Image acquisition method, device, terminal and medium

Abstract

The application provides an image acquisition method and device, a terminal and a medium. The image acquisition method comprises the following steps: acquiring a collection pose relationship based on an image acquisition device, a motion mechanism, a target object and a bearing mechanism; acquiring three-dimensional model data of the target object, and acquiring a first collection track and a second collection track based on the three-dimensional model data; the first collection track connects each first sampling point, the second collection track connects each second sampling point, and the first sampling point and the second sampling point correspond to each other; acquiring each target image based on the first collection track and the second collection track, and acquiring spatial information corresponding to each target image based on a first sampling point and a second sampling point corresponding to each target image and in combination with the collection pose relationship; and storing each target image and the corresponding spatial information as target object collection data, so as to effectively improve the collection efficiency and accuracy of the target object collection data.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence and relates to an image acquisition technology, particularly an image acquisition method, device, terminal and medium. Background Technology

[0002] In the process of artificial intelligence image generation, in order to generate images with high realism and accurate conformity to the laws of the physical world, it is usually necessary to acquire real-world images so that the model can generate images based on these real-world images. Among them, the spatial information of the real-world images is used to reflect the spatial state of the real-world images at the time of acquisition, including but not limited to information such as position, angle, and focal length, so that the model can understand the spatial state of the content in the image.

[0003] Current image acquisition methods typically involve capturing a large number of images using cameras or other image acquisition devices, and then using computer vision algorithms such as structure-of-motion reconstructive motion (SOR) and multi-view stereo vision to estimate the corresponding spatial information based on these images. This approach makes the image acquisition and storage process cumbersome and inconvenient, and subsequent spatial information estimation may contain errors, thus affecting the final image generation result. Furthermore, because the image acquisition process is relatively independent of the AI ​​image generation process, it is impossible to use the results of AI image generation to provide feedback on the image acquisition process, further impacting the final image generation result. Summary of the Invention

[0004] The purpose of this application is to provide an image acquisition method, device, terminal, and medium to solve the problems in the existing image acquisition process, where spatial information is subsequently estimated through computer vision algorithms, which is complex, cumbersome, inconvenient, and has low accuracy. At the same time, the image acquisition process and the artificial intelligence image synthesis process are relatively independent, and no effective feedback mechanism has been established.

[0005] In a first aspect, this application provides an image acquisition method, wherein an image acquisition device is fixedly connected to a motion mechanism, and a target object is fixed to a supporting mechanism, and the image acquisition device acquires an image of the target object, comprising:

[0006] Based on the image acquisition device, the motion mechanism, the target object, and the supporting mechanism, the acquisition pose relationship is obtained;

[0007] The three-dimensional model data of the target object is acquired, and a first acquisition trajectory and a second acquisition trajectory are acquired based on the three-dimensional model data; wherein, the first acquisition trajectory connects each of the first sampling points, the second acquisition trajectory connects each of the second sampling points, and the first sampling points and the second sampling points correspond one-to-one;

[0008] Based on the first acquisition trajectory and the second acquisition trajectory, each target image is acquired, and based on the first sampling point and the second sampling point corresponding to each target image, combined with the acquisition pose relationship, the spatial information corresponding to each target image is obtained; each target image and the corresponding spatial information are stored as target object acquisition data.

[0009] In one embodiment of this application, the method for acquiring the pose relationship includes:

[0010] Based on the structure of the motion mechanism, the pose state of the image acquisition device is obtained;

[0011] Based on the structure of the bearing mechanism, the position and orientation state of the target object is obtained;

[0012] Based on the pose state of the target object and the pose state of the image acquisition device, the acquisition pose relationship is obtained.

[0013] In one embodiment of this application, obtaining the pose state of the image acquisition device based on the structure of the motion mechanism includes:

[0014] Obtain the motion state of the motion mechanism;

[0015] Hand-eye calibration is performed based on the image acquisition device and the motion mechanism, and the calibration results are obtained;

[0016] Based on the motion state of the motion mechanism and the calibration results, the pose state of the image acquisition device is obtained.

[0017] In one embodiment of this application, obtaining the first acquisition trajectory and the second acquisition trajectory based on the three-dimensional model data includes:

[0018] Surface complexity analysis is performed based on the three-dimensional model data to obtain the distribution of viewpoint importance.

[0019] Construct an observation sphere, and obtain each first sampling point and each second sampling point by combining the distribution of the importance of the viewpoint;

[0020] Based on each of the first sampling points, a first acquisition trajectory is generated, and based on each of the second sampling points, a second acquisition trajectory is generated.

[0021] In one embodiment of this application, the method for constructing the three-dimensional model data includes:

[0022] Based on preset shooting angles, acquire each real-shot image of the target object;

[0023] Based on the captured images, features of the target object are extracted, and three-dimensional reconstruction is performed based on the features to generate the three-dimensional model data.

[0024] In one embodiment of this application, the data collected from the target object is synthesized to obtain a synthesized image, and it is determined whether to reshoot based on the synthesized image. If so, each reshoot angle is obtained.

[0025] Based on each of the aforementioned reshooting perspectives, new target images are acquired, and the spatial information corresponding to each of the new target images is obtained, in order to store and update the target object acquisition data.

[0026] In one embodiment of this application, determining whether to perform a reshoot based on the synthesized image, and if so, obtaining each reshoot angle, includes:

[0027] The quality of the synthesized image is checked; if the check passes, no reshoot is required.

[0028] Otherwise, based on the synthesized image, a synthesized pose vector is obtained, and combined with the target object acquisition data, the minimum matching distance is calculated. If the minimum matching distance does not exceed the first matching threshold, no reshoot is performed.

[0029] Otherwise, based on the synthesized pose vector, a preset number of reshoot angles are generated.

[0030] Secondly, this application provides an image acquisition device for acquiring images of a target object, including an image acquisition device, a motion mechanism, and a support mechanism;

[0031] The supporting mechanism is used to fix the target object; the motion mechanism is used to fix the image acquisition device and drive the image acquisition device to move; the acquisition pose relationship is obtained based on the image acquisition device, the motion mechanism, the target object and the supporting mechanism;

[0032] The image acquisition device is used to acquire images of the target object; obtain three-dimensional model data of the target object, and obtain a first acquisition trajectory and a second acquisition trajectory based on the three-dimensional model data; acquire each target image based on the first acquisition trajectory and the second acquisition trajectory, and obtain spatial information corresponding to each target image based on the first sampling point and the second sampling point corresponding to each target image, combined with the acquisition pose relationship; and store each target image and the corresponding spatial information as target object acquisition data.

[0033] The first acquisition trajectory connects each of the first sampling points, and the second acquisition trajectory connects each of the second sampling points, with each of the first sampling points and the second sampling points corresponding one-to-one.

[0034] Thirdly, this application provides a terminal, including: a processor and a memory, wherein the memory and the processor are communicatively connected;

[0035] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to enable the terminal to perform the image acquisition method as described above.

[0036] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a processor, implements the image acquisition method as described above.

[0037] As described above, this application provides an image acquisition method, apparatus, terminal, and medium. By recording the first sampling point of the target object and the second sampling point of the image acquisition device during each image acquisition, the spatial information corresponding to each target image is calculated based on each first sampling point and each second sampling point. This eliminates the need for subsequent estimation using computer vision algorithms, avoiding complex subsequent image calculation processes, improving image acquisition efficiency, and enhancing the accuracy of spatial information. Consequently, it improves the image quality of the synthesized image generated based on each target image. Simultaneously, the process of generating the synthesized image based on the artificial intelligence model provides feedback on the target object acquisition data to determine whether reshooting and updating the target object acquisition data are necessary. This ensures the quality of the generated synthesized image. When the image quality is poor, the updated target object acquisition data enables the artificial intelligence model to generate a higher-quality synthesized image, improving the image synthesis effect and possessing high industrial value. Attached Figure Description

[0038] Figure 1 The diagram shown is a structural schematic of a multi-view imaging device according to an embodiment of this application.

[0039] Figure 2 The diagram shown is a flowchart illustrating an image acquisition method according to an embodiment of this application.

[0040] Figure 3 The diagram shows a flowchart illustrating a method for acquiring three-dimensional model data of a target object as described in an embodiment of this application.

[0041] Figure 4 The diagram shows a flowchart illustrating a method for acquiring a first acquisition trajectory and a second acquisition trajectory as described in an embodiment of this application.

[0042] Figure 5 The diagram shows a flowchart illustrating a method for determining whether to perform a reshoot and obtaining various reshoot angles, as described in an embodiment of this application.

[0043] Figure 6The diagram shown is a structural schematic of a terminal as described in an embodiment of this application.

[0044] Explanation of reference numerals in the attached figures

[0045] 11: Robotic arm for shooting; 12: Horizontal movement rail; 13: Lifting assembly; 14: Pallet; 15: Vertical movement rail; 50: Terminal; 51: Processor; 52: Memory; 521: Operating system; 522: Application program; 53: User interface; 54: Network interface; 55: Bus system. Detailed Implementation

[0046] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0047] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0048] Currently, in the image generation process of artificial intelligence models, it is usually necessary to acquire real-world images to assist the AI ​​model in image generation. Specifically, images of the target object are acquired from multiple perspectives, and after the images are acquired, the corresponding spatial information is estimated to form real-world data of the target object. However, because existing spatial information estimation methods still have certain errors, and the estimation process usually requires a large amount of computation, it reduces the processing efficiency of the system, resulting in low image synthesis efficiency of the AI ​​model and low accuracy of the generated images.

[0049] To address the technical problems existing in the prior art, the following embodiments of this application provide an image acquisition method, device, terminal, and medium. By acquiring the acquisition trajectory and performing image acquisition, target images are obtained, so that the spatial information of the target images can be obtained based on the sampling points corresponding to each target image without subsequent estimation. Each target image and its corresponding spatial information are stored as target object acquisition data, which can not only effectively improve the acquisition efficiency of target object acquisition data, but also improve the accuracy of the spatial information corresponding to each target image, thereby improving the accuracy of the image synthesized by the artificial intelligence model.

[0050] The following embodiments of this application provide an image acquisition method, apparatus, terminal, and medium, including but not limited to applications such as acquiring real-shot images required for image synthesis by artificial intelligence models, acquiring target object information in industrial inspection, and performing 3D reconstruction of target objects. This application does not impose specific limitations on these applications. For example, the following will use the acquisition of real-shot images required for image synthesis by artificial intelligence models as an example to specifically illustrate this application.

[0051] The technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0052] This application provides an image acquisition method for acquiring target images of an object from various perspectives to assist an artificial intelligence model in synthesizing images. The image acquisition device, including but not limited to optical cameras, laser scanners, infrared imagers, and X-ray imagers, is used to acquire images of the target object. The image acquisition device is fixedly connected to a motion mechanism; when the motion mechanism moves, it drives the image acquisition device to move, thereby changing the perspective from which the image acquisition device acquires images of the target object. The target object is the object acquired by the image acquisition device, enabling the artificial intelligence model to generate a synthesized image of the target object based on the acquired target images. The target object is placed on a support mechanism to facilitate image acquisition by the image acquisition device. In some optional embodiments, the target object is fixed to the support mechanism, which is movable, thereby driving the target object to move and further increasing the range of selectable perspectives for the image acquisition device when acquiring images of the target object.

[0053] Optionally, the motion mechanism and the support mechanism can be different mechanical structures of the same device, which can be used to fix the target object and the image acquisition device and perform image acquisition based on the device; for example, the motion mechanism and the support mechanism are two robotic arms of the same device, which are fixedly connected to the target object and the image acquisition device respectively, and the image acquisition device can acquire images of the target object from various perspectives through the mutual movement between the two robotic arms.

[0054] In some alternative implementations, such as Figure 1 As shown, this is a multi-view imaging device. In this device, the robotic arm 11, the horizontal motion slide rail 12, and the lifting assembly 13 serve as motion mechanisms. The robotic arm 11 is fixedly connected to the image acquisition device, which then moves the image acquisition device. The tray 14 and the vertical motion slide rail 15 serve as support mechanisms. The target object is placed on or fixed on the tray 14, and the target object moves by rotating the tray 14 or moving it along the vertical motion slide rail 15.

[0055] Based on this, using image acquisition equipment, the target object, the motion mechanism, and the supporting mechanism, an image acquisition method is employed to acquire multi-view images of the target object, thereby obtaining target object acquisition data. Specifically, such as... Figure 2 As shown, the image acquisition methods include:

[0056] S100 acquires the pose relationship based on the image acquisition device, motion mechanism, target object, and supporting mechanism.

[0057] Among them, the pose relationship is used to characterize the pose relationship between the image acquisition device and the target object during the image acquisition process, that is, to characterize the spatial state when acquiring the image.

[0058] It should be noted that during image acquisition, the changes in spatial state are caused by the mutual movement between the target object and the image acquisition device. Therefore, the acquisition pose relationship is actually determined by the pose states of the target object and the image acquisition device.

[0059] Specifically, the pose of the target object is actually determined by the motion state of the supporting mechanism, which includes the motion state of each joint. For example, if the supporting mechanism is a fixed structure, the pose of the target object is actually a fixed position and angle.

[0060] For example, each joint of the support mechanism is equipped with an encoder, such as a rotary encoder and / or a linear encoder, to read out the state of each joint, thereby obtaining the motion state of the support mechanism, that is, obtaining the pose state of the target object.

[0061] For example, taking a pallet as an example of a supporting mechanism, the target object is fixed at the center point of the pallet. At this time, the center point of the pallet is set as a reference point, and the motion state of the reference point is obtained as the pose state of the target object.

[0062] In some alternative implementations:

[0063]

[0064]

[0065] in, Characterizes the target pose state. The total number of joints representing the load-bearing mechanism. Characterizing the first in the bearing mechanism The mobility of each joint Characterizing the first The length of the link in each joint Characterizing the first The torsion angle of each joint Characterizing the first Linkage offset of each joint. Characterizing the first The joint rotation angle of each joint, Characterizing the first The degree of rotation of each joint Characterizing the first The degree of translation of each joint.

[0066] The pose state of the image acquisition device is actually determined by the motion state of the motion mechanism, which includes the motion state of each joint. For example, each joint of the motion mechanism is equipped with an encoder, such as a rotary encoder and / or a linear encoder, to read the state of each joint, thereby obtaining the motion state of the motion mechanism and thus the pose state of the image acquisition device. Specifically, referring to the method for obtaining the pose state of the target object, the motion state of the fixed mounting end of the motion mechanism is obtained to obtain the pose state of the image acquisition device. Here, the fixed mounting end of the motion mechanism refers to the position on the motion mechanism used for fixedly connecting the image acquisition device.

[0067] It should be noted that due to the gaps and positional offsets between the fixed mounting end of the motion mechanism and the camera or image acquisition sensor of the image acquisition device, the motion state of the motion mechanism cannot be directly used as the motion state of the image acquisition device. Therefore, hand-eye calibration is performed on the image acquisition device to obtain calibration results, which are then combined with the motion state of the fixed mounting end of the motion mechanism to obtain the pose state of the image acquisition device.

[0068] The hand-eye calibration of the image acquisition device includes: fixing the image acquisition device to the motion mechanism, performing multiple image acquisitions on a calibration target of known shape and size, such as a checkerboard calibration board or a circular dot matrix calibration board. During each image acquisition, the corresponding pose of the motion mechanism is obtained, that is, the state of each joint of the motion mechanism is obtained. Based on the acquired images, the corresponding pose of the image acquisition device is calculated. Based on the pose of the motion mechanism and the pose of the image acquisition device, a set of equations is constructed to solve and fit the positional relationship between the motion mechanism and the image acquisition device, which is used as the calibration result.

[0069] For example, the target's pose remains unchanged during hand-eye calibration. A fixed point on the support mechanism is selected as the reference point to establish a three-dimensional Cartesian coordinate system, thereby converting the state of each joint into position coordinates and angles, and then constructing equations. For example, taking a support mechanism including a tray as an example, the target is fixed at the center point of the tray. At this time, the center point of the tray is set as the reference point, and the constructed system of equations is as follows:

[0070]

[0071] in, Used to characterize the number of times images are acquired on the calibration target. Used to characterize the During the second image acquisition, the pose of the fixed mounting end of the motion mechanism relative to the reference point. Used to characterize the pose transformation relationship between the fixed mounting end of the motion mechanism and the image acquisition device. Used to characterize the pose of the calibration target relative to the reference point. Used to characterize the The pose of the image acquisition device relative to the calibration target during the next image acquisition.

[0072] Based on this, based on the The equations are used to solve the pose transformation relationship between the fixed mounting end of the corresponding motion mechanism and the image acquisition device. The pose transformation relationship between the fixed mounting end of the motion mechanism and the image acquisition device is then fitted, and the fitted result is used as the calibration result.

[0073] It should be noted that, in order to improve the accuracy of the calibration results, For example, .

[0074] Based on the acquired target pose state and the image acquisition device pose state, the acquisition pose relationship is obtained. Specifically:

[0075]

[0076] in, To collect pose relationships, This refers to the pose state of the image acquisition device. The position and pose of the motion mechanism can be obtained and represented by reading parameters such as link length, link torsion angle, link offset, and joint rotation angle from the encoders of each joint.

[0077] Based on this, during subsequent image acquisition, the spatial information at the time of acquisition can be directly obtained based on the state of each joint of the motion mechanism and the supporting mechanism, such as reading the encoder readings on each joint.

[0078] Furthermore, in actual execution, the acquired pose relationship can be converted into image acquisition device parameters for storage. That is, based on the acquired pose relationship, the intrinsic parameter matrix, rotation matrix, and translation vector of the image acquisition device are extracted and stored as an image acquisition device parameter set. Those skilled in the art should know the specific methods and principles of extracting the intrinsic parameter matrix, rotation matrix, and translation vector of the image acquisition device, which will not be elaborated here in this embodiment.

[0079] S200: Acquire the three-dimensional model data of the target object, and acquire the first acquisition trajectory and the second acquisition trajectory based on the three-dimensional model data.

[0080] Among them, the three-dimensional model data is used to reflect the approximate geometric shape of the target object, such as the overall shape, outline, and concavity and convexity trends of the target object, so as to analyze the surface state of the target object based on its approximate geometric shape, thereby facilitating the analysis of the viewing angle distribution during the target object image acquisition process. For example, for areas with greater surface undulations and more complex shapes, more and denser acquisition viewing angles should be distributed.

[0081] In some alternative implementations, such as Figure 3 As shown, the methods for obtaining the 3D model data of the target object include:

[0082] S211: Based on preset shooting angles, acquire various real-shot images of the target object.

[0083] Since the 3D model data of the target object is only used to reflect the general geometry of the target object and does not need to reflect the surface details of the target object in detail, this embodiment can obtain 3D model data by pre-setting multiple real-shot perspectives and acquiring real-shot images of the target object based on these real-shot perspectives.

[0084] For example, each real-shot viewpoint is an orthogonal viewpoint; for instance, each captured real-shot image is actually a six-view drawing of the target object.

[0085] Alternatively, starting with mutually orthogonal viewpoints, the viewpoints are iteratively increased. Specifically, based on the current viewpoints, a new viewpoint is acquired such that the increment of the joint visible area coverage corresponding to this new viewpoint is greater than the increment of the joint visible area coverage of any remaining viewpoint. This new viewpoint is added to the current viewpoints to serve as the new current viewpoints. If the increment of the joint visible area coverage corresponding to this new viewpoint is lower than a preset increment threshold, or if the number of new current viewpoints exceeds a preset number of reconstructed images, then the new current viewpoints are used as the actual shooting viewpoints; otherwise, new viewpoints are acquired based on the new current viewpoints. The increment of the joint visible area coverage is used to characterize the increase in the joint visible area coverage corresponding to the current viewpoints after adding a new viewpoint. The joint visible area coverage refers to the ratio of the area observable by each viewpoint to the total area of ​​the target object. Specifically, in actual execution, the number of 3D grids of the target object in a three-dimensional state is usually used to reflect its corresponding area; that is, the joint visible area coverage is the ratio of the number of 3D grids observable by each viewpoint to the total number of 3D grids of the target object. It should be noted that at this stage, a 3D model of the target object has not yet been constructed. Instead, a coarse model of the target object is constructed to obtain the number of 3D meshes and the total number of 3D meshes for the target object. Specifically, a coarse 3D model of the target object is constructed, and based on this coarse 3D model, the number of 3D meshes observable from each viewpoint and the total number of 3D meshes for the target object are obtained. The methods for constructing the coarse 3D model of the target object include, but are not limited to, the bounding box approximation method, the convex hull approximation method, or construction based on pre-acquired initial viewpoints. This embodiment does not impose specific limitations on these methods.

[0086] It should be noted that those skilled in the art should set the specific value of the incremental threshold or the number of reconstructed images based on actual needs, and this embodiment does not make specific limitations here.

[0087] Based on this, various real-shot images of the target object are collected from different shooting angles.

[0088] Furthermore, to improve the quality of the acquired 3D model data, multiple image acquisitions were performed for each shooting perspective, and the highest quality image was selected as the final shooting image. For example, for the same shooting perspective, one or more of the following factors—Laplacian variance, histogram distribution, or lighting conditions—were calculated and analyzed for each acquired image to select the highest quality image. A higher Laplacian variance, a more uniform histogram distribution, and lighting conditions closer to diffuse light conditions generally indicate higher image quality.

[0089] S212: Based on each real-shot image, extract the features of the target object, and perform three-dimensional reconstruction based on each feature to generate three-dimensional model data.

[0090] Specifically, based on a pre-trained foreground segmentation model, such as the SAM (Segment Anything Model), each real-world image is input to extract features of the target object in the image, and 3D reconstruction is performed based on these features to generate 3D model data. The methods for 3D reconstruction based on these features include, but are not limited to, MVS (Multi-View Stereo) mesh reconstruction, NeRF (Neural Radiance Field) implicit surface reconstruction, 3DGS (3D Gaussian Splatting) explicit surface reconstruction, and LRM (Large Reconstruction Model) feedforward reconstruction, etc. This embodiment does not impose specific limitations on these methods.

[0091] It should be noted that those skilled in the art should be familiar with the structure and training methods of foreground segmentation models, such as the SAM model, which will not be elaborated upon in this embodiment.

[0092] Furthermore, for each feature of the acquired target object, color correction is performed on each feature before 3D reconstruction to avoid color differences between different grids in the reconstructed 3D model data due to differences in exposure conditions. Specifically, any image is selected as the calibration reference, for example, the front view in the six views is selected as the calibration reference, and histogram matching specification or white balance correction is used to ensure that the colors of each real-shot image are consistent with the front view.

[0093] Furthermore, the quality of the reconstructed 3D model data is verified to further ensure its quality. Specifically, one or more viewpoints other than the actual shooting viewpoints are selected as verification viewpoints, and images of the target object are acquired based on each verification viewpoint to obtain verification images. Based on the 3D model data and each verification viewpoint, rendered images of the 3D model data under that verification viewpoint are obtained. Based on the verification image and rendered image corresponding to the same verification viewpoint, the similarity index between the two images is obtained, and the average similarity index is calculated based on the similarity index corresponding to each verification viewpoint to reflect the degree of similarity between the 3D model data and the actual target object. The 3D model data passes the verification only when the average similarity index exceeds the similarity threshold; otherwise, the 3D model data is reconstructed based on steps S211 to S212.

[0094] For example, the similarity index is obtained based on the structural similarity index and the perceptual loss. The similarity index and the average similarity index are calculated as follows:

[0095]

[0096]

[0097] in, For the first The similarity index corresponds to each verification perspective. To verify the total number of viewpoints, The average similarity index. For the first The rendered image corresponding to each verification viewpoint For the first Verification images corresponding to each verification viewpoint For the first The structural similarity index between the rendered image and the verification image corresponding to each verification viewpoint For the first The perceptual loss between the rendered image and the verification image corresponding to each verification viewpoint The weighting index is preset. Those skilled in the art can set the specific value of the weighting index according to actual needs. This embodiment does not impose specific limitations here. Exemplarily, .

[0098] It should be noted that those skilled in the art can set the specific value of the similarity threshold according to actual needs, and this embodiment does not impose specific limitations here. For example, the similarity threshold is 0.8.

[0099] Based on this, the three-dimensional model data of the target object is acquired, and the image acquisition method of the target object is set according to the complexity of the target object's surface to determine the first acquisition trajectory and the second acquisition trajectory.

[0100] The first acquisition trajectory is used to characterize the motion trajectory of the target object after it is driven by the carrier mechanism during the image acquisition process; the first acquisition trajectory connects each first sampling point, and each first sampling point is used to characterize the pose of the target object during each image acquisition.

[0101] The second acquisition trajectory is used to characterize the motion trajectory of the image acquisition device after it is driven by the motion mechanism during the image acquisition process; the second acquisition trajectory connects each second sampling point, and each second sampling point is used to characterize the pose of the image acquisition device during each image acquisition.

[0102] It should be noted that each first sampling point corresponds to each second sampling point.

[0103] Based on the first and second acquisition trajectories, the target object and the image acquisition device are moved and images are acquired at the corresponding first and second sampling points in each group, so as to obtain images of each target.

[0104] In some alternative implementations, such as Figure 4 As shown, the methods for acquiring the first and second acquisition trajectories include:

[0105] S221, based on 3D model data, performs surface complexity analysis to obtain the distribution of viewpoint importance.

[0106] The viewpoint importance distribution characterizes the importance of each viewpoint when acquiring images of a target object. Specifically, the importance of each viewpoint is determined by the size of the observable area of ​​the target object from that viewpoint, the degree of surface unevenness and texture complexity. The larger the observable area, the higher the degree of surface unevenness and texture complexity, and the more information can be acquired from that viewpoint during image acquisition, meaning the higher the importance of that viewpoint. Generally, for viewpoints with higher importance, more sampling points are set for image acquisition of that viewpoint and its vicinity.

[0107] Based on this, surface complexity analysis is performed using 3D model data. The degree of surface unevenness and texture complexity, as well as the observed area size, are used to obtain the distribution of the importance of the target object from different perspectives. The image acquisition method for the target object is then set based on this distribution of the importance of the target object.

[0108] For example, the viewpoint importance function reflects the distribution of viewpoint importance of a target object. The viewpoint importance function is calculated as follows:

[0109]

[0110]

[0111]

[0112]

[0113] in, For perspective importance function; Let be the vector of the observed viewpoint; for example, for ease of calculation, It is a unit vector; The curvature complexity score is used to characterize the curvature complexity in terms of viewpoint. The degree of unevenness or undulation of the surface observed below; The pre-defined curvature complexity score weights; Texture feature density fraction, used to characterize the viewpoint The complexity of the surface texture observed below; The preset texture feature density score weights; The visible area coverage fraction is used to characterize the viewpoint. The size of the observed area; The visible area coverage score weights are preset. It should be noted that those skilled in the art should set specific values ​​for the curvature complexity score weight, texture feature density score weight, and visible area coverage score weight based on actual needs. This embodiment does not impose specific limitations here; it is merely an example. , , ; From the perspective The set of all observable 3D meshes on the lower 3D model data. The first in the characterization A three-dimensional grid; Used to characterize the The area of ​​a three-dimensional grid; Used to characterize the Gaussian curvature of a three-dimensional grid; Used to characterize the The edge pixel density of a 3D grid; Used to characterize the The unit normal vector of a 3D mesh; Used to represent sets The total number of three-dimensional grids in the data.

[0114] S222, construct the observation sphere, and combine the distribution of the importance of the viewpoint to obtain the first sampling point and the second sampling point.

[0115] For example, the relative motion between the image acquisition device and the target object is as follows: the image acquisition device moves around the spherical surface of the target object. Here, the observed spherical surface is a sphere with the target object as its center and the acquisition distance as its radius. It should be noted that the acquisition distance is actually the distance between the target object and the image acquisition device, and the acquisition distance is set based on the focal length of the image acquisition device; this embodiment does not impose specific limitations on this.

[0116] Based on the distribution of viewpoint importance, sampling positions are set on the observation sphere. Specifically, for viewpoints with higher importance, the probability of setting a sampling position is higher, thereby obtaining each sampling position. In some optional embodiments, the total number of samples is preset, and the probability of setting a sampling position at each location on the observation sphere is obtained based on the distribution of viewpoint importance, thereby obtaining each sampling position. Those skilled in the art should set the specific number of total sampling positions based on actual needs; this embodiment does not impose specific limitations. For example, the total number of samples is no more than 360 and no less than 36.

[0117] To facilitate understanding of the method for obtaining each sampling position in this embodiment by those skilled in the art, the following will take the example of reflecting the distribution of the visual importance of the target object through the visual importance function to specifically explain the method for obtaining each sampling position.

[0118] Specifically, for ease of calculation, the viewpoint importance function is... Normalization is performed to obtain the sampling probability density function, which reflects the probability of setting the sampling position at each location on the observed sphere:

[0119]

[0120] in, Let be the sampling probability density function. The heading angle corresponding to the viewpoint. The pitch angle corresponding to the viewpoint. This represents the maximum pitch angle. This is the minimum pitch angle, because the target object is placed on the support mechanism, and the pitch angle of the image acquisition device is usually limited.

[0121] Based on the sampling probability density function and a preset total number of samples, each sampling position is obtained. For example, by using an inverse transformation sampling method, the observed sphere is divided into multiple small regions, and the sampling probability density function is discretized based on each small region. Thus, a total number of sampling positions are selected according to their probability from largest to smallest. Alternatively, a first number of uniformly distributed initial sampling positions is generated on the observed sphere, which is greater than the total number of samples. Based on the sampling probability density function, the initial sampling position with the lowest probability is reduced each time until the number of remaining initial sampling positions equals the total number of samples. Or, a second number of uniformly distributed initial sampling positions is generated on the observed sphere, which is less than the total number of samples. Based on the sampling probability density function, a new sampling position with the highest probability is added each time until the total number of initial sampling positions and new sampling positions equals the total number of samples. Based on this, each sampling position can be obtained.

[0122] It should be noted that both the target object and the image acquisition device can move during the actual image acquisition process. Based on this, the poses of the target object and the image acquisition device are set according to each sampling position. The pose includes the position and the rotation angle, so that the relative position between the target object and the image acquisition device conforms to the sampling position. The pose of the target object is stored as the first sampling point, and the pose of the image acquisition device is stored as the second sampling point, so as to obtain the corresponding first sampling point and second sampling point for each group.

[0123] S223, based on each first sampling point, generate a first acquisition trajectory, and based on each second sampling point, generate a second acquisition trajectory.

[0124] Specifically, each first sampling point is connected in sequence to form a first acquisition trajectory, and each second sampling point is connected in sequence to form a second acquisition trajectory.

[0125] Based on the first and second acquisition trajectories, the target object and the image acquisition device are moved, and image acquisition is performed when the target object reaches the first sampling point and the image acquisition device reaches the corresponding second sampling point.

[0126] S300 acquires images of each target based on the first acquisition trajectory and the second acquisition trajectory, and obtains spatial information corresponding to each target image based on the first sampling point and the second sampling point corresponding to each target image, combined with the acquisition pose relationship; and stores each target image and the corresponding spatial information as target object acquisition data.

[0127] Spatial information is used to characterize the spatial state corresponding to the acquired target image, including but not limited to position, angle, focal length, and other information. Specifically, spatial information of each target image is obtained based on the pose relationship between the target object and the image acquisition device during the acquisition of each target image.

[0128] It should be noted that the target object is moved by the supporting mechanism. Therefore, to facilitate subsequent calculations, the state of each joint of the supporting mechanism is used to reflect the pose of the target object. Specifically, based on the relationship between the pose of the target object and the states of each joint of the supporting mechanism... , obtain the pose corresponding to each first sampling point.

[0129] The image acquisition device is driven by a motion mechanism. Similarly, the pose of the image acquisition device is reflected by the state of each joint of the motion mechanism. It should be noted that due to the gaps and positional offsets between the fixed mounting end of the motion mechanism and the camera or image acquisition sensor of the image acquisition device, the motion state of the motion mechanism cannot be directly used as the motion state of the image acquisition device. Therefore, based on the calibration results obtained in step S100, and combined with the state of each joint of the motion mechanism, the pose corresponding to each second sampling point is calculated. Specifically:

[0130]

[0131] It should be noted that, in the actual execution process, the first acquisition trajectory connects each first sampling point in sequence, and the state of each joint of the bearing mechanism is obtained based on each first sampling point. Based on the order in the first acquisition trajectory, the state of each joint of the bearing mechanism is changed in sequence so that the target object reaches each first sampling point. Similarly, the state of each joint of the motion mechanism is changed in sequence so that the image acquisition device reaches each second sampling point, so as to realize the acquisition of each target image and record the target object pose state and the image acquisition device pose state corresponding to each target image.

[0132] Furthermore, based on the first and second sampling points corresponding to each target image, the pose state of the target object and the pose state of the image acquisition device corresponding to each target image are obtained, thereby directly calculating the spatial information corresponding to each target image without the need for subsequent estimation through the target image. This not only increases the accuracy of the stored spatial information, but also makes the acquisition of spatial information convenient and fast, effectively improving the acquisition efficiency of target object acquisition data.

[0133] To facilitate understanding of the process of acquiring spatial information in this embodiment by those skilled in the art, a specific implementation method will be used for detailed explanation below.

[0134] Specifically, for any target image, the target object's pose state is obtained by reading the values ​​of each joint of the supporting mechanism and the motion mechanism based on its corresponding first and second sampling points. and the pose status of the image acquisition device .

[0135] The calculated pose relationship is as follows: .

[0136] Based on the acquired pose relationship, the parameters of the image acquisition device are extracted to obtain the pose feature vector, which is then stored together with the corresponding target image as target object acquisition data.

[0137] Among them, based on the collected pose relationship Extract the rotation matrix and convert it into a unit quaternion form as a four-dimensional vector; based on the acquired pose relationships... The translation vector is extracted and normalized to obtain a three-dimensional vector; the horizontal focal length and reference focal length are extracted based on the image acquisition device and logarithmically mapped to obtain a one-dimensional vector, i.e. , It is a one-dimensional vector. Horizontal focal length Using the reference focal length, an eight-dimensional vector is obtained by concatenating four-dimensional, three-dimensional, and one-dimensional vectors as the pose feature vector corresponding to the target image.

[0138] Furthermore, the acquisition time is added as a new vector element to the eight-dimensional vector to obtain a nine-dimensional vector as the pose feature vector.

[0139] Based on this, the pose feature vectors corresponding to each target image are obtained and stored together with each target image as target object acquisition data, so that synthetic images can be generated by artificial intelligence models based on the target acquisition data in the future.

[0140] It should be noted that by associating pose feature vectors with target images for storage, spatial information can be used for retrieval of target images, thereby achieving faster target image retrieval efficiency. For example, in the image synthesis process of artificial intelligence models, target images within the vicinity of the synthesized image's viewpoint can be retrieved, thus quickly and accurately acquiring target images and improving the accuracy of the generated synthesized images.

[0141] Furthermore, based on the results of the synthesized image, the quality of the acquired target data is fed back so that when the quality of the synthesized image is poor, reshooting is performed to acquire a new target image and regenerate the synthesized image, thereby improving the quality of the synthesized image.

[0142] Specifically, based on the data collected from the target object, the data is synthesized using an artificial intelligence model to obtain a synthesized image. Based on the synthesized image, it is determined whether to reshoot. If so, the reshoot angles are obtained.

[0143] Based on each re-shooting perspective, new target images are acquired, and the spatial information corresponding to each new target image is obtained to store and update the target object acquisition data. That is, the new target images and their corresponding spatial information are stored in the target object acquisition data for updating. It should be noted that the specific execution method for acquiring and storing new target images and their corresponding spatial information based on each re-shooting perspective is described in steps S200 to S300 above, and will not be repeated here.

[0144] In some alternative implementations, such as Figure 5 As shown, the specific execution methods for determining whether to perform reshoots and obtaining the perspectives for each reshoot include:

[0145] S401 performs quality verification based on the synthesized image. If the verification passes, no reshoot is required.

[0146] For example, the quality verification includes geometric consistency verification. Based on the three-dimensional model data of the target object and the viewpoint corresponding to the synthesized image, a projected image is obtained. Several key points are extracted from both the projected image and the synthesized image, and geometric distortion analysis is performed based on these key points to reflect the geometric consistency of the synthesized image.

[0147] And / or, quality verification includes light and shadow direction consistency verification. Based on the composite image, the main light direction of the background area and the main light direction of the target object in the composite image are obtained to analyze whether the main light directions of the background area and the target object are consistent, reflecting the light and shadow direction consistency of the composite image.

[0148] And / or, quality verification includes edge fusion quality verification, which, based on the synthesized image, obtains the boundary between the background region and the target object in the synthesized image, and calculates the amplitude of color and brightness jumps on the boundary to reflect the boundary fusion of the synthesized image.

[0149] It should be noted that when the synthesized image passes the quality check, the synthesized image quality is high and no reshoot is required.

[0150] S402, otherwise, based on the synthesized image, obtain the synthesized pose vector, and combine it with the target object acquisition data to calculate the minimum matching distance. If the minimum matching distance does not exceed the first matching threshold, then no reshoot is performed.

[0151] The synthesized pose vector is used to characterize the spatial information corresponding to the synthesized image. Specifically, the observation viewpoint, position, focal length, etc. of the target object in the synthesized image are obtained, and the rotation matrix, translation vector, and focal length of the synthesized image are estimated to obtain an eight-dimensional vector, which serves as the synthesized pose vector. For the specific method of obtaining the synthesized pose vector, please refer to the method of obtaining the pose feature vectors corresponding to the target images mentioned above, which will not be repeated here.

[0152] Based on the spatial information corresponding to each target image in the target object acquisition data, the matching distance between each target image and the synthesized image is calculated, and the minimum value is selected as the minimum matching distance. Each matching distance is the vector distance between the pose feature vector corresponding to each target image and the synthesized pose vector, i.e., the sum of the squares of the differences between corresponding elements, or the weighted sum of the squares of the differences between corresponding elements, to reflect the degree of difference in spatial information between each target image and the synthesized image.

[0153] The first matching threshold is used to characterize the maximum difference in spatial information between the target image and the synthesized image required for the artificial intelligence model to generate a synthesized image that passes quality verification. Specifically, those skilled in the art should set the corresponding first matching threshold based on actual needs, and this embodiment does not limit this. For example, the first matching threshold is 0.05.

[0154] When the minimum matching distance does not exceed the first matching threshold, the difference in spatial information between each target image and the synthesized image still meets the requirements for generating the synthesized image. At this time, the quality verification of the synthesized image fails. This is not due to the large difference between the spatial information of each target image and the synthesized image, but may be due to poor image quality caused by lighting and shadow processing or edge fusion. Therefore, no reshoot is performed.

[0155] When the minimum matching distance exceeds the first matching threshold, the difference in spatial information between each target image and the synthesized image is too great, and reshooting is required.

[0156] S403, otherwise, based on the synthesized pose vector, generate a preset number of reshoot angles.

[0157] Specifically, based on the synthesized pose vector, the viewpoint corresponding to the synthesized image is obtained, and based on the preset field of view, the viewpoint range of each supplementary viewpoint is obtained, and based on the preset number of supplementary viewpoints, each supplementary viewpoint is obtained. Optionally, the angular interval between each supplementary viewpoint is the same, that is, each supplementary viewpoint is evenly distributed.

[0158] Those skilled in the art can set specific field of view and number of reshoot angles according to actual needs. For example, the preset field of view is centered on the angle corresponding to the synthesized image, with a horizontal field of view of 30° and a vertical field of view of 30°, and the preset number is 5.

[0159] Based on this, images of each target and its corresponding spatial information are reacquired from each reshot perspective to update the target acquisition data.

[0160] Furthermore, a second matching threshold is set, which characterizes the maximum difference in spatial information between the target image and the synthesized image required for the AI ​​model to generate a synthetic image. Before the AI ​​model generates the synthetic image, the viewpoint of the desired synthetic image is obtained, thereby obtaining the synthetic pose vector. Based on the synthetic pose vector, the minimum matching distance between each target image and the synthetic image is obtained. If the minimum matching distance exceeds the second matching threshold, reshooting is required. That is, based on the synthetic pose vector, a preset number of reshoot viewpoints are generated to obtain new target images and their corresponding spatial information, updating the target object acquisition data. Those skilled in the art should set the corresponding second matching threshold based on actual needs; this embodiment does not limit this setting. For example, the second matching threshold is 0.15.

[0161] For the specific execution method and principle of obtaining the synthetic pose vector of the synthetic image and obtaining the minimum matching distance based on the synthetic pose vector, please refer to step S402; for the specific execution method and principle of obtaining each reshooting viewpoint based on the synthetic pose vector for reshooting and updating the target object acquisition data, please refer to step S403. This embodiment will not elaborate further here.

[0162] Furthermore, the spatial information corresponding to each target image used to assist in generating the synthesized image is recorded each time the artificial intelligence model generates a synthesized image, in order to obtain the viewpoint query frequency distribution of the target object, which is used to characterize the query frequency of the target object from each viewpoint. If the query frequency exceeds the query frequency threshold within any viewpoint range, and the number of target images within that viewpoint range is lower than the coverage density threshold, then supplementary shooting is performed based on that viewpoint range. The size of each viewpoint range is a preset value; those skilled in the art can set the specific range size according to actual needs, and this embodiment does not impose specific limitations. Similarly, the query frequency threshold and coverage density threshold for each viewpoint are also preset values; those skilled in the art should set them specifically based on the size of each viewpoint range and in conjunction with actual needs, and this embodiment does not impose specific limitations.

[0163] It should be noted that for the specific execution method and principle of reshooting based on the field of view, please refer to step S403. This embodiment does not impose specific limitations here.

[0164] Based on this, after acquiring the target object data, this embodiment analyzes whether each target image meets the requirements for generating a high-quality composite image by generating a composite image based on an artificial intelligence model. In this way, the target object data is updated by taking additional photos. The process of generating composite images based on an artificial intelligence model realizes the feedback update of the target object data, which is conducive to obtaining higher-quality composite images and improving the image acquisition effect.

[0165] Furthermore, this embodiment also provides an image acquisition device for acquiring images of a target object based on the aforementioned image acquisition method.

[0166] The image acquisition device includes an image acquisition device, a motion mechanism, and a support mechanism.

[0167] The supporting mechanism is used to fix the target object; the motion mechanism is used to fix and connect the image acquisition device and drive the image acquisition device to move; the acquisition pose relationship is obtained based on the image acquisition device, the motion mechanism, the target object and the supporting mechanism;

[0168] The image acquisition device is used to acquire images of a target object; acquire three-dimensional model data of the target object, and acquire a first acquisition trajectory and a second acquisition trajectory based on the three-dimensional model data; acquire images of each target object based on the first acquisition trajectory and the second acquisition trajectory, and acquire spatial information corresponding to each target image based on the first sampling point and the second sampling point corresponding to each target image, combined with the acquisition pose relationship; and store each target image and the corresponding spatial information as target object acquisition data.

[0169] Based on the same technical concept, the image acquisition method provided in the embodiments of the present invention can be implemented on the terminal side or the server side.

[0170] like Figure 6 The diagram illustrates an optional hardware structure of a terminal according to an embodiment of the present invention. The terminal 50 can be a mobile phone, computer device, tablet device, personal digital processing device, factory back-end processing device, etc. The terminal 50 includes at least one processor 51, a memory 52, at least one network interface 54, and a user interface 53. The various components in the device are coupled together via a bus system 55. It is understood that the bus system 55 is used to realize communication between these components. In addition to a data bus, the bus system 55 also includes a power bus, a control bus, and a status signal bus.

[0171] The user interface 53 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0172] It is understood that memory 52 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memory characterized in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable categories of memory.

[0173] In this embodiment of the invention, the memory 52 is used to store various types of data to support the operation of the terminal. Examples of this data include: any executable program for operation on the terminal 50, such as the operating system 521 and application programs 522; the operating system 521 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 522 may contain various applications, such as media players, browsers, etc., for implementing various application services. The image acquisition method provided in this embodiment of the invention can be included in the application program 522.

[0174] The methods disclosed in the above embodiments of the present invention can be applied to processor 51, or implemented by processor 51. Processor 51 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 51 or by instructions in the form of software. The processor mentioned above may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 51 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present invention. Processor 51 may be a microprocessor or any conventional processor, etc. The steps of the image acquisition method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0175] In an exemplary embodiment, terminal 50 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to execute the aforementioned method.

[0176] This invention also provides a computer-readable storage medium storing a computer program that, when invoked by a processor, implements the image acquisition method provided by this invention.

[0177] Computer-readable storage media can be tangible devices capable of holding and storing instructions used by an instruction execution device. Computer-readable storage media can be, for example, (but not limited to) electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, and mechanical encoding devices.

[0178] The computer-readable program represented herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network, to an external computer or external storage device. A network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards these instructions to the computer-readable storage medium in the respective computing / processing device.

[0179] In summary, this application obtains the first sampling point of the target object and the second sampling point of the image acquisition device during image acquisition, thereby acquiring the corresponding spatial information during image acquisition. This eliminates the need for subsequent estimation based on the target image, improving not only the efficiency of image acquisition but also the accuracy of the acquired spatial information, which is beneficial for improving the image quality of the synthetic image generated by the artificial intelligence model. Furthermore, the process of generating the synthetic image based on the artificial intelligence model provides feedback on the target object acquisition data, updating the target object acquisition data through re-shooting. This allows the artificial intelligence model to generate higher-quality synthetic images based on the target object acquisition data, improving the overall image synthesis effect.

[0180] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0181] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. An image acquisition method, wherein an image acquisition device is fixedly connected to a motion mechanism, a target object is fixed to a supporting mechanism, and an image of the target object is acquired by the image acquisition device, characterized in that, include: Based on the image acquisition device, the motion mechanism, the target object, and the supporting mechanism, the acquisition pose relationship is obtained; The three-dimensional model data of the target object is acquired, and a first acquisition trajectory and a second acquisition trajectory are acquired based on the three-dimensional model data; wherein, the first acquisition trajectory connects each first sampling point, the second acquisition trajectory connects each second sampling point, and the first sampling point and the second sampling point correspond one-to-one; Based on the first acquisition trajectory and the second acquisition trajectory, each target image is acquired, and based on the first sampling point and the second sampling point corresponding to each target image, combined with the acquisition pose relationship, the spatial information corresponding to each target image is obtained; each target image and the corresponding spatial information are stored as target object acquisition data.

2. The image acquisition method according to claim 1, characterized in that, The method for acquiring the pose relationship includes: Based on the structure of the motion mechanism, the pose state of the image acquisition device is obtained; Based on the structure of the bearing mechanism, the position and orientation state of the target object is obtained; Based on the pose state of the target object and the pose state of the image acquisition device, the acquisition pose relationship is obtained.

3. The image acquisition method according to claim 2, characterized in that, The process of obtaining the pose state of the image acquisition device based on the structure of the motion mechanism includes: Obtain the motion state of the motion mechanism; Hand-eye calibration is performed based on the image acquisition device and the motion mechanism, and the calibration results are obtained; Based on the motion state of the motion mechanism and the calibration results, the pose state of the image acquisition device is obtained.

4. The image acquisition method according to claim 1, characterized in that, The process of obtaining the first acquisition trajectory and the second acquisition trajectory based on the three-dimensional model data includes: Surface complexity analysis is performed based on the three-dimensional model data to obtain the distribution of viewpoint importance. Construct an observation sphere, and obtain each first sampling point and each second sampling point by combining the distribution of the importance of the viewpoint; Based on each of the first sampling points, a first acquisition trajectory is generated, and based on each of the second sampling points, a second acquisition trajectory is generated.

5. The image acquisition method according to claim 1, characterized in that, The construction methods of the three-dimensional model data include: Based on preset shooting angles, acquire each real-shot image of the target object; Based on the captured images, features of the target object are extracted, and three-dimensional reconstruction is performed based on the features to generate the three-dimensional model data.

6. The image acquisition method according to claim 1, characterized in that, Based on the data collected from the target object, a composite image is obtained, and based on the composite image, it is determined whether to reshoot. If so, each reshoot angle is obtained. Based on each of the aforementioned reshooting perspectives, new target images are acquired, and the spatial information corresponding to each of the new target images is obtained, in order to store and update the target object acquisition data.

7. The image acquisition method according to claim 6, characterized in that, The step of determining whether to perform a reshoot based on the synthesized image, and if so, obtaining each reshoot angle, includes: The quality of the synthesized image is checked; if the check passes, no reshoot is required. Otherwise, based on the synthesized image, a synthesized pose vector is obtained, and combined with the target object acquisition data, the minimum matching distance is calculated. If the minimum matching distance does not exceed the first matching threshold, no reshoot is performed. Otherwise, based on the synthesized pose vector, a preset number of reshoot angles are generated.

8. An image acquisition device for acquiring images of a target object, characterized in that, Includes image acquisition equipment, motion mechanism, and load-bearing mechanism; The supporting mechanism is used to fix the target object; the motion mechanism is used to fix the image acquisition device and drive the image acquisition device to move; the acquisition pose relationship is obtained based on the image acquisition device, the motion mechanism, the target object and the supporting mechanism; The image acquisition device is used to acquire images of the target object; obtain three-dimensional model data of the target object, and obtain a first acquisition trajectory and a second acquisition trajectory based on the three-dimensional model data; Based on the first acquisition trajectory and the second acquisition trajectory, each target image is acquired, and based on the first sampling point and the second sampling point corresponding to each target image, combined with the acquisition pose relationship, the spatial information corresponding to each target image is obtained; each target image and the corresponding spatial information are stored as target object acquisition data; The first acquisition trajectory connects each of the first sampling points, and the second acquisition trajectory connects each of the second sampling points, with each of the first sampling points and the second sampling points corresponding one-to-one.

9. A terminal, characterized in that, include: A processor and a memory, wherein the memory and the processor are communicatively connected; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to cause the terminal to perform the image acquisition method as described in any one of claims 1 to 7.

10. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the image acquisition method as described in any one of claims 1 to 7.

Images