An automatic image annotation method, terminal, and readable storage medium

Abstract

This invention discloses an automatic image annotation method, comprising: pre-calibrating a camera array and establishing a camera coordinate system; acquiring a first two-dimensional point in a first image captured by a first camera; wherein: converting the first two-dimensional point into a first three-dimensional point using inverse perspective projection transformation; transforming the first three-dimensional point from the first camera coordinate system to the Nth camera coordinate system using rigid body transformation to obtain an Nth three-dimensional point; and converting the Nth three-dimensional point into an Nth two-dimensional point using perspective projection transformation; wherein the Nth two-dimensional point is a point in the Nth image captured by the Nth camera. This invention achieves automatic annotation by establishing a camera coordinate system and employing keypoint perspective projection transformation and rigid body transformation, reducing the workload of manually adjusting and changing the pose and position of the target object, and eliminating the problems of missed or incorrect annotations, without requiring manual verification.

Description

Technical Field

[0001] This invention relates to the field of terminal applications, and more particularly to an automatic image annotation method, a terminal, and a readable storage medium. Background Technology

[0002] Object detection image acquisition and annotation is a crucial problem in computer vision. Traditional object detection image acquisition devices typically use a single camera to capture images of the target in different scenes and simulate real-world object imaging by changing the target's position and pose. Object detection image annotation is primarily done manually.

[0003] Manual annotation relies entirely on annotators using data annotation tools such as LabelImg, LabelMe, or sprite annotation assistants to label the location and category of targets in object detection images. Annotating the bounding box and class label of a target involves several steps. First, the annotator clicks on one corner of an imaginary rectangle tightly surrounding the target (e.g., the top left corner), then clicks and drags the mouse to the opposite corner to create the labeled rectangle. The annotator may need to further adjust the rectangle by clicking and dragging its edges until the box fits snugly onto the target. Finally, the annotator selects the class label for the labeled target and clicks the "Submit" button. These three steps (clicking the first corner, dragging to the second corner, and labeling the class) constitute three distinct tasks, which the annotator must repeat for each target, consuming a significant amount of time.

[0004] Therefore, improving the efficiency of target detection image collection and the speed of annotation has become an urgent problem to be solved. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an automatic image annotation method, terminal and readable storage medium to address the shortcomings of the prior art, thereby solving the problems of slow target detection image collection speed and low efficiency.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] In a first aspect, the present invention provides an automatic image annotation method, the automatic image annotation method comprising the following steps:

[0008] The camera array is pre-calibrated and a camera coordinate system is established; wherein, the camera coordinate system includes N camera coordinate systems;

[0009] Obtain a first two-dimensional point in a first image captured by a first camera; wherein the first two-dimensional point is a manually labeled point in the first image;

[0010] The first two-dimensional point is transformed into a first three-dimensional point using inverse perspective projection transformation; wherein, the first three-dimensional point is a point in the first camera coordinate system corresponding to the first two-dimensional point;

[0011] The first three-dimensional point is transformed from the first camera coordinate system to the Nth camera coordinate system using rigid body transformation to obtain the Nth three-dimensional point.

[0012] The Nth three-dimensional point is converted into an Nth two-dimensional point using perspective projection transformation; wherein the Nth two-dimensional point is a point in the Nth image captured by the Nth camera.

[0013] In one implementation, the pre-calibration of the camera array and the establishment of the camera coordinate system include:

[0014] A calibration board is used to jointly calibrate each camera in the camera array and establish a corresponding camera coordinate system for each camera.

[0015] In one implementation, the step of acquiring the first two-dimensional point in the first image captured by the first camera includes:

[0016] Obtain the intrinsic parameters of each camera, the extrinsic parameters between them, and the distortion coefficients;

[0017] Obtain the distance of each camera from the base; wherein, the base is the support for the target object;

[0018] A target image set is acquired, which consists of one image containing the target object taken by each camera; wherein, the image taken by the first camera is called the first image, and the image taken by the Nth camera is called the Nth image.

[0019] In one implementation, acquiring the first two-dimensional point in the first image captured by the first camera includes:

[0020] The first image captured by the first camera is labeled as the first two-dimensional point.

[0021] In one implementation, converting the first two-dimensional point into a first three-dimensional point using inverse perspective projection transformation includes:

[0022] The inverse perspective projection transformation formula is:

[0023]

[0024] Where (u, v) are the coordinates of the first two-dimensional point, (X, Y, Z) are the coordinates of the first three-dimensional point, and Z is the coordinate of the first three-dimensional point. c1 K1 represents the distance from the first camera to the base, and K1 is the intrinsic parameter of the first camera.

[0025] The coordinates of the first two-dimensional point are obtained by inputting the inverse perspective projection transformation formula.

[0026] In one implementation, the step of transforming the first 3D point from the first camera coordinate system to the Nth camera coordinate system using rigid body transformation to obtain the Nth 3D point includes:

[0027] The rigid body transformation formula is:

[0028]

[0029] Wherein, (X', Y', Z') are the coordinates of the Nth three-dimensional point, and (R, T) are the extrinsic parameters of the Nth camera relative to the first camera;

[0030] The coordinates of the first three-dimensional point are input into the rigid body transformation formula to obtain the coordinates of the Nth three-dimensional point.

[0031] In one implementation, converting the Nth three-dimensional point into the Nth two-dimensional point using perspective projection transformation includes:

[0032] The perspective projection transformation formula is:

[0033]

[0034] Among them, K n Z is the intrinsic parameter of the Nth camera. cn Let (u', v') be the distance from the Nth camera to the base, and (u', v') be the Nth two-dimensional point.

[0035] The coordinates of the Nth three-dimensional point are input into the perspective projection transformation formula to obtain the coordinates of the Nth two-dimensional point.

[0036] In one implementation, the step of inputting the coordinates of the Nth three-dimensional point into the perspective projection transformation formula to obtain the coordinates of the Nth two-dimensional point further includes:

[0037] The coordinates of the Nth two-dimensional point are automatically labeled as the automatically labeled points in the Nth image;

[0038] Determine if N-1 automatically labeled points have been obtained; if so, end the process.

[0039] If not, then reacquire the automatically labeled points until N-1 automatically labeled points are obtained.

[0040] In a second aspect, the present invention provides a terminal, characterized in that it includes: a processor and a memory, wherein the memory stores an image automatic annotation method program, and the image automatic annotation method program, when executed by the processor, is used to implement any one of the image automatic annotation methods described above.

[0041] Thirdly, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores an image automatic annotation method program, which, when executed by a processor, is used to implement any one of the image automatic annotation methods described above.

[0042] The present invention, by employing the above technical solution, has the following effects:

[0043] This invention discloses an automatic image annotation method, comprising: pre-calibrating a camera array and establishing a camera coordinate system; acquiring a first two-dimensional point in a first image captured by a first camera; wherein: converting the first two-dimensional point into a first three-dimensional point using inverse perspective projection transformation; transforming the first three-dimensional point from the first camera coordinate system to the Nth camera coordinate system using rigid body transformation to obtain an Nth three-dimensional point; and converting the Nth three-dimensional point into an Nth two-dimensional point using perspective projection transformation; wherein the Nth two-dimensional point is a point in the Nth image captured by the Nth camera. This invention achieves automatic annotation by establishing a camera coordinate system and employing keypoint perspective projection transformation and rigid body transformation, reducing the workload of manually adjusting and changing the pose and position of the target object, and eliminating the problems of missed or incorrect annotations, without requiring manual verification. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0045] Figure 1 This is a flowchart of an automatic image annotation method in one implementation of the present invention.

[0046] Figure 2 This is a functional schematic diagram of the terminal in one implementation of the present invention.

[0047] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0048] This invention relates to a method for target detection image acquisition and automatic annotation, belonging to the field of image processing technology. One objective of this invention is to provide a simple acquisition device for industrial target detection image collection. It uses a planar camera array for synchronous image acquisition, allowing for the acquisition of target images at different positions and orientations with a single shot, resulting in fast acquisition speed and high efficiency. The planar camera array consists of several evenly distributed cameras. The target object is placed on a base, and multiple cameras simultaneously capture images. Compared to traditional single-camera acquisition systems, this device can acquire images of the target object in more orientations and positions, reducing the workload of manually adjusting and changing the target object's orientation and position. Furthermore, the acquisition device is simple, easy to operate, and has a wide range of applications.

[0049] Another objective of this invention is to propose an automatic target bounding box annotation method based on perspective projection transformation. Compared with the current machine annotation method based on deep learning, it does not have the problems of missing or incorrect annotations, so no manual review is required. Moreover, it has high robustness and can also annotate solid-color, highly reflective, or transparent objects or objects with occlusion. It also does not require a large number of sensors, but only a certain number of ordinary cameras to form a camera array.

[0050] The automatic target bounding box annotation method includes the following steps: calibrating the intrinsic parameters K and extrinsic parameters (R, T) of the cameras in the planar camera array using a calibration plate, and acquiring multi-pose and multi-position images of the target object using the image acquisition device; manually annotating the target image captured by the reference camera in the planar camera array; and projecting the key points of the target object's bounding box onto the corresponding positions of the other camera's image using the intrinsic parameters K and extrinsic parameters (R, T) between the cameras, thereby transforming the automatic annotation of the bounding box into perspective projection transformation and rigid body transformation of the key points. This method can handle some unavoidable occlusion situations in the acquired target images, has good robustness, and requires no manual verification. Compared with traditional manual annotation, efficiency is significantly improved.

[0051] like Figure 1 As shown, this embodiment of the invention provides an automatic image annotation method, which includes the following steps:

[0052] Step S100: Pre-calibrate the camera array and establish a camera coordinate system; wherein the camera coordinate system includes N camera coordinate systems.

[0053] In this embodiment, the target detection image acquisition device mainly includes a planar camera array and a base. The planar camera array consists of several cameras, which are evenly distributed. The target object is placed on the base, and multiple cameras simultaneously capture images. Compared with traditional single-camera acquisition systems, this device can acquire images of the target object in more poses and positions, reducing the workload of manually adjusting and changing the pose and position of the target object.

[0054] In one implementation of this embodiment, step S100 includes the following steps:

[0055] Step S101: Use a calibration board to perform joint calibration on each camera in the camera array and establish a corresponding camera coordinate system for each camera.

[0056] In this embodiment, a calibration map is acquired. The black and white checkerboard used for camera array calibration is 400×400mm in size and has 25×24 corner points. The camera array (N cameras) is used to simultaneously capture images of the checkerboard in different positions and postures, capturing approximately 20 sets of images.

[0057] In this embodiment, the camera array is calibrated using the MATLAB calibration toolbox or the Zhang Zhengyou calibration method via OpenCV. The camera at the center is used as the reference camera, with its coordinate system serving as the world coordinate system. This world coordinate system is then used as the first camera coordinate system, and the intrinsic parameters K and extrinsic parameters (R, T) for the other N-1 cameras are obtained. The intrinsic parameters include the camera's principal point (cx, cy), focal length (fx, fy), and coordinate axis tilt parameter γ, as shown in formula (1). The extrinsic parameters include the translation vector T and the rotation matrix R. The rotation matrix R describes the direction of the world coordinate system's coordinate axes relative to the camera coordinate axes, and the translation vector T describes the position of the origin in the camera coordinate system. The distortion coefficients include radial distortion coefficients (k1, k2, k3) and tangential distortion coefficients (p1, p2). The camera coordinate systems for the other N-1 cameras are determined based on these intrinsic and extrinsic parameters.

[0058]

[0059] like Figure 1 As shown, in one implementation of this invention, an automatic image annotation method further includes the following steps:

[0060] Step S200: Obtain the first two-dimensional point in the first image captured by the first camera.

[0061] In this embodiment, the images captured by the reference camera are manually labeled. The camera at the center of the planar camera array is selected as the first camera, and its camera coordinate system will also serve as the world coordinate system. The bounding boxes and category labels of the objects in the images captured by this camera are manually labeled.

[0062] In this embodiment, the perspective projection transformation of key points mainly uses the perspective projection model of the camera, namely the pinhole imaging model. Its main purpose is to transform a three-dimensional object into a two-dimensional graphic representation, so that any point P in space can be approximately represented on the image plane. The matrix used in the transformation process is called the projection matrix.

[0063] In one implementation of this embodiment, step S200 includes the following steps:

[0064] Step S201: Obtain the intrinsic parameters of each camera, the extrinsic parameters between them, and the distortion coefficients;

[0065] Step S202: Obtain the distance of each camera from the base; wherein, the base is the support for the target object;

[0066] Step S203: Obtain a target image set, which consists of one image containing the target object taken by each camera; wherein, the image taken by the first camera is called the first image, and the image taken by the Nth camera is called the Nth image.

[0067] In this embodiment, the distance Z from each camera to the base is obtained.

[0068] In this embodiment, target object image acquisition involves placing the calibration object on a base and simultaneously capturing images of the calibration object at different positions and orientations using a camera array. The camera array comprises N cameras that simultaneously capture images, each photographing a different target object position and orientation. This reduces the workload of manually placing the target object and allows for the acquisition of more target object images within the same timeframe. Compared to traditional single-camera target detection dataset acquisition, this acquisition device increases the diversity of the target detection dataset and better simulates its imaging in real-world environments.

[0069] In this embodiment, the camera array calibration is performed using the MATLAB calibration toolbox (camera calibrator) or OpenCV with the Zhang Zhengyou calibration method. The camera located at the center is used as the reference camera, and its coordinate system is taken as the world coordinate system. This reference camera coordinate system is called the first camera coordinate system. The intrinsic parameters K and extrinsic parameters (R, T) of the other N-1 cameras are then obtained.

[0070] like Figure 1As shown, in one implementation of this invention, the automatic image annotation method further includes the following steps:

[0071] Step S300: The first two-dimensional point is converted into a first three-dimensional point using inverse perspective projection transformation; wherein, the first three-dimensional point is a point in the first camera coordinate system corresponding to the first two-dimensional point.

[0072] In one implementation of this embodiment, step S300 includes the following steps:

[0073] Step S301, the inverse perspective projection transformation formula is:

[0074]

[0075] Where (u, v) are the coordinates of the first two-dimensional point, (X, Y, Z) are the coordinates of the first three-dimensional point, and Z is the coordinate of the first three-dimensional point. C1 K1 represents the distance from the first camera to the base, and K1 is the intrinsic parameter of the first camera.

[0076] Step S302: Input the coordinates of the first two-dimensional point into the inverse perspective projection transformation formula to obtain the coordinates of the first three-dimensional point.

[0077] In this embodiment, automatic annotation based on perspective projection transformation and rigid body transformation is performed. The four corner points of the target object's bounding box, which were captured by the reference camera and manually annotated, are used as key points. These key points are then transformed onto the image corresponding to another camera using perspective projection transformation and rigid body transformation to complete the automatic annotation. The detailed steps are as follows:

[0078] Inverse perspective projection transformation, given the intrinsic parameter matrix K1 of the reference camera c and the set of key points in the image plane of the reference camera.

[0079]

[0080] in, Represents the four key points of the bounding box: top, bottom, left, and right. This represents the presence of n bounding boxes for objects in an image.

[0081] The key point set Kpts is transformed into the first camera coordinate system using formula (3).

[0082]

[0083] This leads to the 3D point set in the first camera coordinate system:

[0084] like Figure 1 As shown, in one implementation of this invention, the automatic image annotation method further includes the following steps:

[0085] Step S400: Use rigid body transformation to transform the first three-dimensional point from the first camera coordinate system to the Nth camera coordinate system to obtain the Nth three-dimensional point.

[0086] In one implementation of this embodiment, step S400 includes the following steps:

[0087] Step S401, the rigid body transformation formula is:

[0088]

[0089] Wherein, (X', Y', Z') are the coordinates of the Nth three-dimensional point, and (R, T) are the extrinsic parameters of the Nth camera relative to the first camera;

[0090] Step S402: Input the coordinates of the first three-dimensional point into the rigid body transformation formula to obtain the coordinates of the Nth three-dimensional point.

[0091] In this embodiment, a rigid body transformation is performed. Given the translation vector T and rotation matrix R of the Nth camera relative to the first camera, the 3D point set Kpts3 in the coordinate system of the first camera is transformed into the camera coordinate system of the camera to be labeled using formula (4), i.e., rigid body transformation, to obtain the 3D point set in the coordinate system of the Nth camera.

[0092]

[0093] like Figure 1 As shown, in one implementation of this invention, the automatic image annotation method further includes the following steps:

[0094] Step S500: The Nth three-dimensional point is converted into an Nth two-dimensional point using perspective projection transformation; wherein the Nth two-dimensional point is a point in the Nth image captured by the Nth camera.

[0095] In one implementation of this embodiment, step S500 includes the following steps:

[0096] Step S501, the perspective projection transformation formula is:

[0097]

[0098] Among them, K n Z is the intrinsic parameter of the Nth camera. cn Let (u', v') be the distance from the Nth camera to the base, and (u', v') be the Nth two-dimensional point.

[0099] Step S502: Input the coordinates of the Nth three-dimensional point into the perspective projection transformation formula to obtain the coordinates of the Nth two-dimensional point.

[0100] In this embodiment, the key points marked on the image plane corresponding to another camera are finally obtained after another perspective projection transformation. The category label of the keypoint is the same as that of the keypoint corresponding to the benchmark camera.

[0101]

[0102] For all images captured by the first camera, the inverse perspective projection transformation-rigid body transformation-perspective projection transformation process described above is sequentially applied to the corresponding images captured by the other cameras. For each transformed image, the keypoint set is... This means that the graph contains n target bounding boxes.

[0103] In one implementation of this embodiment, step S502 further includes the following steps:

[0104] Step S5021: Automatically label the coordinates of the Nth two-dimensional point as the automatically labeled point in the Nth image;

[0105] Step S5022: Determine whether N-1 automatic annotation points have been obtained. If so, the process ends.

[0106] Step S5023: If not, then reacquire the automatic annotation points until N-1 automatic annotation points are obtained.

[0107] In this embodiment, for each target bounding box, there are 4 key points. Draw straight lines parallel to the pixel coordinate axes along the uv and uv axes of the pixel coordinate system respectively. The resulting pixel coordinates of each vertex of the largest bounding rectangle of the target object in the pixel coordinate system are used as the bounding box coordinates for automatic annotation. The category labels are the same as the original keypoint category labels.

[0108] Repeat the above process: transform the keypoints of N-1 cameras in the camera array (excluding the reference camera), take the largest bounding rectangle, attach category labels, and complete the automatic annotation of all images.

[0109] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 2 As shown.

[0110] The terminal includes: a processor, a memory, an interface, a display screen, and a communication module connected via a system bus; wherein, the processor provides computing and control capabilities; the memory includes a computer-readable storage medium and internal memory; the computer-readable storage medium stores an operating system and computer programs; the internal memory provides an environment for the operation of the operating system and computer programs in the computer-readable storage medium; the interface is used to connect to external terminal devices, such as mobile terminals and computers; the display screen is used to display relevant image automatic annotation method information; and the communication module is used to communicate with a cloud server or mobile terminal.

[0111] This computer program is executed by the processor to implement an automatic image annotation method.

[0112] It will be understood by those skilled in the art that Figure 2 The schematic diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0113] In one embodiment, a terminal is provided, comprising: a processor and a memory, the memory storing an image automatic annotation program, which, when executed by the processor, is used to implement the image automatic annotation method as described above.

[0114] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores an image automatic annotation program, which, when executed by a processor, is used to implement the image automatic annotation method as described above.

[0115] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory.

[0116] In summary, this invention discloses an automatic image annotation method, comprising: pre-calibrating a camera array and establishing a camera coordinate system; acquiring a first two-dimensional point in a first image captured by a first camera; wherein: converting the first two-dimensional point into a first three-dimensional point using inverse perspective projection transformation; transforming the first three-dimensional point from the first camera coordinate system to the Nth camera coordinate system using rigid body transformation to obtain an Nth three-dimensional point; and converting the Nth three-dimensional point into an Nth two-dimensional point using perspective projection transformation; wherein the Nth two-dimensional point is a point in the Nth image captured by the Nth camera. This invention achieves automatic annotation by establishing a camera coordinate system and employing keypoint perspective projection transformation and rigid body transformation, reducing the workload of manually adjusting and changing the pose and position of the target object, and eliminating the problems of missed or incorrect annotations, thus eliminating the need for manual verification.

[0117] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. An image automatic labeling method, characterized by, The image automatic labeling method comprises the following steps: The camera array is calibrated in advance and a camera coordinate system is established; wherein the camera coordinate system comprises N camera coordinate systems; A first two-dimensional point in a first image captured by a first camera is obtained; wherein the first two-dimensional point is a manually labeled point in the first image; The first two-dimensional point is converted into a first three-dimensional point by using inverse perspective projection transformation; wherein the first three-dimensional point is a point in the first camera coordinate system corresponding to the first two-dimensional point; The first three-dimensional point is converted from the first camera coordinate system to an Nth camera coordinate system by using rigid body transformation to obtain an Nth three-dimensional point; The Nth three-dimensional point is converted into an Nth two-dimensional point by using perspective projection transformation; wherein the Nth two-dimensional point is a point in an Nth image captured by an Nth camera; The first two-dimensional point is converted into a first three-dimensional point by using inverse perspective projection transformation, which comprises: The inverse perspective projection transformation formula is: ； wherein (u, v) is the first two-dimensional point coordinate, (X, Y, Z) is the first three-dimensional point, Z c1 is the distance from the first camera to the base, and K1 is the intrinsic parameter of the first camera. The first two-dimensional point coordinates are input into the inverse perspective projection transformation formula to obtain the first three-dimensional point coordinates; The first three-dimensional point is converted from the first camera coordinate system to an Nth camera coordinate system by using rigid body transformation to obtain an Nth three-dimensional point, which comprises: The rigid body transformation formula is: ； wherein (X ’ , Y ’ , Z ’ ) is the Nth three-dimensional point coordinate, and (R, T) is the extrinsic parameter of the Nth camera relative to the first camera; The first three-dimensional point coordinates are input into the rigid body transformation formula to obtain the Nth three-dimensional point coordinates; The Nth three-dimensional point is converted into an Nth two-dimensional point by using perspective projection transformation, which comprises: The perspective projection transformation formula is: wherein, is the intrinsic parameter of the Nth camera, Z cn is the distance from the base to the Nth camera, (u ’ , v ’ ) is the Nth two-dimensional point; The Nth three-dimensional point coordinates are input into the perspective projection transformation formula to obtain the Nth two-dimensional point coordinates; The Nth three-dimensional point coordinates are input into the perspective projection transformation formula to obtain the Nth two-dimensional point coordinates, which further comprises: The Nth two-dimensional point coordinates are automatically labeled as automatic labeling points in the Nth image; It is determined whether N-1 automatic labeling points are obtained, if yes, the process is ended; If not, the automatic labeling points are reacquired until N-1 automatic labeling points are obtained.

2. The image automatic labeling method of claim 1, wherein, The camera array is calibrated in advance and a camera coordinate system is established, which comprises: A calibration board is used to jointly calibrate each camera in the camera array and establish a corresponding camera coordinate system for each camera.

3. The image automatic labeling method of claim 1, wherein, Before the first two-dimensional point in the first image captured by the first camera is obtained, the following steps are included: The intrinsic parameters, mutual extrinsic parameters and distortion coefficients of each camera are obtained; The distance of each camera from the base is obtained; wherein the base is a support for the target object; A target object image set is obtained, which is composed of one image containing the target object captured by each camera; wherein the image captured by the first camera is referred to as the first image, and the image captured by the Nth camera is referred to as the Nth image.

4. The image automatic labeling method of claim 1, wherein, The first two-dimensional point in the first image captured by the first camera is obtained, which comprises: The first image captured by the first camera is labeled as the first two-dimensional point.

5. A terminal, characterized by comprising: It comprises: A processor and a memory, wherein the memory stores an image automatic labeling method program, and the image automatic labeling method program is used to implement the image automatic labeling method of any one of claims 1-4 when executed by the processor.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium stores an image automatic labeling method program, and the image automatic labeling program is used for implementing the image automatic labeling method in any one of claims 1-4 when executed by the processor.

Images