2-5 mm, gradually converging) and normal vector consistency constraints are set, ultimately controlling the average registration error of the overlapping area within 0.04-0.10 mm, resulting in a seamless and geometrically continuous complete inner wall point cloud of the wine jar 200.

[0085] When the driving mechanism 100 drives the outer shell 300 to rotate, it records the rotation angle for point cloud stitching.

[0086] Specifically, the rotary drive unit 120 integrates a high-precision incremental photoelectric encoder (preferably with a resolution of 0.005° to 0.02°). Whenever image acquisition is triggered, it synchronously reads the current cumulative rotation angle θk and timestamps it together with the reading zk of the axial displacement sensor, associating it with the corresponding single-frame point cloud data. This angle information is used to uniformly transform the local point clouds of each single frame to the global coordinate system of the device through a rotation matrix R(θk) around the Z-axis and a translation T(zk) along the Z-axis, achieving continuous stitching in the circumferential and axial directions and avoiding cumulative drift.

[0087] In this embodiment, by performing high-precision laser plane calibration within the camera 414 before scanning, using a calibration sphere to accurately solve and refine the relative pose of the two modules, combining coarse alignment and ICP fine registration during the fusion stage, and recording rotation angles throughout the process to assist in point cloud stitching, this set of technical features effectively ensures the high spatial consistency and geometric accuracy of multi-source point clouds, significantly reduces the measurement deviation in the bottom corner area and the weak texture area of ​​the side wall, and overcomes the technical defects of the prior art, such as the difficulty in unifying multi-sensor data, the accumulation of calibration errors leading to model distortion, and insufficient registration causing surface steps or voids. Thus, a high-quality point cloud of the inner wall of the wine jar 200 with an overall geometric error better than ±0.10 mm, high surface continuity, and suitable for subsequent accurate modeling can be obtained in a single continuous scan, which has high measurement reliability and engineering practical value.

[0088] The above are merely exemplary embodiments of the present invention and do not limit the scope of the patent of the present invention. All equivalent structural transformations made using the contents of the present invention specification and drawings under the technical concept of the present invention, or direct / indirect applications in other related technical fields, are included within the scope of patent protection of the present invention.

Claims

1. A device for modeling the inner wall of a wine jar, characterized in that, include: A driving mechanism is installed at the mouth of the wine jar and extends from the mouth of the jar into the interior of the wine jar. A casing, wherein the casing is mounted at one end of the drive mechanism located inside the wine jar, the casing having a mounting position formed thereon, and the drive mechanism capable of driving the casing to rotate within the wine jar; and... A scanning modeling mechanism is installed at the installation position and is positioned facing the inner wall of the wine jar. The driving mechanism can drive the scanning modeling mechanism to rotate inside the wine jar and scan the inner wall of the wine jar to model the wine jar. The scanning modeling mechanism includes: A scanning component is installed at the installation location, and the scanning component is positioned facing the inner wall of the wine jar; as well as, A modeling terminal is communicatively connected to the scanning component. The modeling terminal is used to receive point cloud data inside the wine jar obtained by the scanning component and to model it. Multiple mounting slots are formed at intervals along the circumference of the outer shell at the mounting position; The scanning component includes multiple first scanning modules and multiple second scanning modules. All first scanning modules and all second scanning modules are installed circumferentially and spaced apart from each other in the mounting slot. All first scanning modules and all second scanning modules are arranged facing the inner wall of the wine jar, and all first scanning modules and all second scanning modules are communicatively connected to the modeling terminal.

2. The wine jar inner wall modeling device as described in claim 1, characterized in that, The first scanning module includes: A first mounting base, which is installed in a corresponding mounting slot; and... A camera is mounted on the first mounting base, and the lens of the camera is positioned facing the inner wall of the wine jar.

3. The wine jar inner wall modeling device as described in claim 2, characterized in that, The second scanning module includes: A second mounting base is installed in a corresponding mounting slot, and the second mounting base is spaced apart from the first mounting base; and, A line laser is mounted on the second mounting base, with the laser output end of the line laser facing the inner wall of the wine jar.

4. The wine jar inner wall modeling device as described in claim 3, characterized in that, The first scanning module and the second scanning module are distributed at intervals along the height direction of the wine jar.

5. The wine jar inner wall modeling device as described in any one of claims 1 to 4, characterized in that, The drive mechanism includes: A support frame is installed at the mouth of the wine jar, and a rotating through hole is formed at the center of the support frame in a vertical direction; A rotary drive member, mounted on the support leg, the output end of the rotary drive member extending downward and rotatably passing through the rotary through-hole; and, A rotating rod is installed at the output end of the rotating drive component. The rotating rod is located inside the wine jar and connected to the outer shell. The rotating drive component can drive the rotating rod to rotate the outer shell so that the scanning modeling mechanism can scan the inner wall of the wine jar to model the wine jar.

6. A method for modeling the inner wall of a wine jar, characterized in that, Using the wine jar inner wall modeling device as described in any one of claims 1 to 5, the scanning modeling mechanism includes at least one first scanning module and at least one second scanning module, the first scanning module including a camera, and the second scanning module including a line laser, the method comprising the following steps: The device is installed at the mouth of the wine jar, with the first scanning module facing the inner side of the wine jar and the second scanning module facing the bottom of the inner wall of the wine jar. When the drive mechanism drives the outer shell to rotate, it controls the line laser to project laser stripes onto the inner wall of the wine jar, and at the same time controls the camera to acquire the laser stripe image; Based on the acquired laser stripe images, the side point clouds and bottom point clouds were reconstructed respectively. Determine the relative pose between the first scanning module and the second scanning module; The side point cloud and the bottom point cloud are fused together according to the relative pose to obtain a complete point cloud of the inner wall of the wine jar. A model of the inner wall of the wine jar is constructed based on the complete point cloud of the inner wall of the wine jar.

7. The method for modeling the inner wall of a wine jar as described in claim 6, characterized in that, The steps of reconstructing the side point cloud and bottom point cloud based on the acquired laser stripe image include: For each laser stripe image acquired by the camera, the pixel coordinates of the center point of the laser stripe are extracted; Based on the camera intrinsic parameters and the laser plane equation, the three-dimensional point corresponding to each pixel is calculated by triangulation to obtain a single frame point cloud. Based on the rotation angle recorded during the rotation process, the point clouds of each single frame are stitched together to obtain the side point cloud and the bottom point cloud respectively.

8. The method for modeling the inner wall of a wine jar as described in claim 7, characterized in that, Before the step of installing the device on the mouth of the wine jar and positioning the first scanning module facing the inner wall of the wine jar and the second scanning module facing the bottom of the inner wall of the wine jar, the method further includes: The camera's intrinsic parameters are calibrated using a calibration board to obtain the camera's intrinsic parameter matrix and distortion coefficients; The calibration plate is placed in different positions, laser stripes are projected, the center point of the stripes is extracted, and the laser plane equation is fitted by the least squares method.

Images