Search

CN-121982250-A - Method and device for modeling volume of wine jar container and related medium

CN121982250ACN 121982250 ACN121982250 ACN 121982250ACN-121982250-A

Abstract

The invention discloses a volumetric modeling method, a volumetric modeling device and a related medium of a wine jar container, wherein the method comprises the steps of carrying out pose regularization treatment based on a rotating surface on original point cloud data to obtain pose alignment point cloud data; the method comprises the steps of carrying out coding feature extraction processing on gesture alignment point cloud data to obtain hidden vector data, carrying out decoding reconstruction processing on the hidden vector data to obtain an initial grid model, carrying out network training processing according to the initial grid model to obtain a network generation model, carrying out reasoning on the gesture alignment point cloud data by utilizing the network generation model to obtain volume calculation grid data, carrying out multi-height slicing and calculating the internal sectional area of each slice to obtain height-volume curve modeling data. According to the invention, the multi-height slicing is carried out through the volume calculation grid data, and the internal sectional area of each slice is calculated, so that the height-volume curve modeling data is obtained, the effective sectional area of the wine jar can be effectively calculated, and the volume metering precision of the wine jar is improved.

Inventors

  • CHEN ZIYANG

Assignees

  • 天翼物联科技有限公司

Dates

Publication Date
20260505
Application Date
20260119

Claims (10)

  1. 1. A method of modeling the volume of a wine jar container comprising: Acquiring original point cloud data of a wine jar container, and performing pose regularization processing on the original point cloud data based on a rotating surface to obtain pose alignment point cloud data; Performing coding feature extraction processing on the attitude alignment point cloud data to perform hierarchical feature aggregation and output fixed dimension representation to obtain hidden vector data; decoding and reconstructing the hidden vector data, and updating layer by layer based on a preset grid template to obtain an initial grid model; performing network training processing according to the initial grid model to obtain a network generation model; reasoning the attitude alignment point cloud data by using the network generation model to obtain a target grid model, and mapping a bottom thin slice of the target grid model to obtain volume calculation grid data; And carrying out multi-height slicing on the basis of the volume calculation grid data, and calculating the internal sectional area of each slice so as to accumulate layer by layer at slice intervals and carry out curve fitting to obtain height-volume curve modeling data.
  2. 2. The method for modeling the volume of a wine jar container according to claim 1, wherein the steps of obtaining the original point cloud data of the wine jar container, and performing pose regularization processing on the original point cloud data based on a rotation surface to obtain pose alignment point cloud data comprise: Calculating directional bounding box data according to the original point cloud data, taking the longest axis of the directional bounding box data as an initial estimated axis of a wine jar center rotating shaft, and performing primary rotation transformation processing on the original point cloud data to obtain primary rotation point cloud data; Performing upper and lower endpoint cloud removal processing on the preliminary rotation point cloud data based on wine jar structural features to obtain middle blank belly point cloud data, and performing layering slicing processing on the middle blank belly point cloud data to obtain lamellar point cloud data; Performing circular or arc fitting processing on the lamellar point cloud data to obtain a lamellar point data set, and performing singular value decomposition main direction calculation based on the lamellar point data set to obtain a main direction axis to obtain initial pose data of the rotating shaft; establishing an optimization target value according to the initial pose data of the rotating shaft, and carrying out iterative optimization solving processing on the optimization target value to obtain central rotating shaft data; and calculating rigid transformation matrix data by utilizing the central rotation axis data, so that the central rotation axis data is transformed to be overlapped with the vertical axis of the world coordinate system, and posture alignment point cloud data is output.
  3. 3. The method for modeling the volume of a wine jar container according to claim 1, wherein the processing of extracting the coding feature of the attitude alignment point cloud data to perform hierarchical feature aggregation and output a fixed dimension representation to obtain hidden vector data comprises: constructing a hierarchical structure of the encoder by utilizing the attitude alignment point cloud data, and setting a plurality of set abstraction layers to execute neighborhood division layer by layer to obtain hierarchical point set data; performing attention-enhancing feature learning in each of the set abstraction layers according to the hierarchical point set data to aggregate point features in a local region into aggregated point feature data; And performing global maximum pooling on the aggregation point characteristic data after step-by-step processing by a plurality of aggregation abstract layers so as to aggregate the final point characteristic map into hidden vector data with fixed dimension.
  4. 4. The method for modeling the volume of a wine jar container according to claim 1, wherein the decoding and reconstructing the hidden vector data, and updating layer by layer based on a preset grid template to obtain an initial grid model, comprises: inputting and arranging the hidden vector data and a preset standard wine jar template grid, and constructing a decoder as a graph convolution network to obtain a decoded input object; performing multi-layer graph convolution updating operation on the decoding input object to obtain a layer-by-layer updating grid object; Performing anisotropic convolution operation on the layer-by-layer updated grid object, and configuring direction parameters of the characteristics propagating along the preset direction of the grid surface to obtain a direction propagation grid object; And outputting a vertex coordinate matrix according to the direction transmission grid object, and combining the vertex coordinate matrix with the topological structure of the standard wine jar template grid to obtain an initial grid model.
  5. 5. The method for modeling the volume of a wine jar container according to claim 1, wherein the performing a network training process according to the initial mesh model to obtain a network generation model comprises: Collecting three-dimensional scanning data of real wine jars with different models as a real data set, and constructing truth model data based on the real data set; performing data enhancement processing on the truth model data to generate damage input data matched with the truth model data; calculating according to the damage input data to generate prediction grid data, and inputting the prediction grid data and the truth model data into a composite loss function to perform loss calculation to obtain total loss data; and carrying out end-to-end parameter optimization updating on the initial grid model based on the total loss data to obtain a network generation model.
  6. 6. The method for modeling the volume of a wine jar container according to claim 1, wherein the reasoning the attitude alignment point cloud data by using the network generation model to obtain a target grid model, and mapping a bottom thin slice of the target grid model to obtain volume calculation grid data comprises the following steps: Extracting one or more thin slice sections with the height coordinates close to zero in the target grid model, and integrating to obtain polygon directed point set data; performing density clustering on the polygon directed point set data to obtain clustered data; Executing concave-package outline calculation processing by utilizing each cluster in the cluster data to obtain outline polygon set data; and performing nested relation analysis processing on the outline polygon set data to output volume calculation grid data.
  7. 7. The method for modeling the volume of a wine jar container according to claim 1, wherein said calculating the grid data based on the volume to perform multi-height slicing and calculate the internal sectional area of each slice to accumulate and perform curve fitting layer by layer at slice intervals to obtain height-volume curve modeling data, comprising: Setting equally-spaced horizontal slice planes, and cutting the volume calculation grid data from bottom to top at preset slice intervals to obtain multi-height slice sequence data; calculating an internal closed contour set which is obtained by intersecting the volume calculation grid data according to each height slice in the multi-height slice sequence data, and performing internal sectional area calculation processing on the internal closed contour set to obtain slice sectional area sequence data; And calculating the volume of each slice according to the slice interval by the slice sectional area sequence data, accumulating the slice volume layer by layer from bottom to top to obtain a discrete height-volume data point set, and performing function fitting processing on the discrete height-volume data point set to generate continuous curve data to obtain height-volume curve modeling data.
  8. 8. A volumetric modeling apparatus for a wine jar container, comprising: the data acquisition unit is used for acquiring original point cloud data of the wine jar container, and carrying out pose regularization processing on the original point cloud data based on a rotating surface to obtain pose alignment point cloud data; The feature extraction unit is used for carrying out coding feature extraction processing on the attitude alignment point cloud data so as to carry out hierarchical feature aggregation and output fixed dimension representation to obtain hidden vector data; the decoding reconstruction unit is used for decoding reconstruction processing of the hidden vector data and updating layer by layer based on a preset grid template to obtain an initial grid model; the network training unit is used for carrying out network training processing according to the initial grid model to obtain a network generation model; The mapping processing unit is used for reasoning the attitude alignment point cloud data by utilizing the network generation model to obtain a target grid model, and mapping the bottom thin slice of the target grid model to obtain volume calculation grid data; And the data output unit is used for carrying out multi-height slicing on the basis of the volume calculation grid data and calculating the internal sectional area of each slice so as to accumulate layer by layer at slice intervals and carry out curve fitting to obtain height-volume curve modeling data.
  9. 9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing a method of modelling the volume of a wine jar container according to any of claims 1 to 7 when the computer program is executed by the processor.
  10. 10. A computer readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements a method of volumetric modeling of a wine jar container according to any of claims 1 to 7.

Description

Method and device for modeling volume of wine jar container and related medium Technical Field The invention relates to the technical field of data processing, in particular to a volumetric modeling method and device for a wine jar container and a related medium. Background In the field of white spirit base wine production and storage management, in order to meet the high-precision and batched volume metering requirements of a large number of large-scale wine jars, a non-contact measurement scheme based on three-dimensional scanning point clouds has been proposed and applied in the industry, wherein a geometric slice integration thought is more commonly used, namely, the point clouds are sliced along the height direction, the section outline is extracted, the section area is calculated, and then the volumes of all layers are accumulated to obtain the volume. However, under the condition of industrial field collection, the wine jar point cloud is often accompanied with noise, shielding loss and asymmetric interference introduced by an accessory structure, and complicated forms such as pits, foot rings or multiple fulcrums may exist at the bottom of the wine jar, so that the conventional slice product classification scheme is easy to misjudge on the determination of the section profile and the internal effective sectional area, further the volume calculation deviation is obvious, and the requirement on the consistency of the volume metering precision and the automation of the wine jar is difficult to be satisfied. Disclosure of Invention The embodiment of the invention provides a volumetric modeling method and device for a wine jar container and a related medium, and aims to solve the technical problem that in the prior art, the effective sectional area of the interior of a wine jar in an industrial site is easy to be calculated by mistake, so that the volumetric metering accuracy of the wine jar is insufficient. In a first aspect, an embodiment of the present invention provides a method for modeling a volume of a wine jar container, including: Acquiring original point cloud data of a wine jar container, and performing pose regularization processing on the original point cloud data based on a rotating surface to obtain pose alignment point cloud data; Performing coding feature extraction processing on the attitude alignment point cloud data to perform hierarchical feature aggregation and output fixed dimension representation to obtain hidden vector data; decoding and reconstructing the hidden vector data, and updating layer by layer based on a preset grid template to obtain an initial grid model; performing network training processing according to the initial grid model to obtain a network generation model; reasoning the attitude alignment point cloud data by using the network generation model to obtain a target grid model, and mapping a bottom thin slice of the target grid model to obtain volume calculation grid data; And carrying out multi-height slicing on the basis of the volume calculation grid data, and calculating the internal sectional area of each slice so as to accumulate layer by layer at slice intervals and carry out curve fitting to obtain height-volume curve modeling data. In a second aspect, an embodiment of the present invention provides a volumetric modeling apparatus for a wine jar container, comprising: the data acquisition unit is used for acquiring original point cloud data of the wine jar container, and carrying out pose regularization processing on the original point cloud data based on a rotating surface to obtain pose alignment point cloud data; The feature extraction unit is used for carrying out coding feature extraction processing on the attitude alignment point cloud data so as to carry out hierarchical feature aggregation and output fixed dimension representation to obtain hidden vector data; the decoding reconstruction unit is used for decoding reconstruction processing of the hidden vector data and updating layer by layer based on a preset grid template to obtain an initial grid model; the network training unit is used for carrying out network training processing according to the initial grid model to obtain a network generation model; The mapping processing unit is used for reasoning the attitude alignment point cloud data by utilizing the network generation model to obtain a target grid model, and mapping the bottom thin slice of the target grid model to obtain volume calculation grid data; And the data output unit is used for carrying out multi-height slicing on the basis of the volume calculation grid data and calculating the internal sectional area of each slice so as to accumulate layer by layer at slice intervals and carry out curve fitting to obtain height-volume curve modeling data. In a third aspect, an embodiment of the present invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the proce