Digital coding points and a dynamic pose measurement method based on digital coding points

Abstract

The present application discloses a digital coding point and a dynamic pose measurement method based on the digital coding point. The method includes: capturing a target image of a target object, where each target image at each moment includes digital coding points, the second surface of the digital coding point is in contact with the surface of the target object, the first surface has a gray background, a digital pattern centered thereon, and a concentric circle pattern, and the concentric circle pattern includes white rings and a black solid circle; extracting the contour of the digital coding point from the image, recognizing the concentric circle pattern of the digital coding point based on the contour, and determining the central two-dimensional coordinates of the digital coding point according to the concentric circle pattern; determining the central three-dimensional coordinates by combining the central two-dimensional coordinates and the device parameters of the capturing device; and determining the dynamic pose information of the target object according to the central three-dimensional coordinates of multiple images. This solution extracts the contour of the concentric circle pattern from the digital coding point in the image to achieve accurate recognition of the center of the digital coding point, and better obtains the dynamic pose information based on the recognized center.

Description

Technical Field

[0001] The present application relates to the technical fields of computer vision and optical detection, and particularly relates to a digital coding point and a dynamic pose measurement method based on the digital coding point. Background Art

[0002] In the process of reconstructing the coordinates of a certain point in space by using a binocular stereo vision system, and in the process of identifying and tracking the dynamic pose of a moving object, it is a common method to make coding points at the position to be measured. At present, coding points have been designed into many types. For the coding points in three-dimensional reconstruction, two main tasks need to be completed: reading its identity information and accurately positioning the center position of its image. By simultaneously taking pictures of the image containing the coding points with two cameras and combining photogrammetry algorithms, the three-dimensional coordinates of the coding points can be reconstructed. By continuously obtaining the three-dimensional coordinates of the coding points and analyzing the spatial changes at multiple time points, the dynamic pose of the object can be tracked and measured in real time.

[0003] The existing technologies usually have the following problems: the decoding speed of the coding points is slow, the coding capacity is small, the recognition accuracy rate is low, the correction cost of incorrect decoding is high, and the application in complex dynamic scenes is limited. Summary of the Invention

[0004] For this reason, the present application discloses the following technical solutions:

[0005] The first aspect of the present application provides a digital coding point. The digital coding point is a square thin sheet, and the square thin sheet includes a first surface with a gray background and a second surface opposite to the first surface. A digital pattern is centrally arranged on the first surface;

[0006] The first surface further includes a concentric circle pattern. The center of the concentric circle pattern coincides with the center of the first surface. The concentric circle pattern includes a white ring and a black solid circle surrounded by the white ring;

[0007] The digital coding point is used to be pasted on the surface of a target object to assist in tracking the pose information of the target object at different times;

[0008] Wherein, when the digital coding point is pasted on the surface of the target object, the second surface is in contact with the surface of the target object;

[0009] The digital patterns of the respective digital coding points pasted on the surface of the target object are different from each other.

[0010] Optionally, the digital coding point further includes a white positioning strip located below the digital pattern.

[0011] Optionally, the digital coding point further includes longitudinal dividing strips respectively located on both sides of the concentric circle pattern;

[0012] The longitudinal dividing strip is used to divide the digital patterns located on both sides of the concentric circle pattern.

[0013] Optionally, in the digital coding point where the digital pattern is 13, the part of pattern 3 located on the left side of the longitudinal dividing strip is the same color as the background of the first side.

[0014] The second aspect of the present application provides a dynamic pose measurement method based on digital coding points, including:

[0015] Taking target images of the target object at different times, each of the target images at each time includes at least one digital coding point, the second side of the digital coding point is in contact with the surface of the target object, the first side of the digital coding point has a gray background and includes a digital pattern and a concentric circle pattern arranged in the center, and the concentric circle pattern includes a white ring and a black solid circle surrounded by the white ring;

[0016] Performing image distortion correction and Gaussian filtering on the target image to obtain a processed target image;

[0017] Extracting the contour of each digital coding point in the processed target image according to the edge detection algorithm, and identifying the concentric circle pattern of the digital coding point within the contour to determine the central two-dimensional coordinates of the digital coding point according to the concentric circle pattern;

[0018] Based on binocular stereo vision technology, combining the central two-dimensional coordinates and the device parameters of the shooting device that shoots the target image to determine the central three-dimensional coordinates of the digital coding point;

[0019] Determining the dynamic pose information of the target object according to the central three-dimensional coordinates of each digital coding point in the target images at different times.

[0020] Optionally, the taking of the target images of the target object at different times includes:

[0021] Taking target images of the target object at different times through a pre-calibrated binocular camera;

[0022] The device parameters include the internal parameters and external parameters of the binocular camera obtained by calibrating the binocular camera.

[0023] Optionally, the extracting of the contour of each digital coding point in the processed target image according to the edge detection algorithm includes:

[0024] Process the processed target image according to a convolutional neural network to identify digital coding points included in the processed target image; the convolutional neural network includes an anchor fine-tuning module and a target detection module connected to each other, and the target detection module includes a multi-level feature fusion block and a detector cascade block;

[0025] Extract the contour of the identified digital coding points according to an edge detection algorithm.

[0026] Optionally, the extracting the contour of the identified digital coding points according to an edge detection algorithm includes:

[0027] Calculate the target image based on a Sobel operator to obtain the gray gradient distribution of the target image;

[0028] Based on a non-maximum suppression algorithm, determine local maximum values of pixel point gray gradients in the gray gradient distribution to obtain a preliminary target contour composed of the local maximum values;

[0029] Screen the preliminary target contour based on a preset target gray level to obtain a secondary target contour;

[0030] Based on a double-threshold algorithm, use the gray gradient distribution to perform connection processing on the secondary target contour to obtain the contour of the digital coding points.

[0031] Optionally, the identifying digital coding points included in the processed target image is implemented by training the convolutional neural network according to a sample image set, the sample image set is composed of multiple frames of images including the digital coding points, and the method for obtaining the sample image set includes:

[0032] Shoot multiple frames of original images including the digital coding points;

[0033] Perform image transformation, projection transformation, and pose adjustment processing on the original images to obtain multiple frames of processed sample images, and the processed sample images and the original images constitute the sample image set.

[0034] Optionally, the identifying concentric circle patterns of the digital coding points within the contour to determine the central two-dimensional coordinates of the digital coding points includes:

[0035] Screen a central positioning circle from within the contour according to an image morphology method;

[0036] Detect the gray centroid of the central positioning circle to obtain the central two-dimensional coordinates of the digital coding points.

[0037] In this solution, a concentric circle pattern composed of a white ring and a black solid circle enclosed within the white ring is centered within the square digital coding points, enabling the binocular stereo vision system to more accurately extract the contour of the concentric circle pattern in the captured image, thereby achieving more precise identification of the center of the digital coding points. Furthermore, the center of the digital coding points can more quickly and accurately obtain the dynamic pose information of the corresponding object. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on the provided drawings.

[0039] Figure 1 is a schematic diagram of a digital coding point provided by an embodiment of the present application;

[0040] Figure 2 is a flowchart of a dynamic pose measurement method based on digital coding points provided by an embodiment of the present application;

[0041] Figure 3 is a schematic diagram of a model structure provided by an embodiment of the present application;

[0042] Figure 4 is a schematic diagram of a system structure for implementing the dynamic pose measurement method provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of them. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0044] The first aspect of the present application provides a digital coding point. Please refer to Figure 1 , which is a schematic diagram of the digital coding point provided in this embodiment. Figure 1 Twenty digital coding points are shown, corresponding to the numbers 1 to 20 in sequence.

[0045] It can be seen that a digital coding point is a square thin sheet, and the square thin sheet includes a first side with a gray background (i.e., Figure 1 the side with the number shown) and a second side opposite to the first side (i.e., Figure 1 the back of the surface shown).

[0046] On the first side of each digital coding point, a digital pattern is centered, for example Figure 1 the digital patterns in the first row from left to right in

[0047] are 1 to 5 in sequence.

[0048] The digital coding point is used to be pasted on the surface of the target object to assist in tracking the pose information of the target object at different times. That is to say, when it is necessary to detect the pose information of any target object, one or more digital coding points can be pasted on the surface of the target object, and then the target object can be photographed, and the pose information of the target object can be obtained by identifying these digital coding points in the photographed image.

[0049] Wherein, when the digital coding point is pasted on the surface of the target object, the second side is in contact with the surface of the target object. Therefore, in the photographed image, the pattern on the first side of the digital coding point as shown in Figure 1 can be seen.

[0050] The digital patterns of the respective digital coding points pasted on the surface of the target object are different from each other.

[0051] In this solution, by centering a concentric circle pattern composed of a white ring and a black solid circle surrounded by the white ring in the square digital coding point, the binocular stereo vision system can more accurately extract the contour of the concentric circle pattern in the photographed image, thereby realizing more accurate center recognition of the digital coding point, and further obtaining the dynamic pose information of the corresponding object more quickly and accurately for the center of the digital coding point.

[0052] Optionally, as shown in Figure 1 the digital coding point further includes a white positioning strip located below the digital pattern. It is used to distinguish the front and back of the number, avoid confusion of numbers such as 6 and 9, 18 and 81 during rotation, and ensure the uniqueness of the digital coding when rotation occurs.

[0053] Optionally, the digital coding point further includes longitudinal dividing strips respectively located on both sides of the concentric circle pattern;

[0054] The longitudinal dividing strips are used to divide the digital patterns located on both sides of the concentric circle pattern.

[0055] The color of the longitudinal dividing strips can be the same as the background color of the digital coding point. For example, both the longitudinal dividing strips and the background of the digital coding point are gray. The longitudinal dividing strips can longitudinally penetrate the left and right two numbers, making the center positioning circle in the digital coding point the only complete circle in the entire coding point.

[0056] Optionally, see Figure 1 The digital coding point with the digital pattern 13, that is, the third digital coding point in the third row (counting from top to bottom) and the third one (counting from left to right). It can be seen that in the digital coding points with the digital pattern 13, the part of pattern 3 located on the left side of the vertical dividing strip is the same color as the background of the first side.

[0057] The purpose of doing this is to distinguish the digital patterns 18 and 13 with partially similar digital coding points, and avoid confusion between the 13th digital coding point and the 18th digital coding point during recognition.

[0058] Based on the above digital coding points, this embodiment provides a dynamic pose measurement method based on digital coding points. Please see Figure 2 , which is the flowchart of this method. This method may include the following steps.

[0059] S201, Capture target images of the target object at different times. Each target image contains at least one digital coding point. The second side of the digital coding point is in contact with the surface of the target object. The first side of the digital coding point has a gray background and includes a centrally arranged digital pattern and a concentric circle pattern. The concentric circle pattern includes a white ring and a black solid circle enclosed within the white ring.

[0060] S202, Perform image distortion correction and Gaussian filtering on the target image to obtain the processed target image.

[0061] S203, Extract the contour of each digital coding point in the processed target image according to the edge detection algorithm, and identify the concentric circle pattern of the digital coding point within the contour to determine the central two-dimensional coordinates of the digital coding point according to the concentric circle pattern.

[0062] S204, Based on the binocular stereo vision technology, combine the central two-dimensional coordinates and the device parameters of the device for capturing the target image to determine the central three-dimensional coordinates of the digital coding point.

[0063] S205, Determine the dynamic pose information of the target object according to the central three-dimensional coordinates of each digital coding point in the target images at different times.

[0064] The above technical solutions have the following advantages or beneficial effects: An Arabic numeral coding point with intuitive coding features, rich coding types, and large coding capacity is designed, a sample set based on actual photographed coding point images, and training and rapid decoding of digital coding point types are realized based on a convolutional neural network. An improved Sobel edge detection algorithm is provided, which improves the speed and robustness of coding point center positioning. Through binocular stereo vision technology, combined with the central two-dimensional coordinates of the identified digital coding points, the position and pose of an object at different time points are tracked, and real-time tracking and precise measurement of the object's dynamic pose are realized.

[0065] Among them, step S201 may include:

[0066] Taking target images of a target object at different times by a pre-calibrated binocular camera;

[0067] The device parameters include the internal parameters and external parameters of the binocular camera obtained by calibrating the binocular camera.

[0068] Specifically, before taking images with the binocular camera, the binocular camera can be calibrated based on any calibration method (such as Zhang Zhengyou calibration method) to obtain the internal parameters and external parameters of the camera; then the binocular camera is used to synchronously take pictures of a target object with several digital coding points pre-pasted on it to obtain a surface image of the target object containing the digital coding points.

[0069] In step S202, first, according to the calibration results of the binocular camera (i.e., the internal parameters and external parameters of the binocular camera), distortion correction of each frame of the captured image can be performed, and a two-dimensional Gaussian function is selected as a smoothing filter to perform image filtering on each frame of the image to obtain several processed target images.

[0070] In a preferred embodiment, distortion correction can be performed using the undistortImage function in MATLAB according to the camera calibration results. Since the pixel coordinates calculated by the distortion correction algorithm are non-integer values, interpolation is required to generate pixel coordinate grayscales at integer positions. In this embodiment, the bilinear interpolation method is selected, and the size of the corrected image is the same as that of the original image. A two-dimensional Gaussian function is selected as a smoothing filter for image filtering. The rotational symmetry of the two-dimensional Gaussian function ensures that there will be no problem of different smoothing degrees in each direction during the filtering process, ensuring that there will be no direction deviation in subsequent edge detection. The method of performing image filtering based on the two-dimensional Gaussian function can refer to relevant existing technologies and will not be elaborated.

[0071] Step S203 may include:

[0072] A1. Process each frame of the processed target image according to a convolutional neural network to identify the digital coding points contained in the processed target image. The convolutional neural network includes an interconnected anchor fine-tuning module and an object detection module. The object detection module includes a multi-level feature fusion block and a detector cascade block.

[0073] A2. Extract the contour of the identified digital coding points according to an edge detection algorithm.

[0074] In step A1, any convolutional neural network (CNN) can be used to process each frame of the processed target image to identify the digital coding points contained in the image of each frame.

[0075] The convolutional neural network can be trained according to a set of sample images. The set of sample images consists of multiple frames of images containing digital coding points. The method for obtaining the set of sample images includes: shooting multiple frames of original images containing digital coding points; performing image transformation, projection transformation, and pose adjustment processing on the original images to obtain multiple frames of processed sample images. The processed sample images and the original images form the set of sample images.

[0076] By performing image transformation, projection transformation, and pose adjustment on the original images, the images of the coding points in different poses and on different surfaces can be simulated, ensuring the diversity of training data and the high accuracy of recognition.

[0077] In a preferred embodiment, the convolutional neural network for identifying digital coding points can be, for example, Figure 3 as shown, a digital coding point recognition network constructed based on the RefineDet framework. This network is based on a feed-forward convolutional neural network and mainly consists of two interconnected anchor fine-tuning modules (Anchor Refining Module, ARM) and an object detection module (Objection Detection Module, ODM). Among them, ODM includes multiple multi-level feature fusion blocks (MultipleFeatures Fusion Block, MFFB) and multiple detector cascade blocks (Detector Cascade Block, DCB), thereby obtaining the classification and regression of objects, which helps to improve the accuracy of object position regression and the accuracy of anchor boxes when predicting multiple class labels, and improves the recognition accuracy of small-size and complex-background coding points. The other structures and specific working principles of ARM and ODM can refer to relevant existing technologies and will not be elaborated.

[0078] Take Figure 3 as an example, ODM can include four MFFB modules and four DCB modules.

[0079] The ARM can receive the image features output by the convolutional neural network as input, and the convolutional neural network can process the input image to obtain the image features of the input image.

[0080] As Figure 3 shown, the convolutional neural network used in this embodiment can be the VGG16 neural network developed by the Visual Geometry Group (VGG). In this embodiment, the VGG16 neural network can include 6 convolutional layers and two fully connected layers. The 6 convolutional layers are sequentially denoted as Conv1 to Conv6, and the two fully connected layers are respectively denoted as Fc_6 and Fc_7. For the connection relationship between the above convolutional layers and fully connected layers, refer to Figure 3 , which will not be elaborated.

[0081] The above convolutional layers and fully connected layers can all include activation functions. Taking Figure 3 as an example, the activation function can be the Rectified Linear Unit (ReLU) activation function. Alternatively, the activation function can also be other forms of activation functions in the prior art.

[0082] For each processed target image, the processed target image can be input into the Figure 3 shown network. In the output result, it can include the type of each encoded point in the processed target image (that is, the number of the encoded point, such as 12, 6, etc.) and the position of its recognition box (that is, the position where the digital encoded point is located), pictures, and files.

[0083] Step A2 can include:

[0084] 1. Calculate the target image based on the Sobel operator to obtain the gray gradient distribution of the target image;

[0085] 2. Based on the non-maximum suppression algorithm, determine the local maximum of the pixel gray gradient in the gray gradient distribution to obtain a preliminary target contour composed of local maxima;

[0086] 3. Screen the preliminary target contour based on a preset target gray level to obtain a secondary target contour;

[0087] 4. Based on the double-threshold algorithm, use the gray gradient distribution to connect the secondary target contour to obtain the contour of the digital encoded point.

[0088] In step 1, the Sobel operator can be used to calculate each pixel point of the processed target image to obtain the gray gradient of each pixel point. The set of gray gradients of all pixel points of the processed target image is the gray gradient distribution of the processed target image.

[0089] In step 2, non-maximum suppression can be adopted to traverse the gray gradients of each pixel point in the gray gradient distribution to determine the local maximum of the gray gradient. Then, the gray gradient as the local maximum can be retained, and the gray gradients of other pixel points within a certain range around the pixel point corresponding to the local maximum can be set to zero. Thus, the target contour composed of the pixel points corresponding to the local maximum can be initially obtained, that is, the initial target contour of step 2 is obtained.

[0090] In step 3, based on the preset target gray value, and according to the particularity of the center gray value of the digital coding point in the image, each obtained initial target contour can be further screened to eliminate non-target contours, and the remaining contours are used as the secondary target contours of step 3. Specifically, in step 3, the pixel points with gray values greater than the preset target gray value can be deleted from the initial target contour, and the contour composed of the remaining pixel points is used as the secondary target contour.

[0091] In step 4, based on the double-threshold algorithm, the secondary target contour can be further judged and connected according to the gray gradient, so that the secondary target contour is closer to the true contour of the digital coding point, and finally the true contour of the digital coding point is obtained, denoted as the target contour. The specific principle of the double-threshold algorithm can refer to the relevant existing technologies and will not be elaborated here.

[0092] Optionally, the method for identifying the concentric circle pattern of the digital coding point within the contour to determine the central two-dimensional coordinates of the digital coding point may include:

[0093] Screen out the central positioning circle from within the contour according to the image morphology method;

[0094] Detect the gray centroid of the central positioning circle to obtain the central two-dimensional coordinates of the digital coding point.

[0095] In this embodiment, first, for the target contour of each digital coding point identified in the previous step, within the closed area enclosed by the target contour, all circular areas can be screened out from this closed area of the processed target image through image morphology operations. Since Figure 1 due to the special design of the digital coding point shown, the circular area within a digital coding point is obviously the area where the black solid circle (i.e., the central positioning circle) in the concentric circle pattern of the digital coding point is located. Therefore, the circular area screened out within the target contour of a digital coding point can be used as the area where the central positioning circle of this digital coding point is located.

[0096] After determining the areas where the central positioning circles of all digital coding points in the image are located, the center point coordinate positioning can be completed by calculating the gray centroids of all central positioning circles in the processed target image, and thus the two-dimensional coordinates of the center of each digital coding point in the image coordinate system can be correctly obtained.

[0097] Among them, the specific methods of image morphological operations and calculating the gray center of gravity can be referred to the relevant existing technologies and will not be elaborated here.

[0098] In step S205, the dynamic pose information of the target object may include the moving speed and moving direction of the target object, the rotational speed and rotational direction when the pose changes, and other information.

[0099] In a preferred embodiment, when performing dynamic pose measurement, it is necessary to calculate the three-dimensional coordinates of the digital coding points based on the binocular stereo vision technology and in combination with the central two-dimensional coordinates of the identified digital coding points. Through the three-dimensional point cloud data of consecutive frames, the position and pose of the object at different time points can be tracked to achieve the dynamic pose measurement of the object.

[0100] Specifically, in step S205, for two frames of images captured by the binocular camera at the same moment, the three-dimensional coordinates (i.e., the central three-dimensional coordinates) of the center of the digital coding points pasted on the surface of the target object when these two frames of images are captured can be calculated through the binocular stereo vision technology and in combination with the central two-dimensional coordinates of the digital coding points in these two frames of images identified in step S204.

[0101] Thus, the central three-dimensional coordinates of the digital coding points on the surface of the target object at multiple consecutive moments can be obtained, and these three-dimensional coordinates constitute the three-dimensional point cloud data of consecutive frames.

[0102] On this basis, according to the three-dimensional coordinates of the center of each digital coding point in the three-dimensional point cloud data at different moments, the position and pose of the object at different time points can be tracked to achieve the real-time tracking and accurate measurement of the dynamic pose of the object, and the dynamic pose information of the target object can be obtained.

[0103] For example, the three-dimensional coordinates of the No. 2 digital coding point (i.e., the digital coding point with the digital pattern 2) one second ago are (x1, y1, z1), and the three-dimensional coordinates of the No. 2 digital coding point one second later are (x2, y1, z1), where x2 is greater than x1 and the difference between them is 1 meter. Then it can be determined that the moving direction of the target object is moving horizontally to the right, and the moving speed is (x2 - x1) meters per second.

[0104] For another example, according to the three-dimensional coordinates of the No. 2 digital coding point one second ago and one second later, and the three-dimensional coordinates of the No. 3 digital coding point one second ago and one second later, it is found that both of these two digital coding points have rotated 90 degrees clockwise relative to the center of the target object within one second. Then it can be determined that the rotational direction of the target object is clockwise rotation, and the rotational speed is 90 degrees per second.

[0105] The method provided in this embodiment can be performed by Figure 4The system shown is executed, specifically by a computer of the system. It can be seen that the system includes a binocular camera, a computer, a target object with digital coding points pasted on it, and a synchronous triggering device.

[0106] Among them, the computer can, through the synchronous triggering device, control the binocular camera to simultaneously capture the target object, thereby obtaining a number of pairs of target images. Each pair of target images corresponds to a shooting moment, and each pair of target images includes two frames of target images simultaneously captured by the binocular camera at the corresponding moment.

[0107] Then, the computer can obtain a number of pairs of target images captured by the binocular camera at multiple moments, and based on the target images among them, Figure 2 perform processing according to the method of the embodiment shown, so as to obtain the dynamic pose information of the target object at any moment.

[0108] It should be noted that the various embodiments in this specification are all described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other.

[0109] For the convenience of description, when describing the above system or device, it is divided into various modules or units according to functions for separate description. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0110] From the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application, in essence, or the part that makes a contribution to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0111] Finally, it should also be noted that in this text, relational terms such as first, second, third, and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the said element.

[0112] The above are only the preferred embodiments of the present application. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A digital coding point, characterized in that, The digital coding point is a square thin sheet, the square thin sheet includes a first side with a gray background and a second side opposite to the first side, a digital pattern is centered on the first side, and a white positioning strip is provided below the digital pattern; The first side further includes a concentric circle pattern, the center of the concentric circle pattern coincides with the center of the first side, and the concentric circle pattern includes a white ring and a black solid circle surrounded by the white ring; The digital coding point is used to be pasted on the surface of the target object to assist in tracking the pose information of the target object at different times through the center of the digital coding point; Wherein, when the digital coding point is pasted on the surface of the target object, the second side is in contact with the surface of the target object; The digital patterns of the respective digital coding points pasted on the surface of the target object are different from each other.

2. The digital coding point according to claim 1, wherein The digital coding point further includes longitudinal dividing strips respectively located on both sides of the concentric circle pattern; The longitudinal dividing strips are used to divide the digital patterns located on both sides of the concentric circle pattern.

3. The digital coding point according to claim 2, wherein In the digital coding point with the digital pattern of 13, the part of pattern 3 located on the left side of the longitudinal dividing strip is the same color as the background of the first side.

4. A dynamic pose measurement method based on digital coding points, characterized in that, For performing dynamic pose measurement according to the digital coding point according to any one of claims 1 to 3, the method includes: Taking target images of the target object at different times, each target image at each time includes at least one digital coding point, the second side of the digital coding point is in contact with the surface of the target object, the first side of the digital coding point has a gray background and includes a centered digital pattern and a concentric circle pattern, the concentric circle pattern includes a white ring and a black solid circle surrounded by the white ring, the digital patterns of the respective digital coding points pasted on the surface of the target object are different from each other, and a white positioning strip is provided below the digital pattern; Performing image distortion correction and Gaussian filtering processing on the target image to obtain a processed target image; Extracting the contour of each digital coding point in the processed target image according to an edge detection algorithm, screening out the black solid circle from within the contour according to an image morphology method, and detecting the gray center of gravity of the black solid circle to obtain the center two-dimensional coordinates of the digital coding point; Based on binocular stereo vision technology, combining the center two-dimensional coordinates and the device parameters of the imaging device that captures the target image to determine the center three-dimensional coordinates of the digital coding point; Determining the dynamic pose information of the target object according to the center three-dimensional coordinates of the respective digital coding points in the target images at different times; The digital coding points included in the processed target image are recognized by a convolutional neural network, the convolutional neural network includes an anchor fine-tuning module and a target detection module connected to each other, the target detection module includes a plurality of multi-level feature fusion blocks and a plurality of detector cascade blocks, and the plurality of detector cascade blocks are used to output multi-class classification and regression information of the image.

5. The method according to claim 4, characterized in that The taking of the target images of the target object at different times includes: Taking target images of the target object at different times by a pre-calibrated binocular camera; The device parameters include the internal parameters and external parameters of the binocular camera calibrated for the binocular camera.

6. The method according to claim 4, wherein The extracting the contour of each digital coding point in the processed target image according to the edge detection algorithm includes: Processing the processed target image by a convolutional neural network to identify the digital coding points included in the processed target image; Extracting the contour of the identified digital coding points according to the edge detection algorithm.

7. The method according to claim 6, characterized in that, The extracting the contour of the identified digital coding points according to the edge detection algorithm includes: Calculating the target image based on the Sobel operator to obtain the gray gradient distribution of the target image; Determining the local maximum of the pixel gray gradient in the gray gradient distribution based on the non-maximum suppression algorithm to obtain a preliminary target contour composed of the local maximums; Screening the preliminary target contour based on a preset target gray level to obtain a secondary target contour; Based on the double-threshold algorithm, connecting the secondary target contour by using the gray gradient distribution to obtain the contour of the digital coding point.

8. The method according to claim 6, wherein The identifying the digital coding points included in the processed target image is realized by training the convolutional neural network according to a sample image set, and the sample image set is composed of multiple frames of images including the digital coding points. The method for obtaining the sample image set includes: Taking multiple frames of original images including the digital coding points; Performing image transformation, projective transformation and pose adjustment processing on the original images to obtain multiple frames of processed sample images, and the processed sample images and the original images form the sample image set.

Images