WO2019113912A1

WO2019113912A1 - Structured light-based three-dimensional image reconstruction method and device, and storage medium

Info

Publication number: WO2019113912A1
Application number: PCT/CN2017/116321
Authority: WO
Inventors: 宋展; 曾海
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2017-12-15
Filing date: 2017-12-15
Publication date: 2019-06-20

Abstract

A structured light-based three-dimensional image reconstruction method and device, and a storage medium, applicable to the field of image processing. The method comprises: extracting main encoding feature points of an input object encoding image; constructing a topological network of the main encoding feature points, and extracting all encoding element images comprised in the object encoding image according to the topological network; positioning the initial position of a graphical feature point in each of the encoding element images, and calculating an auxiliary encoding feature point of the encoding element image; identifying all encoding element images by means of a deep learning network according to the main encoding feature points and the auxiliary encoding feature points; matching encoding information corresponding to each of the identified encoding element images with encoding information corresponding to a pre-stored encoding pattern according to a preset polar line encoding policy to implement decoding; and performing three-dimensional image reconstruction on the object according to pre-acquired calibration parameters of a three-dimensional image reconstruction system and decoding information obtained by the matching to obtain a three-dimensional image of the object.

Description

Three-dimensional image reconstruction method, device and storage medium based on structured light

Technical field

The invention belongs to the technical field of image processing, and in particular relates to a method, a device and a storage medium for reconstructing a three-dimensional image based on structured light.

Background technique

The structured light three-dimensional reconstruction system refers to the problem of matching in the stereo vision by projecting the optical pattern containing the specific coded information to the surface of the object, and then obtaining the correspondence by decoding, and then recovering the three-dimensional space coordinates at the projection point by the optical triangulation principle. . According to different projection patterns, structured light can be divided into point structure light, line structure light, multi-line structure light and surface structure light. Among them, point structure light method, line structure light method and multi-line structure light method are relatively mature and simple, but Each reconstruction requires multiple images to be taken, with low efficiency and a small measurement range. The surface structured light method uses a projector to project one or more coding patterns onto the surface of a three-dimensional object, and photographs the coded pattern of the surface of the three-dimensional object with a camera, and then uses the characteristics of the projected coded structured light to perform image matching, and finally utilizes the triangle. The principle of law calculates the point cloud coordinates of the object surface.

The coding method adopted by the existing structured light three-dimensional reconstruction technology can be roughly divided into a time coding method and a spatial coding method, and the time coding is performed according to the time sequence of the projected image, and then the coded image is continuously projected onto the surface of the object in chronological order. It has the advantages of high measurement accuracy and high measurement resolution, but its measurement speed is slow, so it is suitable for 3D information acquisition of static targets and scenes. The latter only needs to project a coding pattern, and the measurement speed is fast, so it is suitable for three-dimensional information acquisition of dynamic targets and scenes. Spatial coding aims to realize three-dimensional reconstruction of the surface of an object by projecting a single-encoded image. The coding information is generated by spatial coding features or different arrangement and combination thereof. The encoding and decoding processes are all completed in a single image, which has the advantage of real-time. . The existing spatial coding structure light is often encoded by color information and gray information, but the decoding effect of the existing method is easily affected by the surface color and color channel crosstalk of the object. The impact is not robust. From the current research situation in this field, spatial coding using black and white geometric features has become a trend, but there is a contradiction between the coding density of such techniques and the size of the coding window and the types of coding elements used, ie, to obtain high density. Structured optical coding patterns can only increase the type of coding elements or increase the coding window, and these two measures will significantly increase the difficulty of decoding, resulting in a reduction in decoding success rate.

Summary of the invention

It is an object of the present invention to provide a three-dimensional image reconstruction method, apparatus and storage medium based on structured light, which aims to solve the problem of low decoding success rate due to the existing three-dimensional image reconstruction method based on structured light.

In one aspect, the present invention provides a three-dimensional image reconstruction method based on structured light, the method comprising the steps of:

When receiving a three-dimensional image reconstruction request of the object input by the user, extracting a main coding feature point of the input object coded image, the coded element pattern of the object coded image has rotational symmetry and includes a preset number of auxiliary coded feature points;

Constructing a topology network of the primary coded feature points according to the primary coded feature points of the object coded image, and extracting all coded element images included in the object coded image according to the topology network;

Positioning an initial position of a graphical feature point in each of the encoded element images using a preset corner detection algorithm, and calculating an auxiliary of the encoded element image according to the initial position and a gray value of the encoded element image Coding feature points;

Identifying all of the encoded element images using a pre-trained deep learning network based on the primary encoded feature points and the auxiliary encoded feature points of all of the encoded element images;

And matching the encoded information corresponding to each of the identified coding element images with the coding information corresponding to the pre-stored coding pattern according to a preset polar coding strategy, so as to implement corresponding primary coding feature points and auxiliary coding features. Decoding of points;

The object is subjected to three-dimensional image reconstruction according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information to obtain a three-dimensional image of the object.

In another aspect, the present invention provides a three-dimensional image reconstruction apparatus based on structured light, the apparatus comprising:

a feature point extracting unit, configured to: when receiving a three-dimensional image reconstruction request of the object input by the user, extract a main coding feature point of the input object coded image, where the coded element graphic of the object coded image has rotational symmetry and includes a preset quantity Auxiliary coding feature points;

An element image extracting unit, configured to construct a topological network of the main coded feature points according to the main coded feature points of the object coded image, and extract all coded element images included in the object coded image according to the topology network;

a feature point calculation unit, configured to locate an initial position of a graphic feature point in each of the encoded element images using a preset corner detection algorithm, and calculate the grayscale value according to the initial position and the encoded element image The auxiliary coding feature point of the encoded element image;

An image recognition unit, configured to identify, according to the primary coding feature point and the auxiliary coding feature points of all the coding element images, the pre-trained deep learning network to identify all the coding element images;

a decoding unit, configured to match the encoded coding information corresponding to each of the coded element images and the coded information corresponding to the pre-stored coding pattern according to a preset polar coding strategy, to implement corresponding primary coding features Decoding of point and auxiliary coded feature points;

And an image reconstruction unit configured to perform three-dimensional image reconstruction on the object according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information to obtain a three-dimensional image of the object.

In another aspect, the present invention also provides a computer readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described above.

When performing the three-dimensional image reconstruction of the object, the main coding feature point and the auxiliary coding feature point of the object coded image are extracted, and all the coded element images are identified by using the pre-trained deep learning network, and according to the preset polar line coding strategy. Matching the encoded information corresponding to each of the identified coding element images with the coding information corresponding to the previously stored coding pattern to implement the corresponding primary coding The decoding of the feature points and the auxiliary coding feature points, and finally the three-dimensional image reconstruction of the object according to the pre-acquired three-dimensional image reconstruction system calibration parameters and the matched decoding information, thereby improving the success rate of image decoding, thereby improving the effect of three-dimensional image reconstruction. .

DRAWINGS

1 is a flowchart showing an implementation of a method for reconstructing a three-dimensional image based on structured light according to Embodiment 1 of the present invention;

2 is a schematic diagram of images of eight coding elements provided by Embodiment 1 of the present invention;

3 is a schematic diagram of main coding feature points and auxiliary coding feature points of a tessellated coding element image according to Embodiment 1 of the present invention;

4 is a coded image for projection provided by Embodiment 1 of the present invention;

5 is a schematic structural diagram of a three-dimensional image reconstruction apparatus based on structured light according to Embodiment 2 of the present invention;

FIG. 6 is a schematic structural diagram of a three-dimensional image reconstruction apparatus based on structured light according to Embodiment 3 of the present invention.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The specific implementation of the present invention is described in detail below in conjunction with specific embodiments:

Embodiment 1:

FIG. 1 is a flowchart showing an implementation process of a structured light-based three-dimensional image reconstruction method according to Embodiment 1 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown, which are described in detail as follows:

In step S101, when the object three-dimensional image reconstruction request input by the user is received, the main coded feature point of the input object coded image is extracted.

In an embodiment of the invention, preferably, the object encoded image is obtained by:

(1) rotating a coded element pattern having rotational symmetry and including a preset number of feature points to obtain a plurality of code word patterns of the coded element;

(2) using a plurality of code words and combining black and white backgrounds to respectively generate corresponding coded element images;

In an embodiment of the invention, the encoded element graphic is a coding element graphic having rotational symmetry and including a preset number of feature points, for example, an "L" shaped coding element graphic, a "Δ" shaped coding element graphic. By way of example, the embodiment of the present invention is described herein by using an "L"-shaped coded element pattern. By rotating the "L"-shaped coded element pattern by 90°, 180°, and 270°, respectively, four codeword graphics can be obtained. By combining these four codeword graphics with the black and white background block, eight encoded element images can be obtained, as shown in FIG. The eight coded element images shown in Figure 2 represent eight code words (1, -1, 2, -2, 3, -3, 4, -4), respectively, where positive digital words represent codes in white background. Element image, negative digital word represents a coded element image with black background. In this way of encoding, only one kind of coding element graphic is needed, but through its rotation and background color change, eight different coding element images can be generated, which will greatly reduce the difficulty of subsequent feature detection and coding feature recognition. Significantly improve coding efficiency.

(3) according to the polar line coding strategy, using the coded element image to be encoded along the polar line direction with a preset coding window to obtain a coded image for projection of a preset resolution;

In the embodiment of the present invention, according to the polar line coding strategy, the entire encoding process is performed only in the direction of the polar line. Since one coding dimension is reduced, a smaller coding window can be obtained based on the same number of coding elements, or more A small number of coding elements obtain the same size coding window. By defining the number of polar lines, the entire coding capacity can be adjusted to meet the projection needs of different resolutions.

Preferably, the black and white checkerboard is used as the coding basic frame in the coding, and by filling the different coded element images, the checkerboard corner points are uniquely coded in the polar direction, and the checkerboard corner points are defined as the main coding feature points, due to the coding The element graphic itself has obvious geometric features, such as "L" shape or "Δ" shape. Therefore, at least three auxiliary coding feature points can be defined in each coding element graphic, as shown in Fig. 3, in which the middle The four corner points of the square checkerboard are the main coding feature points, and the three corner points of the "L" shape in the checkerboard are auxiliary coding feature points. Coding feature points by the definition of such mixed coded feature points The number can be increased by a factor of 3, which greatly increases the density of coded feature points. As an example, a coded image for projection as shown in FIG. 4 can be obtained by the above-described polar line coding strategy.

(4) Projecting the encoded image for projection onto the object, and photographing the object through the camera to obtain an object encoded image.

In an embodiment of the invention, when projecting the encoded image for projection onto an object, preferably using a diffractive optical element of a three-dimensional image reconstruction system comprising a projection device (consisting of a diffractive optical element and a laser) And a camera. Specifically, when projecting the encoded image for projection onto the object, the encoded image for projection can be obtained by laser and diffractive optical elements in the three-dimensional image reconstruction system according to the principle of laser diffraction, and projected onto the surface of the object, thereby being reconstructed by the three-dimensional image reconstruction system. The camera in the camera captures the object encoded image.

In the embodiment of the present invention, when extracting the main coded feature points of the input object coded image, preferably, the pixels of the object coded image are subjected to template convolution, and the candidate main coded feature points of the object coded image are obtained according to the template convolution result. Calculating a degree of symmetry of each candidate main coding feature point, and culling the candidate primary coding feature points whose degree of symmetry is less than a preset threshold to obtain a main coding feature point of the object coded image.

Further preferably, when the black and white checkerboard is used as the coding base frame, if the main coded feature point of the input object coded image is to be extracted, it is equivalent to detecting the corner point of the checkerboard. Therefore, first, for each pixel in each image, the "+" type template is used to calculate the volume in the neighborhood (the size of the neighborhood is generally 2/3 of the image size of one coding element in the object coded image). Product value, the convolution value of the "+" type template

Where f(x, y) represents the pixel value of the object-encoded image at point (x, y), and N represents the size of the template. Next, it is judged whether it is a maximum value in a small area centered on the pixel point, and if it is a maximum value, the corresponding pixel point is a candidate main coding feature point, that is, calculated by using a "+" type template. The convolution value is the largest in the local area. Finally, the erroneous candidate primary coding feature points are culled by rotational symmetry, and the true primary coding feature points are found from the candidate primary coding feature points.

In step S102, the main coding feature point is constructed according to the main coding feature point of the object coded image. The topology network extracts all coded element images included in the object encoded image according to the topology network.

In step S103, the initial position of the graphic feature point in each encoded element image is located using a preset corner detection algorithm, and the auxiliary coding feature point of the encoded element image is calculated according to the initial position and the gray value of the encoded element image. .

In step S104, all of the encoded element images are identified using the pre-trained deep learning network based on the primary encoded feature points and the auxiliary encoded feature points of all encoded element images.

In the embodiment of the present invention, the initially established deep learning network is trained in advance. Preferably, when training the deep learning network, a preset number of object encoded images having different colors, textures, illuminations, and scenes are first acquired. Samples, using a sample of each object encoded image to train a pre-established deep learning network to obtain a trained deep learning network. Specifically, after acquiring the object coded image sample, extracting the main coded feature point of the object coded image sample, constructing a topological network of the main coded feature point according to the main coded feature point of the object coded image sample, and extracting the object coded image sample according to the topology network All the encoded element images are included, thereby extracting a large number of coded element image samples, performing Gaussian blurring, occlusion, etc. on the coded element image samples, further expanding the sample size, and then training the deep learning network using the expanded coded element image samples. To get a trained deep learning network.

In step S105, the encoded information corresponding to each encoded element image is matched with the encoding information corresponding to the pre-stored coding pattern according to the preset polar coding strategy, so as to implement the corresponding primary coding feature point and Auxiliary coding of the decoding of feature points.

In the embodiment of the present invention, according to the polar line constraint in the polar line coding strategy (ie, the projection point of the spatial point on the two image planes, if the mapping point of the spatial point on the left image plane is known to be left on the left image plane) On the polar line, then the mapping point of the spatial point on the right image plane is on the right polar line of the right image plane, and vice versa. This constraint relationship is called the polar line constraint), and each encoded element image will be recognized. The corresponding coding information is matched with the coding information corresponding to the pre-stored coding pattern to implement decoding of the corresponding primary coding feature point and the auxiliary coding feature point. Further, the decoding can be error-corrected according to constraints such as continuity and smoothness to improve the accuracy of decoding.

In step S106, the object is subjected to three-dimensional image reconstruction according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information to obtain a three-dimensional image of the object.

In the embodiment of the present invention, the camera and the projection device in the three-dimensional image reconstruction system are calibrated by using the Zhang Zhengyou camera calibration method to obtain the calibration parameters of the camera and the projection device, and then according to the acquired camera and the calibration parameters of the projection device. The positional relationship parameter between the camera and the projection device is calculated to obtain the calibration parameters of the three-dimensional image reconstruction system. Then, the object is subjected to three-dimensional image reconstruction according to the pre-acquired three-dimensional image reconstruction system calibration parameters and the matched decoding information to obtain a three-dimensional image of the object.

In the embodiment of the present invention, a coding element graphic having a rotational symmetry and including a preset number of feature points is combined with a black and white background to generate a corresponding coding element image, and according to the polar coding strategy, the coding element image is used to pre The coding window is coded along the direction of the polar line to obtain a coded image for projection with a preset resolution, thereby implementing a smaller coding window with fewer coding element types, and greatly reducing the subsequent decoding difficulty. The success rate of decoding. Corresponding to decoding, by extracting the main coding feature points and the auxiliary coding feature points of the object coded image, all the coded element images are identified using a pre-trained deep learning network, and the recognition is obtained according to a preset polar line coding strategy. The coding information corresponding to each coding element image is matched with the coding information corresponding to the pre-stored coding pattern to implement decoding of the corresponding primary coding feature point and the auxiliary coding feature point, and finally, according to the pre-acquired three-dimensional image reconstruction system calibration parameter And the decoded information obtained by the matching performs three-dimensional image reconstruction on the object, thereby improving the success rate of image decoding, thereby improving the effect of three-dimensional image reconstruction.

Embodiment 2:

FIG. 5 shows a structure of a structured light-based three-dimensional image reconstruction apparatus according to Embodiment 2 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown, including:

The feature point extracting unit 51 is configured to: when receiving the object three-dimensional image reconstruction request input by the user, extract a main coding feature point of the input object coded image, the coded element graphic of the object coded image has rotational symmetry and includes a preset quantity Auxiliary coding feature points;

The element image extracting unit 52 is configured to construct an editor based on the main coded feature points of the object coded image. a topological network of code feature points, extracting all coded element images included in the object coded image according to the topology network;

The feature point calculation unit 53 is configured to locate an initial position of the graphic feature point in each of the encoded element images using a preset corner detection algorithm, and calculate an auxiliary of the encoded element image according to the initial position and the gray value of the encoded element image. Coding feature points;

The image recognition unit 54 is configured to identify all the encoded element images by using a pre-trained deep learning network according to the primary encoded feature points and the auxiliary encoded feature points of all the encoded element images;

The decoding unit 55 is configured to match, according to the preset polar coding strategy, the encoded information corresponding to each of the identified coding element images with the coding information corresponding to the pre-stored coding pattern, to implement the corresponding primary coding feature point. And decoding of the auxiliary coded feature points;

The image reconstruction unit 56 is configured to perform three-dimensional image reconstruction on the object according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information to obtain a three-dimensional image of the object.

In the embodiment of the present invention, each unit of the three-dimensional image reconstruction apparatus may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into one soft and hardware unit, and there is no need to limit the present. invention. For a specific implementation of each unit, reference may be made to the specific description of the corresponding steps in the first embodiment, and details are not described herein again.

Embodiment 3:

FIG. 6 shows a structure of a structured light-based three-dimensional image reconstruction apparatus according to Embodiment 3 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown, including:

The first parameter calibration unit 601 is configured to calibrate the camera and the projection device in the three-dimensional image reconstruction system by using Zhang Zhengyou camera calibration method to obtain calibration parameters of the camera and the projection device;

The second parameter calibration unit 602 is configured to calculate a positional relationship parameter between the camera and the projection device according to the acquired camera and the calibration parameter of the projection device.

An element rotation unit 603, configured to rotate a coding element having rotational symmetry and including a preset number of feature points to obtain a plurality of codeword patterns of the coding element;

The element image generating unit 604 is configured to use a plurality of code word graphics and a black and white background to combine, Generating corresponding coded element images respectively;

The image encoding unit 605 is configured to perform encoding along the polar line direction by using a coded element image in a preset coding window according to the polar line coding strategy to obtain a coded image for projection of a preset resolution;

a coded image projecting unit 606, configured to project a coded image for projection onto an object;

The feature point extracting unit 607 is configured to: when receiving the object three-dimensional image reconstruction request input by the user, extract a main coding feature point of the input object coded image, where the coded element graphic of the object coded image has rotational symmetry and includes a preset quantity Auxiliary coding feature points;

The element image extracting unit 608 is configured to construct a topological network of the main coded feature points according to the main coded feature points of the object coded image, and extract all the coded element images included in the object coded image according to the topology network;

The feature point calculation unit 609 is configured to locate an initial position of the graphic feature point in each of the encoded element images by using a preset corner detection algorithm, and calculate an auxiliary of the encoded element image according to the initial position and the gray value of the encoded element image. Coding feature points;

The image recognition unit 610 is configured to identify all the encoded element images by using a pre-trained deep learning network according to the primary encoded feature points and the auxiliary encoded feature points of all the encoded element images;

The decoding unit 611 is configured to match, according to the preset polar coding strategy, the encoded information corresponding to each of the identified coding element images with the coding information corresponding to the pre-stored coding pattern, to implement the corresponding primary coding feature point. And decoding of the auxiliary coded feature points;

The image reconstruction unit 612 is configured to perform three-dimensional image reconstruction on the object according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information to obtain a three-dimensional image of the object.

The feature point extraction unit 607 includes:

a candidate feature point acquiring unit, configured to perform template convolution on pixels of the object encoded image, and obtain candidate main coding feature points of the object encoded image according to the template convolution result;

The feature point culling unit 6072 is configured to calculate a degree of symmetry of each candidate primary coded feature point, and cull the candidate primary coded feature points whose degree of symmetry is less than a preset threshold to obtain a primary coded feature point of the object coded image.

Embodiment 4:

In an embodiment of the present invention, there is provided a computer readable storage medium storing a computer program, which when executed by a processor, implements the steps in the foregoing method embodiments, for example, FIG. Steps S101 to S106 are shown. Alternatively, the computer program, when executed by the processor, implements the functions of the various units in the various apparatus embodiments described above, such as the functions of units 51 through 56 shown in FIG.

In the embodiment of the present invention, when the computer program is executed by the processor, the steps in the foregoing method embodiment are implemented, and the pre-trained deep learning network is used to extract all the main coding feature points and the auxiliary coding feature points of the object coded image. The coded element image is identified, and the coded information corresponding to each of the identified coded element images is matched with the coded information corresponding to the previously stored coded pattern according to a preset polar line coding strategy, so as to implement the corresponding primary coded feature point. And the decoding of the auxiliary coding feature points, and finally the three-dimensional image reconstruction of the object according to the pre-acquired three-dimensional image reconstruction system calibration parameters and the matched decoding information, thereby improving the success rate of image decoding, thereby improving the effect of three-dimensional image reconstruction.

The computer readable storage medium of the embodiments of the present invention may include any entity or device capable of carrying computer program code, a recording medium such as a ROM/RAM, a magnetic disk, an optical disk, a flash memory, or the like.

The above is only the preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection of the present invention. Within the scope.

Claims

A three-dimensional image reconstruction method based on structured light, characterized in that the method comprises the following steps:

When receiving a three-dimensional image reconstruction request of the object input by the user, extracting a main coding feature point of the input object coded image, the coded element graphic in the object coded image has rotational symmetry and includes a preset number of auxiliary coded feature points;

Constructing a topology network of the primary coded feature points according to the primary coded feature points of the object coded image, and extracting all coded element images included in the object coded image according to the topology network;

Positioning an initial position of a graphical feature point in each of the encoded element images using a preset corner detection algorithm, and calculating an auxiliary of the encoded element image according to the initial position and a gray value of the encoded element image Coding feature points;

Identifying all of the encoded element images using a pre-trained deep learning network based on the primary encoded feature points and the auxiliary encoded feature points of all of the encoded element images;

And matching the encoded information corresponding to each of the identified coding element images with the coding information corresponding to the pre-stored coding pattern according to a preset polar coding strategy, so as to implement corresponding primary coding feature points and auxiliary coding features. Decoding of points;

The object is subjected to three-dimensional image reconstruction according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information to obtain a three-dimensional image of the object.
The method according to claim 1, wherein before the step of performing three-dimensional image reconstruction on the object according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information, the method further comprises:

Calibrating the camera and the projection device in the three-dimensional image reconstruction system by using Zhang Zhengyou camera calibration method to obtain calibration parameters of the camera and the projection device;

Calculating a positional relationship parameter between the camera and the projection device according to the acquired camera and the calibration parameter of the projection device.
The method of claim 1 wherein the master of the input object encoded image is extracted The steps of encoding feature points include:

And performing template convolution on the pixels of the object encoded image, and acquiring candidate main coding feature points of the object encoded image according to the template convolution result;

Calculating a degree of symmetry of each of the candidate primary coding feature points, and punctifying the candidate primary coding feature points whose degree of symmetry is less than a preset threshold to obtain a primary coding feature point of the object encoded image.
The method of claim 1 wherein before the step of extracting the main coded feature points of the input object coded image, the method further comprises:

Rotating a coded element graphic having rotational symmetry and including a preset number of feature points to obtain a plurality of code word patterns of the coded element;

Combining the plurality of codeword graphics and the black and white background to generate corresponding coding element images respectively;

According to the polar line coding strategy, the coded element image is encoded along a polar line direction with a preset coding window to obtain a coded image for projection of a preset resolution;

The encoded image for projection is projected onto the object.
The method of claim 1 wherein prior to the step of identifying all of the encoded element images using a pre-trained deep learning network, the method further comprising:

Obtaining a preset number of object coded image samples having different colors, textures, illuminations, and scenes;

A pre-established deep learning network is trained using each of the object encoded image samples to obtain the trained deep learning network.
A three-dimensional image reconstruction device based on structured light, characterized in that the device comprises:

a feature point extracting unit, configured to: when receiving a three-dimensional image reconstruction request of the object input by the user, extract a main coding feature point of the input object coded image, where the coded element graphic of the object coded image has rotational symmetry and includes a preset quantity Auxiliary coding feature points;

An element image extracting unit, configured to construct a topological network of the main coded feature points according to the main coded feature points of the object coded image, and extract all coded element images included in the object coded image according to the topology network;

a feature point calculation unit, configured to locate an initial position of a graphic feature point in each of the encoded element images using a preset corner detection algorithm, and calculate the grayscale value according to the initial position and the encoded element image The auxiliary coding feature point of the encoded element image;

An image recognition unit, configured to identify, according to the primary coding feature point and the auxiliary coding feature points of all the coding element images, the pre-trained deep learning network to identify all the coding element images;

a decoding unit, configured to match the encoded coding information corresponding to each of the coded element images and the coded information corresponding to the pre-stored coding pattern according to a preset polar coding strategy, to implement corresponding primary coding features Decoding of point and auxiliary coded feature points;

And an image reconstruction unit configured to perform three-dimensional image reconstruction on the object according to the pre-acquired three-dimensional image reconstruction system calibration parameter and the matched decoding information to obtain a three-dimensional image of the object.
The device of claim 6 wherein said device further comprises:

a first parameter calibration unit, configured to calibrate a camera and a projection device in the three-dimensional image reconstruction system by using a Zhang Zhengyou camera calibration method to obtain calibration parameters of the camera and the projection device;

And a second parameter calibration unit, configured to calculate a positional relationship parameter between the camera and the projection device according to the acquired camera and the calibration parameter of the projection device.
The device according to claim 6, wherein the feature point extracting unit comprises:

a candidate feature point acquiring unit, configured to perform template convolution on pixels of the object encoded image, and obtain candidate main coding feature points of the object encoded image according to the template convolution result;

a feature point culling unit, configured to calculate a degree of symmetry of each of the candidate primary coded feature points, and cull the candidate primary coded feature points whose degree of symmetry is less than a preset threshold to obtain a primary coding feature of the object coded image point.
The device of claim 6 wherein said device further comprises:

An element rotation unit, configured to rotate a coding element having rotational symmetry and including a preset number of feature points to obtain a plurality of codeword patterns of the coding element;

An element image generating unit, configured to use the plurality of codeword graphics and black and white backgrounds to generate corresponding coding element images;

An image encoding unit, configured to perform encoding along a polar line direction with a preset encoding window according to the polar line encoding strategy to obtain a coded image for projection of a preset resolution;

An encoded image projecting unit for projecting the encoded image for projection onto the object.
A computer readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the method of any one of claims 1 to 5.