Three-dimensional model construction method and device, equipment and storage medium

Abstract

This application provides a method, apparatus, device, and storage medium for constructing a 3D model. The method includes: performing image recognition on each original image to determine the target region corresponding to a target object in each original image; performing image segmentation on each original image based on the target region corresponding to the target object to obtain a target image containing only the target object; determining feature information of feature points corresponding to each target image; determining at least one feature point matching the feature points corresponding to the target image based on the feature information; constructing a corresponding point cloud model based on the feature information of the feature points corresponding to the at least one target image and the feature information of the matching feature points; and constructing a corresponding 3D model based on the point cloud model. The method of this application avoids the influence of background information on the model construction process and can improve the accuracy of model construction.

Description

Technical Field

[0001] This application relates to communication technology, and more particularly to a method, apparatus, device and storage medium for constructing a three-dimensional model. Background Technology

[0002] As people's living standards improve, their demand for 3D printing is increasing. The basic principle of 3D printing is to create three-dimensional objects by printing or laying continuous layers of material. 3D printers, also known as 3D printers, construct three-dimensional entities by printing layer by layer.

[0003] Before printing, the object to be printed is photographed using a camera. The model is reconstructed based on the photographed images, and then modeled using computer modeling software. The constructed 3D model is then divided into layers of cross-sections, i.e. slices, to guide the 3D printer to print layer by layer.

[0004] However, the images captured not only contain the printed object but also other background information, which affects the modeling accuracy and speed, thus affecting the final printed 3D entity. Summary of the Invention

[0005] This application provides a method, apparatus, device, and storage medium for constructing three-dimensional models, in order to solve the problem that the presence of other background information in captured images affects the accuracy of modeling.

[0006] Firstly, this application provides a method for constructing a three-dimensional model, including:

[0007] Multiple raw images are acquired, and image recognition is performed on each raw image to determine the target region corresponding to the target object in each raw image;

[0008] Based on the target region corresponding to the target object, perform image segmentation on each original image to obtain a target image containing only the target object;

[0009] Determine the feature information of the feature points corresponding to each target image, and determine the feature points that match the feature points of at least one target image based on the feature information;

[0010] A corresponding point cloud model is constructed based on the feature information of the feature points corresponding to the at least one target image and the feature information of the matched feature points, and a corresponding three-dimensional model is constructed based on the point cloud model.

[0011] Secondly, this application also provides a 3D printing method, comprising:

[0012] Obtain the 3D model;

[0013] 3D printing is performed based on a 3D model, which is obtained by the method described in the first aspect.

[0014] Thirdly, this application provides a three-dimensional model building apparatus, comprising:

[0015] The image acquisition unit is used to acquire multiple raw images;

[0016] An image recognition unit is used to perform image recognition on each original image in order to determine the target region corresponding to the target object in each original image.

[0017] An image segmentation unit is used to segment each original image based on the target region corresponding to the target object to obtain a target image containing only the target object.

[0018] The processing unit is used to determine the feature information of the feature points corresponding to each target image, and to determine at least one feature point matching the feature points of the target image based on the feature information.

[0019] The processing unit is further configured to construct a corresponding point cloud model based on the feature information of the feature points corresponding to the at least one target image and the feature information of the matched feature points, and to construct a corresponding three-dimensional model based on the point cloud model.

[0020] Fourthly, this application also provides a 3D printing apparatus, comprising:

[0021] Acquisition unit, used to acquire 3D models;

[0022] A printing unit for performing 3D printing based on a 3D model, which is obtained by the method described in the first aspect.

[0023] Fifthly, the present invention provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0024] The memory stores computer-executed instructions;

[0025] The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in the first aspect.

[0026] In a sixth aspect, the present invention provides a three-dimensional printing device, comprising: a processor, and a memory communicatively connected to the processor;

[0027] The memory stores computer-executed instructions;

[0028] The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in the second aspect.

[0029] In a seventh aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method as described in the first or second aspect.

[0030] The 3D model construction method, apparatus, device, and storage medium provided in this application acquire multiple original images, perform image recognition on each original image to determine the target region corresponding to the target object in each original image; perform image segmentation on each original image based on the target region corresponding to the target object to obtain a target image containing only the target object; determine the feature information of the feature points corresponding to each target image, and determine the feature points matching the feature points corresponding to at least one target image based on the feature information; construct a corresponding point cloud model based on the feature information of the feature points corresponding to the at least one target image and the feature information of the matching feature points, and construct a corresponding 3D model based on the point cloud model. By performing target detection and image segmentation on the original images, a target image containing only the target object is obtained, removing background information from the original images, avoiding the influence of background information on the model construction process, reducing interference to the model, and using a target image containing only the target object to further construct the 3D model can improve the accuracy of model construction, thereby obtaining a better 3D entity. Attached Figure Description

[0031] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0032] Figure 1 This is a schematic diagram of the network architecture of the three-dimensional model construction method provided by the present invention;

[0033] Figure 2 This is a flowchart illustrating the three-dimensional model construction method provided in Embodiment 1 of the present invention;

[0034] Figure 3 This is a flowchart illustrating the three-dimensional model construction method provided in Embodiment 3 of the present invention;

[0035] Figure 4 This is a flowchart illustrating the three-dimensional model construction method provided in Embodiment 5 of the present invention;

[0036] Figure 5 This is a flowchart illustrating the three-dimensional model construction method provided in Embodiment Six of the present invention;

[0037] Figure 6 This is a schematic diagram of the structure of a three-dimensional model building device provided in an embodiment of the present invention;

[0038] Figure 7This is a schematic diagram of the structure of a three-dimensional printing device provided in an embodiment of the present invention;

[0039] Figure 8 This is a block diagram of an electronic device used to implement the three-dimensional model construction method of the embodiments of the present invention;

[0040] Figure 9 This is a block diagram of a three-dimensional printing device used to implement the three-dimensional printing method of the embodiments of the present invention.

[0041] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to the featured embodiments. Detailed Implementation

[0042] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0043] To clearly understand the technical solution of this application, the solutions of the prior art will be described in detail first.

[0044] A 3D printer, also known as a three-dimensional printer, constructs a three-dimensional object by printing layer by layer. Before printing, the object to be printed is photographed using a photographic device. Based on the photographed images, a model is reconstructed, and a model is created using computer modeling software. The constructed three-dimensional model is then "divided" into layer-by-layer sections, or slices, to guide the 3D printer in printing layer by layer.

[0045] However, the images taken not only contain the printed items but also other background information, such as other items that do not need to be printed. This background information can affect the modeling accuracy, thus affecting the final printed 3D entity.

[0046] Therefore, addressing the issue that background information in existing images affects modeling accuracy, the inventors discovered in their research that acquiring multiple original images and performing image recognition on each image helps determine the target region corresponding to the target object in each image. Based on the target region corresponding to the target object, image segmentation is performed on each original image to obtain a target image containing only the target object. Feature information of feature points corresponding to each target image is determined, and feature points matching at least one feature point corresponding to the target image are identified based on this feature information. A corresponding point cloud model is constructed based on the feature information of the feature points corresponding to at least one target image and the feature information of the matching feature points. A corresponding 3D model is then constructed based on the point cloud model. By performing target detection and image segmentation on the original images, a target image containing only the target object is obtained, removing background information from the original images and avoiding its influence on the model construction process, thus reducing interference with the model. Using a target image containing only the target object to further construct the 3D model improves the accuracy of model construction, resulting in a better-quality 3D entity.

[0047] Therefore, based on the above-mentioned inventive discoveries, the inventors proposed the technical solutions of the embodiments of the present invention. The network architecture and application scenarios of the three-dimensional model construction method provided by the embodiments of the present invention will be introduced below.

[0048] like Figure 1As shown, the network architecture of the 3D model construction method provided in this embodiment of the invention includes: a photo-scanning device 1, an electronic device 2, and a 3D printing device 3. The electronic device 2 is communicatively connected to both the photo-scanning device 1 and the 3D printing device 3. The photo-scanning device 1 is placed in a photography studio and photographs the user from multiple different angles to obtain multiple original images. These original images may include not only the user but also background information such as the photography studio. The electronic device 2 acquires the original images captured by the photo-scanning device 1 and first performs target detection on each original image, i.e., image recognition, to determine the target region corresponding to the target object in each original image. The electronic device 2 performs image segmentation on each original image based on the target region corresponding to the target object, obtaining target images containing only the user. The electronic device 2 determines the feature information of the feature points corresponding to each target image and, based on the feature information, determines at least one feature point matching the feature points corresponding to the target image. The electronic device 2 constructs a corresponding point cloud model based on the feature information of the feature points corresponding to at least one target image and the feature information of the matching feature points, and constructs a corresponding 3D model based on the point cloud model. The 3D printing device 3 acquires the 3D model, performs 3D printing based on the 3D model, and outputs a 3D entity of the user. By performing object detection and image segmentation on the original image, a target image containing only the target object is obtained. This removes background information from the original image, avoids the influence of background information on the model building process, and reduces interference with the model. Using the target image containing only the target object to further build the 3D model can improve the accuracy and speed of model building, thereby obtaining a better 3D entity.

[0049] In another possible application scenario, electronic device 2 acquires multiple original images, performs image recognition on each original image to determine the target region corresponding to the target object in each original image; electronic device 2 performs image segmentation on each original image based on the target region corresponding to the target object to obtain a target image containing only the target object; electronic device 2 performs image recognition on the target image to determine the first local target region corresponding to the local target object in each target image; electronic device 2 determines the feature points corresponding to the first local target region and determines the feature points corresponding to the second local target region, where the second local target region is the region outside the first local target region in the target image. For example, the region where the user's face is located is determined as the first local target region corresponding to the local target object in the target image, and the region outside the user's face is determined as the second local target region. The feature point density of the first local target region is greater than that of the second local target region; feature information of the feature points corresponding to each target image is determined based on the feature points corresponding to the first and second local target regions; electronic device 2 determines the feature points matching the feature points corresponding to at least one target image based on the feature information; electronic device 2 constructs a corresponding point cloud model based on the feature information of the feature points corresponding to at least one target image and the feature information of the matching feature points, and constructs a corresponding 3D model based on the point cloud model. The target image is divided into two local target regions by secondary target detection and image segmentation. Feature points are then extracted from the two local target regions. The feature point density of one local target region is greater than that of the other local target region. This can improve the modeling accuracy of local feature regions without significantly affecting the modeling speed.

[0050] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0051] Example 1

[0052] Figure 2 This is a flowchart illustrating the three-dimensional model construction method provided in Embodiment 1 of the present invention, as shown below. Figure 2 As shown, the execution subject of the three-dimensional model construction method provided in this embodiment is a three-dimensional model construction device, which is located in an electronic device. Therefore, the three-dimensional model construction method provided in this embodiment includes the following steps:

[0053] Step 101: Acquire multiple original images and perform image recognition on each original image to determine the target region corresponding to the target object in each original image.

[0054] In this embodiment, a photographic scanning device is used to take pictures of the target object from multiple angles to obtain multiple original images. The original images include the target object image and other background information. Image recognition is performed on each original image to determine the target area of ​​the target object in each original image.

[0055] Among them, the photo-scanning device can be a camera, which moves the camera to take pictures of the target object from multiple angles. The photo-scanning device can also be a matrix camera, which is multiple cameras set at different angles to take pictures of the target object from multiple angles.

[0056] Step 102: Perform image segmentation on each original image based on the target region corresponding to the target object to obtain a target image containing only the target object.

[0057] In this embodiment, image segmentation is performed on each original image based on the target region corresponding to the target object, which can further exclude background information within the target region and obtain a target image containing only the target object, thereby reducing the impact of background information on the subsequent model reconstruction process.

[0058] Step 103: Determine the feature information of the feature points corresponding to each target image, and determine the feature points that match the feature points of at least one target image based on the feature information.

[0059] In this embodiment, feature information of feature points corresponding to each target image is determined. Feature points are prominent points that do not disappear due to factors such as lighting, scale, or rotation, such as corner points, edge points, bright spots in dark areas, and dark spots in bright areas. Each feature point has corresponding position information, scale information, and orientation information. Descriptors are used to describe the information of pixels surrounding the feature point. The feature point information includes the position information, orientation information, and orientation information of the descriptor corresponding to the feature point. Based on the feature information, at least one matching feature point for a feature point corresponding to a target image is determined. The matching feature point is the feature point that matches the feature point for a feature point in one target image within the feature points of another target image.

[0060] It should be noted that all feature points of a target image may have matching feature points in another target image, or only some feature points of a target image may have matching feature points in another target image. Furthermore, a target image can be matched with multiple other target images separately to determine the matching feature points in each of the different target images.

[0061] Step 104: Construct a corresponding point cloud model based on the feature information of feature points corresponding to at least one target image and the feature information of matched feature points.

[0062] In this embodiment, a corresponding point cloud model is constructed based on the feature information of feature points corresponding to at least one target image and the feature information of matching feature points. Specifically, a sparse point cloud model is constructed based on the feature information of feature points corresponding to at least one target image and the feature information of matching feature points, a dense point cloud model is constructed based on the sparse point cloud model, and a three-dimensional model is constructed based on the dense point cloud model.

[0063] In this embodiment, multiple original images are acquired, and image recognition is performed on each original image to determine the target region corresponding to the target object in each original image. Based on the target region corresponding to the target object, each original image is segmented to obtain a target image containing only the target object. The feature information of the feature points corresponding to each target image is further determined. Based on the feature point information, at least one feature point corresponding to the target image is matched. A corresponding point cloud model is constructed based on the feature information of the feature points corresponding to at least one target image and the feature information of the matched feature points. A three-dimensional model is further constructed based on the point cloud model. By performing target detection and image segmentation on the original images, a target image containing only the target object is obtained, removing the background information in the original images, avoiding the influence of background information on the model construction process, and reducing interference to the model. Using a target image containing only the target object to further construct a three-dimensional model can improve the accuracy and speed of model construction, thereby obtaining a three-dimensional entity with better performance.

[0064] Example 2

[0065] Based on the three-dimensional model construction method provided in Embodiment 1 of the present invention, the step 101 of performing image recognition on each original image to determine the target region corresponding to the target object in each original image has been further refined, including the following steps:

[0066] Step 101a: Input the original images into the preset recognition model in sequence to obtain the recognition result of the modeling object corresponding to the original image through the preset recognition model.

[0067] In this embodiment, the YOLO algorithm is trained in advance to obtain the trained YOLO algorithm. The trained YOLO algorithm is determined as the preset recognition model. The original image is input into the preset recognition model in sequence so as to obtain the recognition result of the modeling object corresponding to the original image through the preset recognition model. The modeling object is the object that can be modeled in the original image, such as any object, animal or person that is recognized, which can be used as the modeling object.

[0068] Step 102b: Select the target object from the recognition results of the modeling object, and determine the region where the target object is located in each original image as the target region corresponding to the target object in the original image.

[0069] In this embodiment, a target object is selected from the recognition results of the corresponding modeling object. For example, if an object, animal, or person is identified, a person is selected as the target object. Furthermore, the area where the target object is located in each original image is determined as the target area corresponding to the target object in the original image.

[0070] In this embodiment, a preset recognition model is used to perform image recognition on the original image, which can obtain a relatively accurate recognition result and automatically select the target object for modeling.

[0071] Example 3

[0072] Figure 3 This is a flowchart illustrating the three-dimensional model construction method provided in Embodiment 3 of the present invention, as shown below. Figure 3 As shown, based on the three-dimensional model construction method provided in Embodiment 1 of the present invention, the image recognition of each original image in step 101 to determine the target region corresponding to the target object in each original image has been further refined, including the following steps:

[0073] Step 1011: Input the original images into the preset recognition model in sequence to obtain the recognition result of the modeling object corresponding to the original image through the preset recognition model.

[0074] In this embodiment, the YOLO algorithm is trained in advance to obtain the trained YOLO algorithm. The trained YOLO algorithm is determined as the preset recognition model. The original image is input into the preset recognition model in sequence, and the recognition result of the modeling object corresponding to the original image is output. The modeling object is the object that can be modeled in the original image, such as any object, animal or person that is recognized.

[0075] YOLO (You Only Look Once) redefines object detection as a regression problem. It applies a single convolutional neural network (CNN) to the entire image, dividing it into a grid and predicting the class probability and bounding box for each grid. For each grid, a bounding box and the probability corresponding to each class (car, pedestrian, traffic light, etc.) are predicted.

[0076] It's important to note that the algorithm isn't limited to YOLO; other one-stage or two-stage algorithms can also be used. One-stage algorithms include SSD (Single Shot MultiBox Detector), which performs dense sampling at different locations on the image, using different scales and aspect ratios. Then, it uses a CNN to extract features and directly performs classification and regression. The entire process requires only one step, making it fast. Two-stage algorithms, such as the R-CNN series, generate a series of sparse candidate boxes using Selective Search or a CNN network, and then classify and regress these boxes. Two-stage algorithms offer high accuracy.

[0077] Step 1012: Display the recognition results of the modeled object on the display interface.

[0078] In this embodiment, different types of modeling objects are identified from the original image and the location of each modeling object is obtained. The location of each modeling object is marked with wireframes and all modeling objects in the original image are displayed on the display interface. Users can select the objects to be modeled through the display interface.

[0079] Step 1013: In response to the modeling object selection operation triggered by the display interface, the selected modeling object is determined as the target object.

[0080] In this embodiment, the user can select at least one target object from the modeling objects. In response to the modeling object selection operation triggered by the display interface, the modeling object selected by the user is determined as the target object.

[0081] Step 1014: Determine the region where the target object is located in each original image as the target region corresponding to the target object in the original image.

[0082] In this embodiment, the region where the target object is located in each original image is determined as the target region corresponding to the target object in the image. Then, based on the target region corresponding to the target object, the original images are segmented to remove the background information in the original images and obtain a target image containing only the target object.

[0083] In this embodiment, a preset recognition model is used to perform image recognition on the original image, which can obtain a relatively accurate recognition result. The recognition result is output and displayed. The user can select the target object from the modeling object according to their needs, which can meet the user's needs.

[0084] Example 4

[0085] Based on the three-dimensional model construction method provided in Embodiment 1 of the present invention, step 102 has been further refined, specifically including the following steps:

[0086] Step 1021: Input each original image into the preset segmentation model in sequence to remove the background region in the target region and output the target image containing only the target object.

[0087] In this embodiment, the Grabcut algorithm is pre-trained to obtain the trained Grabcut algorithm, which is then determined as the preset segmentation model. The target image is input into the preset segmentation model, the area containing the target object in the target region is marked as the foreground region, and other areas in the target region are marked as background regions. The background regions in the target region are removed by the preset segmentation model, and the output is a target image containing only the target object, which is the image corresponding to the foreground region.

[0088] GrabCut is an image segmentation algorithm based on Graph Cut. It starts with the bounding box around the object to be segmented, uses a Gaussian mixture model to estimate the color distribution of the target object and the background, and then performs segmentation.

[0089] It should be noted that the preset segmentation model can also use other algorithms capable of segmenting images, not just those mentioned above.

[0090] In this embodiment, the use of a preset segmentation model can remove background information and segment the target image containing only the target object, effectively avoiding the influence of background information on the model building process, and improving the accuracy and speed of modeling. In addition, by pre-determining the target region, there is no need to consider the region outside the target region when performing image segmentation in the original image, which can improve the efficiency of image segmentation.

[0091] Example 5

[0092] Figure 4 This is a flowchart illustrating the three-dimensional model construction method provided in Embodiment 5 of the present invention, as shown below. Figure 4 As shown, based on the three-dimensional model construction method provided in Embodiment 1 of the present invention, the feature information of the feature points corresponding to each target image in step 103 has been further refined, specifically including the following steps:

[0093] Step 1031: Use the first preset scale-invariant feature transformation (Sift) algorithm to extract the feature points corresponding to each target image.

[0094] In this embodiment, the Scale Invariant Feature Transform (SiFT) algorithm is used to extract feature points corresponding to each target image. Feature points are prominent points that do not disappear due to factors such as illumination, scale, or rotation, such as corner points, edge points, bright spots in dark areas, and dark spots in bright areas. Each feature point has corresponding positional, scale, and orientation information. However, using only this information is insufficient for effective feature point matching, so more detailed information is needed to distinguish feature points. This is where descriptors come in. Descriptors can eliminate the scale and orientation changes in the image caused by changes in viewpoint, enabling better matching between images. Descriptors describe the information of pixels surrounding a feature point, using a set of vectors to describe the feature point. Descriptors include pixels around the feature point that contribute to it. Descriptors possess invariance, robustness, and discriminability. Invariance means that the feature does not change with image scaling or rotation; robustness is insensitive to noise, illumination, or other small deformations; discriminability means that each feature descriptor is unique and exclusive, minimizing similarity between them. The corresponding descriptor information, including orientation information, is determined based on the feature point.

[0095] Step 1032: Determine the feature information of the feature points based on the feature points corresponding to each target image. The feature information includes the position information, orientation information, and descriptor information corresponding to the feature points.

[0096] In this embodiment, the feature information of the feature points is determined based on the feature points corresponding to each target image. The feature information includes the position information of the feature points, the orientation information of the feature points, and the descriptor information corresponding to the feature points.

[0097] In this embodiment, feature points are extracted from the target image to transform image matching into feature point matching, thereby improving matching efficiency.

[0098] Example 6

[0099] Figure 5 This is a flowchart illustrating the five-dimensional model construction method provided in Embodiment Six of the present invention, as shown below. Figure 5 As shown, based on the three-dimensional model construction method provided in Embodiment 1 of the present invention, the feature points for determining at least one feature point matching corresponding to a target image based on feature information in step 103 have been further refined, specifically including the following steps:

[0100] Step 103a: Select one target image from the target images in sequence, and calculate the distance between the feature points corresponding to the selected target image and the feature points corresponding to the unselected target images based on the feature information of the feature points corresponding to the selected target image and the feature information of the feature points corresponding to at least one unselected target image.

[0101] In this embodiment, a target image is selected sequentially from the target images. The distance between each feature point in the selected target image and each feature point in each unselected target image is calculated based on the feature information of the feature points corresponding to the selected target image and the feature information of feature points corresponding to at least one unselected target image. The feature information of the feature points includes the position information, orientation information, and corresponding descriptor information of the feature points. The distances mentioned above can be represented using Hamming distance. It should be noted that "selecting a target image sequentially from the target images" includes pre-numbering the target images according to their orientation and selecting them sequentially according to the numbering order, such as selecting one target image sequentially according to the numbering order, or selecting one target image sequentially at intervals according to the numbering order; or pre-sorting the target images according to their timestamps and selecting one target image sequentially according to the timestamp order, or selecting one target image sequentially at intervals according to the timestamp order.

[0102] Step 103b: Determine the first matching feature point between the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image, based on the distance.

[0103] In this embodiment, the first matching feature point is determined between the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image, based on the distance. During the feature point matching process, there may be mismatched feature points, that is, the first matching feature point may be a mismatched feature point. Therefore, it is necessary to verify the first matching feature point.

[0104] It should be noted that the first matching point refers to the point with the shortest distance between the feature points corresponding to the unselected target image and a feature point in the selected target image. In this embodiment, the first matching point between each feature point corresponding to the selected target image and each feature point corresponding to the unselected target image can be determined based on the distance. That is, each feature point of the selected target image has a first matching point in each unselected target image. In other embodiments, the determination of whether to determine the first matching feature point in the unselected target image can also be based on the positional relationship or similarity between the selected and unselected target images. For example, when the selected and unselected target images are in relative positions, the feature points in the target image do not have corresponding matching feature points in the unselected target image. Even if the first matching feature point is determined according to the above method, the first matching feature point may not be the correctly matched feature point. Therefore, the first matching feature point corresponding to the feature point of the target image may not be determined in the unselected target image. In practice, specific rules can be used to set unselected target images for performing the above feature matching process with the selected target image. For example, when numbering target images according to their orientation, the feature matching process can be performed only between the selected target image and unselected target images within a certain range around it. For example, two unselected target images located above, below, to the left, and to the right of the selected target image.

[0105] Step 103c: Verify the first matching feature points, and determine the first matching feature points that pass the verification as the matched feature points.

[0106] In this embodiment, the first matching feature points are verified, the first matching feature points that pass the verification are retained, the first matching feature points that fail the verification are deleted, and the first matching feature points that pass the verification are determined as the matching feature points.

[0107] Specifically, a feature point matching set can be established. A feature point corresponding to the selected target image and a feature point in an unselected target image that first matches the selected feature point are stored as a matching feature point pair in the feature point matching set. If the first matching feature point passes the verification, the matching feature point pair is saved in the feature point matching set. If the first matching feature point fails the verification, the matching feature point pair is removed from the feature point matching set. The matching feature point pairs that are finally retained in the feature point matching set are used to construct the 3D model.

[0108] In this embodiment, the distance between feature points is calculated based on the feature information of the feature points. Based on the distance, the first matching feature point is determined among the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image. The first matching feature point is further verified to obtain the matching feature points. The feature points extracted after removing background information are feature points of the target image containing only the target object, and not feature points corresponding to other background information. The model is constructed by further matching feature points, which can obtain a more accurate model.

[0109] Optionally, to improve matching efficiency, a combination of ANN (Approximate Nearest Neighbor) and FLANN (Fast Library for Approximate Nearest Neighbors) can be used to determine matching feature points. Specifically, the ANN algorithm divides the selected target image into multiple sub-regions. During the search, it can quickly find regions in the unselected target images that match these sub-regions. Then, it traverses these matching regions and uses the FLANN algorithm for feature matching. FLANN constructs a KD-tree from all points by using one dimension as the partitioning criterion for the high-dimensional data, thus quickly determining the best-matching feature points from the set. In other words, it can calculate only the distances between some feature points in the selected target image and some feature points in the unselected target images, without having to traverse the entire target image.

[0110] Example 7

[0111] Based on the three-dimensional model construction method provided in Embodiment Six of the present invention, step 103b has been further refined, specifically including the following steps:

[0112] Step 103b1: Sort the distances between the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image.

[0113] In this embodiment, the distances between the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image are sorted, and the first matching feature point is determined based on the distance sorting results.

[0114] Step 103b2: Determine the feature point that is closest to the feature point corresponding to at least one unselected target image as the first matching feature point.

[0115] In this embodiment, the feature point closest to the feature point corresponding to at least one unselected target image is determined, and the closest feature point is determined as the first matching feature point.

[0116] In this embodiment, the nearest feature point is determined as the first matching feature point. The first matching feature point is verified, and the first matching feature point that passes the verification is determined as the matching feature point. The feature points extracted after removing background information are further used to build a model based on the feature points and their matching feature points, which can result in a more accurate model.

[0117] Example 8

[0118] Based on the three-dimensional model construction method provided in Embodiment Six of the present invention, the verification of the first matched feature points in step 103c has been further refined, specifically including the following steps:

[0119] Step 103c1: Determine the distance between the feature point corresponding to the selected target image and the first matching feature point as the first distance, and determine the distance between the feature point corresponding to the selected target image and the second matching feature point as the second distance.

[0120] In this embodiment, the distance between the selected feature point corresponding to the target image and the first matched feature point is determined as the first distance, and the distance between the selected feature point corresponding to the target image and the second matched feature point is determined as the second distance. The ratio of the first distance to the second distance is calculated, and the ratio is compared with a preset threshold. Based on the comparison result, it is determined whether the first matched feature point passes the verification.

[0121] Step 103c2: If the ratio of the first distance to the second distance is less than a preset threshold, then the first matched feature point is determined to have passed the verification.

[0122] In this embodiment, if the ratio of the first distance to the second distance is less than a preset threshold, it is considered that there is no matching error, and the first matched feature point is determined to have passed verification. The preset threshold can be set according to actual needs.

[0123] Step 103c3: If the ratio of the first distance to the second distance is greater than a preset threshold, then the first matched feature point is determined to have failed the verification.

[0124] In this embodiment, if the ratio of the first distance to the second distance is less than a preset threshold, it is considered a mismatch, and the feature point of the first match is determined to have passed the verification.

[0125] The preset threshold is less than or equal to 0.8, and preferably, the preset threshold is greater than or equal to 0.4 and less than or equal to 0.6.

[0126] It should be noted that the first matching feature point refers to the closest feature point, and the second matching feature point refers to the second closest feature point.

[0127] In this embodiment, the ratio of the first distance to the second distance can be used to accurately determine whether the first matching feature point passes the verification.

[0128] Optionally, the verification of the first matched feature points in step 103c has been further refined, specifically including the following steps:

[0129] Step 103c4: Select a preset number of first-matched feature points from the first-matched feature points, and determine the initial data model based on the preset number of first-matched feature points.

[0130] In this embodiment, the RANSAC algorithm is used for verification. A preset number of first-matching feature points are selected from the first-matching feature points. The preset number can be set to 4. Four first-matching feature points are selected, and the corresponding transformation matrix is ​​calculated based on the four first-matching feature points. The transformation matrix is ​​then used as the initial data model.

[0131] Step 103c5: Optimize the initial data model to obtain an optimized data model.

[0132] In this embodiment, the initial error probability corresponding to the initial data model is calculated. If the initial error probability is greater than a preset probability, the initial model is optimized, and the initial projection error between each first-matched feature point and the initial model is calculated. If the initial projection error between the first-matched feature point and the initial model is less than a preset value, the first-matched feature point is added to the inlier set. If the initial projection error between the first-matched feature point and the initial model is greater than or equal to a preset value, the first-matched feature point is added to the outlier set. If the number of first-matched feature points in the inlier set is greater than a preset number, a new data model is calculated based on the first-matched feature points in the inlier set, and the error probability corresponding to the new data model is calculated. If the error probability is less than a preset probability, the new data model is determined as the optimized data model. If the error probability is greater than a preset threshold, the above steps are repeated. The preset probability can be set to 0.005.

[0133] Step 1103c6: Calculate the projection error between each first-matched feature point and the optimized data model.

[0134] In this embodiment, the projection error between each first-matched feature point and the optimized data model is calculated, and the validation status of the first-matched feature points is determined based on the projection error. Validated first-matched feature points are retained, while unvalidated first-matched feature points are deleted.

[0135] Step 103c7: If the projection error between the first matched feature point and the optimized data model is less than a preset value, then the first matched feature point is determined to have passed the verification.

[0136] In this embodiment, if the projection error between the first matched feature point and the optimized data model is less than a preset value, it is considered that there is no matching error. Then, the first matched feature point is determined to have passed the verification, and the first matched feature point that has passed the verification is determined as the matched feature point.

[0137] Step 103c8: If the projection error between the first matched feature point and the optimized data model is greater than or equal to a preset value, then the first matched feature point is determined to be verified.

[0138] In this embodiment, if the projection error between the first matched feature point and the optimized data model is greater than or equal to a preset value, it is considered a mismatch, and the first matched feature point is determined to have failed verification.

[0139] In this embodiment, the RANSAC algorithm can accurately determine whether the first matched feature point passes the verification.

[0140] Example 9

[0141] Based on the three-dimensional model construction method provided in Embodiments 1 to 8 of this invention, the step 104 of constructing the corresponding point cloud model based on the feature information of feature points corresponding to at least one target image and the feature information of matched feature points has been further refined, specifically including:

[0142] Step 1041: The Structure for Motion Recovery (SFM) algorithm is used to construct a model based on the feature information of feature points corresponding to at least one target image and the feature information of matched feature points, thereby obtaining a sparse point cloud model.

[0143] In this embodiment, the SFM algorithm is used to construct a model based on the feature information of feature points corresponding to at least one target image and the feature information of matched feature points. Specifically, the intrinsic parameters of the camera module are obtained, and the corresponding intrinsic parameter matrix is ​​determined based on the intrinsic parameters of the camera module. The intrinsic parameters include the camera's focal length and pixel information. Based on the feature information of feature points corresponding to at least one target image, the feature information of matched feature points, and the corresponding intrinsic parameter matrix, the corresponding basis matrix and essential matrix are determined. The essential matrix is ​​decomposed by Singular Value Decomposition (SVD) to obtain the pose information of the camera module. Furthermore, based on the feature information of feature points corresponding to the target image and the feature information of matched feature points, the three-dimensional coordinates corresponding to each feature point are determined. Nonlinear optimization is performed on the intrinsic parameters of the camera module, the pose information of the camera module, and the three-dimensional coordinates corresponding to the feature points to obtain a sparse point cloud model. The nonlinear optimization uses the BundleAdjustment algorithm.

[0144] Step 1041: Construct a dense point cloud model based on the sparse point cloud model.

[0145] In this embodiment, the PMVS algorithm is used to construct a dense point cloud model. Specifically, the PMVS (patch-based multi-view stereo) algorithm is used to construct a dense point cloud model based on each target image, the intrinsic parameters of the camera module, the pose information of the camera module, and the sparse point cloud model.

[0146] In this embodiment, the target image obtained after segmenting the original image contains only the target object. The model is constructed using the target image with background information removed, which can improve the accuracy of the model.

[0147] Example 10

[0148] Based on the three-dimensional model construction method provided in Embodiment 9 of the present invention, the step 104 of constructing the corresponding three-dimensional model based on the point cloud model has been further refined, specifically including the following steps:

[0149] Step 104a: Perform patch processing on the dense point cloud model to obtain the initial 3D model.

[0150] In this embodiment, the dense point cloud model is processed into patches to output an initial patched model; the patched model is then processed into a mesh to output an initial 3D model, wherein the initial 3D model has higher accuracy than the initial patched model.

[0151] Step 104b: Perform surface refinement and / or texture mapping on the initial 3D model to obtain the processed 3D model.

[0152] In this embodiment, the initial 3D model is subjected to surface refinement processing to make the processed 3D model more closely resemble the actual target object, and / or the initial 3D model is subjected to texture mapping processing to make the processed 3D model more similar to the target object.

[0153] In this embodiment, the initial 3D model is optimized to obtain a more accurate 3D model.

[0154] Example 11

[0155] Based on the three-dimensional model construction method provided in Embodiment 1 of the present invention, after step 102, the following steps are also included:

[0156] Step 105: Perform image recognition on the target images to determine the first local target region corresponding to the local target object in each target image.

[0157] In this embodiment, image recognition is performed on the target image to determine the first local target region corresponding to the local target object in each target image. Specifically, the original image is sequentially input into a preset recognition model to obtain the recognition result of the local modeling object corresponding to the original image. The local modeling object is one of multiple objects in the target image that can be locally modeled. For example, if the image recognition identifies the user's whole body, the local modeling objects include the user's head, hands, and face. A local target object is selected from the recognition results of the local modeling objects, and the region where the local target object is located is determined as the first local target region corresponding to the local target object in the target image. For example, the region where the user's face is located is determined as the first local target region corresponding to the local target object in the target image. Alternatively, in some other embodiments, the regions where the user's face and hands are located can also be determined as the first local target region corresponding to the local target object in the target image. The first local target region can be one region or multiple independent regions, and the density of feature points in the multiple regions can also be different.

[0158] The feature information of the feature points corresponding to each target image in step 103 has been further refined, specifically including the following steps:

[0159] Step 103A: Determine the feature points corresponding to the first local target region and the feature points corresponding to the second local target region. The second local target region is the region outside the first local target region in the target image. The feature point density of the first local target region is greater than the feature point density of the second local target region.

[0160] In this embodiment, the region outside the first local target region in the target image is defined as the second local target region. For example, the region outside the user's face in the target image is defined as the second local target region. The first local target region and the second local target region together constitute the target image. Feature points corresponding to the first local target region and the second local target region are determined respectively, wherein the feature point density of the first local target region is greater than that of the second local target region.

[0161] Step 103B: Determine the feature information of the feature points corresponding to each target image based on the feature points corresponding to the first local target region and the feature points corresponding to the second local target region.

[0162] In this embodiment, feature points are prominent points that do not disappear due to factors such as lighting, scale, or rotation, such as corner points, edge points, bright spots in dark areas, and dark spots in bright areas. Each feature point has corresponding position information, scale information, and orientation information. Descriptors are used to describe the information of pixels surrounding the feature point. The feature point information includes the feature point's position information, orientation information, and the orientation information of the descriptor corresponding to the feature point. The feature information of the feature points corresponding to the first local target region and the feature points corresponding to the second local target region of each target image is determined as the feature information of the feature points corresponding to each target image. Further, feature points matching at least one feature point corresponding to the target image are determined based on the feature information; a corresponding point cloud model is constructed based on the feature information of the feature points corresponding to at least one target image and the feature information of the matching feature points; and a corresponding 3D model is constructed based on the point cloud model. In this embodiment, target detection and image segmentation are performed on the target image, dividing it into two local target regions. Feature points are then extracted from these two local target regions, where the feature point density of one local target region is greater than that of the other. This improves the modeling accuracy of the local feature regions without significantly impacting the modeling speed.

[0163] Example 12

[0164] Based on the three-dimensional model construction method provided in Embodiment Eleven of the present invention, the determination of feature points corresponding to the first local target region in step 103A has been further refined, specifically including the following steps:

[0165] Step 103A1: Use the second preset scale-invariant feature transformation (SiFT) algorithm to extract feature points corresponding to the first local target region in each target image.

[0166] In this embodiment, the first local target region and the second local target region together constitute the target image. Different algorithms are used to extract feature points of the two local target regions. The second preset scale-invariant feature transformation (Sift) algorithm is used to extract feature points corresponding to the first local target region in each target image. The use of the second preset scale-invariant feature transformation (Sift) algorithm can make the feature point density of the first local target region greater than that of the second local target region.

[0167] Furthermore, the determination of the feature points corresponding to the second local target region in step 103A was further refined, specifically including the following steps:

[0168] Step 103A2: The third preset scale-invariant feature transformation (SiFT) algorithm is used to extract the feature points corresponding to the second local target region in each target image.

[0169] In this embodiment, the third preset scale-invariant feature transformation (Sift) algorithm is used to extract feature points corresponding to the second local target region in each target image. The feature point density extracted by the third preset scale-invariant feature transformation (Sift) algorithm is different from that of the second preset scale-invariant feature transformation (Sift) algorithm.

[0170] Specifically, the parameters of the second preset scale-invariant feature transformation (Sift) algorithm and the third preset scale-invariant feature transformation (Sift) algorithm are different. For example, the standard deviation of the normal distribution of the Gaussian filter in the second preset scale-invariant feature transformation (Sift) algorithm is lower than the standard deviation of the normal distribution of the Gaussian filter in the third preset scale-invariant feature transformation (Sift) algorithm, thereby making the feature point density of the first local target region greater than that of the second local target region.

[0171] It should be noted that the first preset scale-invariant feature transformation (Sift) algorithm and the third preset scale-invariant feature transformation (Sift) algorithm may be the same or different.

[0172] In this embodiment, the feature point densities obtained by using two different Sift algorithms are different. During modeling, this can improve the modeling accuracy of local feature regions without significantly affecting the modeling speed.

[0173] Example 13

[0174] The 3D printing method provided in this embodiment is executed by a 3D printing device, which is located in a 3D printing equipment. Therefore, the 3D printing method provided in this embodiment includes the following steps:

[0175] Step 201: Obtain the 3D model.

[0176] In this embodiment, the 3D model building device is connected to the 3D printing device. The 3D printing device acquires the 3D model built by the 3D model building device. Specifically, it acquires multiple original images, performs image recognition on each original image to determine the target region corresponding to the target object in each original image, performs image segmentation on each original image based on the target region corresponding to the target object to obtain a target image containing only the target object, determines the feature information of the feature points corresponding to each target image, and determines at least one feature point matching the feature point corresponding to the target image based on the feature information, constructs a corresponding point cloud model based on the feature information of the matched feature points, and constructs a corresponding 3D model based on the point cloud model.

[0177] Step 202: Perform 3D printing based on the 3D model.

[0178] In this embodiment, 3D printing is performed based on a 3D model to output the target object entity. During the construction of the 3D model, the original image is segmented and a target image containing only the target object is used. The target image does not contain background information, which improves the accuracy and speed of model construction. This makes the entity printed based on the 3D model fit the target object better and results in a better 3D entity.

[0179] Figure 6 This is a schematic diagram of the structure of a three-dimensional model building device provided in an embodiment of the present invention, as shown below. Figure 6 As shown, the three-dimensional model construction device 200 provided in this embodiment includes an image acquisition unit 201, an image recognition unit 202, an image segmentation unit 203, and a processing unit 204.

[0180] The system includes an image acquisition unit 201 for acquiring multiple original images. An image recognition unit 202 performs image recognition on each original image to determine the target region corresponding to the target object in each original image. An image segmentation unit 203 segments each original image based on the target region corresponding to the target object to obtain a target image containing only the target object. A processing unit 204 determines the feature information of feature points corresponding to each target image and determines at least one matching feature point for a feature point corresponding to a target image based on the feature information. The processing unit 204 also constructs a corresponding point cloud model based on the feature information of the feature points corresponding to at least one target image and the feature information of the matching feature points, and constructs a corresponding 3D model based on the point cloud model.

[0181] Optionally, the image recognition unit is further configured to sequentially input the original images into a preset recognition model to obtain the recognition result of the modeling object corresponding to the original image through the preset recognition model; select the target object from the recognition result of the modeling object, and determine the area where the target object is located in each original image as the target area corresponding to the target object in the original image.

[0182] Optionally, the image recognition unit is further configured to sequentially input the original images into a preset recognition model to obtain the recognition result of the modeling object corresponding to the original image through the preset recognition model; display the recognition result of the modeling object on the display interface; in response to the modeling object selection operation triggered by the display interface, determine the selected modeling object as the target object; and determine the region where the target object is located in each original image as the target region corresponding to the target object in the original image.

[0183] Optionally, the image segmentation unit is also used to sequentially input each original image into a preset segmentation model, so as to remove the background region in the target region through the preset segmentation model and output a target image containing only the target object.

[0184] Optionally, the processing unit is further configured to extract feature points corresponding to each target image using a first preset scale-invariant feature transformation (SiFT) algorithm; and determine the feature information of the feature points based on the feature points corresponding to each target image, wherein the feature information includes the position information, orientation information, and descriptor information corresponding to the feature points.

[0185] Optionally, the processing unit is further configured to sequentially select a target image from the target images, calculate the distance between the feature points corresponding to the selected target image and the feature points corresponding to the unselected target images based on the feature information of the feature points corresponding to the selected target image and the feature information of the feature points corresponding to at least one unselected target image; determine a first matching feature point among the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image based on the distance; verify the first matching feature point, and determine the first matching feature point that passes the verification as the matching feature point.

[0186] Optionally, the processing unit is further configured to sort the distances between the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image; and to determine the feature point closest to the feature point corresponding to the selected target image among the feature points corresponding to the feature points of the at least one unselected target image as the first matching feature point.

[0187] Optionally, the processing unit is further configured to determine the distance between the feature point corresponding to the selected target image and the first matching feature point as a first distance, and to determine the distance between the feature point corresponding to the selected target image and the second matching feature point as a second distance; if the ratio of the first distance to the second distance is less than a preset threshold, then the first matching feature point is determined to have passed the verification; if the ratio of the first distance to the second distance is greater than the preset threshold, then the first matching feature point is determined to have failed the verification.

[0188] Optionally, the processing unit is further configured to use the Structure for Motion Recovery (SFM) algorithm to construct a model based on the feature information of feature points corresponding to at least one target image and the feature information of matched feature points, thereby obtaining a sparse point cloud model; and to construct a dense point cloud model based on the sparse point cloud model.

[0189] Optionally, the processing unit is also used to perform patch processing on the dense point cloud model to obtain an initial three-dimensional model; and to perform surface refinement processing and / or texture mapping processing on the initial three-dimensional model to obtain a processed three-dimensional model.

[0190] Optionally, the image recognition unit is further configured to perform image recognition on the target image to determine a first local target region corresponding to a local target object in each target image. The processing unit is further configured to determine feature points corresponding to the first local target region and determine feature points corresponding to a second local target region, wherein the second local target region is a region in the target image outside the first local target region, and the feature point density of the first local target region is greater than the feature point density of the second local target region; and determine the feature information of the feature points corresponding to each target image based on the feature points corresponding to the first local target region and the feature points corresponding to the second local target region.

[0191] Optionally, the processing unit is further configured to extract feature points corresponding to the first local target region in each target image using a second preset scale-invariant feature transformation (Sift) algorithm; the processing unit is further configured to extract feature points corresponding to the second local target region in each target image using a third preset scale-invariant feature transformation (Sift) algorithm.

[0192] Figure 7 This is a schematic diagram of the structure of a three-dimensional printing device provided in an embodiment of the present invention, as shown below. Figure 7 As shown, the 3D printing device 300 provided in this embodiment includes an acquisition unit 301 and a printing unit 302.

[0193] The acquisition unit 301 is used to acquire a three-dimensional model. The printing unit 302 is used to perform three-dimensional printing based on the three-dimensional model, which is obtained through any of the above embodiments.

[0194] Figure 8 This is a block diagram of an electronic device used to implement the three-dimensional model construction method of the embodiments of the present invention, such as... Figure 8 As shown, the electronic device 400 includes: a memory 401 and a processor 402.

[0195] Memory 401 stores computer-executed instructions;

[0196] The processor 402 executes computer execution instructions stored in the memory 401, causing the processor 402 to perform the method provided in any one of embodiments one to twelve.

[0197] Figure 9 This is a block diagram of a three-dimensional printing device used to implement the three-dimensional printing method of the embodiments of the present invention, such as... Figure 9 As shown, the 3D printing device 500 includes: a memory 501 and a processor 502.

[0198] Memory 501 stores computer-executed instructions;

[0199] The processor 502 executes computer execution instructions stored in the memory 501, causing the processor 502 to perform the method provided in Embodiment Thirteen.

[0200] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores computer-executable instructions that are executed by a processor using the methods in any of the above embodiments.

[0201] In an exemplary embodiment, a computer program product is also provided, including a computer program that is executed by a processor using the methods of any of the above embodiments.

[0202] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0203] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for constructing a three-dimensional model, characterized in that, The method includes: Multiple raw images are acquired, and image recognition is performed on each raw image to determine the target region corresponding to the target object in each raw image; Based on the target region corresponding to the target object, perform image segmentation on each original image to obtain a target image containing only the target object; Image recognition is performed on the target images to determine the first local target region corresponding to the local target object in each target image; The feature points corresponding to the first local target region are determined, and the feature points corresponding to the second local target region are determined. The second local target region is the region outside the first local target region in the target image. The feature point density of the first local target region is greater than the feature point density of the second local target region. The feature information of each target image is determined based on the feature points corresponding to the first local target region and the feature points corresponding to the second local target region. Based on the feature information, at least one feature point matching the feature point corresponding to the target image is determined; A corresponding point cloud model is constructed based on the feature information of the feature points corresponding to the at least one target image and the feature information of the matched feature points, and a corresponding three-dimensional model is constructed based on the point cloud model.

2. The method according to claim 1, characterized in that, The step of performing image recognition on each original image to determine the target region corresponding to the target object in each original image includes: The original images are sequentially input into a preset recognition model to obtain the recognition result of the modeling object corresponding to the original image through the preset recognition model; Select the target object from the recognition results of the modeling object, and determine the region where the target object is located in each original image as the target region corresponding to the target object in the original image.

3. The method according to claim 1, characterized in that, The step of performing image recognition on each original image to determine the target region corresponding to the target object in each original image includes: The original images are sequentially input into a preset recognition model to obtain the recognition result of the modeling object corresponding to the original image through the preset recognition model; The recognition results of the modeled object are displayed on the display interface; In response to a modeling object selection operation triggered by the display interface, the selected modeling object is determined as the target object; The region where the target object is located in each original image is determined as the target region corresponding to the target object in the original image.

4. The method according to claim 1, characterized in that, The step of segmenting each original image based on the target region corresponding to the target object to obtain a target image containing only the target object includes: Each original image is sequentially input into a preset segmentation model to remove the background region in the target area and output a target image containing only the target object.

5. The method according to claim 1, characterized in that, The feature information includes the location information, orientation information, and descriptive information corresponding to the feature points.

6. The method according to claim 1, characterized in that, The step of determining at least one feature point matching the feature point corresponding to the target image based on the feature information includes: Select one target image at a time from the target images, and calculate the distance between the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image based on the feature information of the feature points corresponding to the selected target image and the feature information of the feature points corresponding to at least one unselected target image. Based on the distance, determine the first matching feature point among the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image; The first matched feature points are verified, and the first matched feature points that pass the verification are determined as the matched feature points.

7. The method according to claim 6, characterized in that, The step of determining the first matching feature point among the feature points corresponding to the selected target image and the feature points corresponding to at least one unselected target image based on distance includes: Sort the distances between the feature points corresponding to the selected target images and the feature points corresponding to at least one unselected target image; The feature point that is closest to the feature point corresponding to at least one unselected target image is determined as the first matched feature point.

8. The method according to claim 6, characterized in that, The verification of the first matched feature points includes: The distance between the feature point corresponding to the selected target image and the first matching feature point is determined as the first distance, and the distance between the feature point corresponding to the selected target image and the second matching feature point is determined as the second distance; If the ratio of the first distance to the second distance is less than a preset threshold, then the first matched feature point is determined to have passed the verification. If the ratio of the first distance to the second distance is greater than a preset threshold, then the first matched feature point is determined to have failed the verification.

9. The method according to any one of claims 1 to 8, characterized in that, The point cloud model includes: a sparse point cloud model and a dense point cloud model; The step of constructing a corresponding point cloud model based on the feature information of the feature points corresponding to the at least one target image and the feature information of the matched feature points includes: The Structure for Motion Recovery (SFM) algorithm is used to construct a model based on the feature information of the feature points corresponding to at least one target image and the feature information of the matched feature points to obtain a sparse point cloud model. Construct a dense point cloud model based on a sparse point cloud model.

10. The method according to claim 9, characterized in that, The step of constructing a corresponding 3D model based on the point cloud model includes: The dense point cloud model is processed into patches to obtain an initial 3D model; The initial 3D model is subjected to surface refinement and / or texture mapping to obtain the processed 3D model.

11. The method according to claim 1, characterized in that, Determining the feature points corresponding to the first local target region includes: The second preset scale-invariant feature transform (SiFT) algorithm is used to extract feature points corresponding to the first local target region in each target image; The determination of the feature points corresponding to the second local target region includes The third preset scale-invariant feature transformation (SiFT) algorithm is used to extract feature points corresponding to the second local target region in each target image.

12. A three-dimensional printing method, characterized in that, The method includes: Obtain the 3D model; 3D printing is performed based on a 3D model, wherein the 3D model is obtained by the 3D model construction method according to any one of claims 1 to 11.

13. A three-dimensional model construction device, characterized in that, The device includes: The image acquisition unit is used to acquire multiple raw images; An image recognition unit is used to perform image recognition on each original image in order to determine the target region corresponding to the target object in each original image. An image segmentation unit is used to segment each original image based on the target region corresponding to the target object to obtain a target image containing only the target object. The processing unit is used to determine the feature information of the feature points corresponding to each target image, and to determine at least one feature point matching the feature points of the target image based on the feature information. The processing unit is further configured to construct a corresponding point cloud model based on the feature information of the feature points corresponding to the at least one target image and the feature information of the matched feature points, and to construct a corresponding three-dimensional model based on the point cloud model. The image recognition unit is further configured to perform image recognition on the target image to determine the first local target region corresponding to the local target object in each target image; The processing unit is specifically used to determine the feature points corresponding to the first local target region and to determine the feature points corresponding to the second local target region. The second local target region is the region outside the first local target region in the target image, and the feature point density of the first local target region is greater than the feature point density of the second local target region. The feature information of each target image is determined based on the feature points corresponding to the first local target region and the feature points corresponding to the second local target region.

14. A three-dimensional printing apparatus, characterized in that, The device includes: Acquisition unit, used to acquire 3D models; A printing unit for performing 3D printing based on a 3D model, wherein the 3D model is obtained by the 3D model construction method according to any one of claims 1 to 11.

15. An electronic device comprising: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1 to 11.

16. A three-dimensional printing apparatus, comprising: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in claim 12.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 12.

Images