CN111899326A - Three-dimensional reconstruction method based on GPU parallel acceleration - Google Patents
Three-dimensional reconstruction method based on GPU parallel acceleration Download PDFInfo
- Publication number
- CN111899326A CN111899326A CN202010557165.4A CN202010557165A CN111899326A CN 111899326 A CN111899326 A CN 111899326A CN 202010557165 A CN202010557165 A CN 202010557165A CN 111899326 A CN111899326 A CN 111899326A
- Authority
- CN
- China
- Prior art keywords
- gpu
- phase
- parallel
- dimensional reconstruction
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 230000001133 acceleration Effects 0.000 title claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 28
- 238000012937 correction Methods 0.000 claims abstract description 22
- 238000004364 calculation method Methods 0.000 claims abstract description 20
- 230000011218 segmentation Effects 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000001914 filtration Methods 0.000 claims description 14
- 238000010587 phase diagram Methods 0.000 claims description 8
- 230000002146 bilateral effect Effects 0.000 claims description 5
- 238000003384 imaging method Methods 0.000 claims description 3
- 230000010363 phase shift Effects 0.000 claims description 3
- 230000008030 elimination Effects 0.000 claims description 2
- 238000003379 elimination reaction Methods 0.000 claims description 2
- 238000011160 research Methods 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
-
- G06T5/70—
-
- G06T5/80—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
Abstract
The invention belongs to the field of machine vision research, and particularly relates to a three-dimensional reconstruction method based on GPU parallel acceleration. The method mainly comprises the following steps: initializing an algorithm; loading an image sequence modulated by the structural light, and copying the image sequence in the memory of the computer to a public memory at a GPU (graphics processing unit) device end in parallel; the GPU is used for carrying out background segmentation on the images shot by the left camera and the right camera in parallel, and only three-dimensional reconstruction is carried out on the segmented foreground targets, so that redundant calculation is reduced, and the calculation efficiency is improved; using a GPU to solve the phase main value of the image sequence in a parallel acceleration mode and perform phase expansion to obtain an absolute phase value; carrying out distortion correction on the phase unwrapped map; performing stereo matching in a mode of combining serial and parallel, namely performing serial processing on local areas of the image, performing parallel processing among the areas, and performing stereo matching; preprocessing a disparity map; and calculating the three-dimensional point cloud according to the disparity map. The method fully utilizes the parallel computing capability of the GPU, improves the overall computing speed of the algorithm and also expands the application scene of three-dimensional reconstruction.
Description
Technical Field
The invention belongs to the field of machine vision research, and particularly relates to a three-dimensional reconstruction method based on GPU parallel acceleration.
Background
Three-dimensional reconstruction technology is one of the important subjects of machine vision research, and means that it recovers the three-dimensional space geometry of a three-dimensional object from an image of the three-dimensional object. In general, three-dimensional reconstruction is performed by triangulation using the principle of binocular disparity of two cameras or by obtaining spatial coding using structured light, and depth information using triangulation.
Due to the inherent complexity of the three-dimensional reconstruction algorithm, the three-dimensional reconstruction algorithm has a large calculation amount, and the reconstruction speed is always a bottleneck for restricting the application of the three-dimensional reconstruction technology. Although the number of cores of the current CPU is increased, the instruction system is more prone to processing tasks and is not suitable for parallel processing of data, and the improvement of the computational performance is not obvious. The GPU is mainly used for graphic and image processing calculation, has the floating point calculation capacity ten times that of the CPU, and can be optimized according to the characteristic so as to greatly improve the speed of three-dimensional reconstruction.
Disclosure of Invention
In order to solve the problems mentioned in the background art, the invention discloses a three-dimensional reconstruction method based on GPU parallel acceleration.
In order to achieve the above purpose, the following technical solutions are provided:
a three-dimensional reconstruction method based on GPU parallel acceleration comprises the following steps
S1: initializing an algorithm;
s2: loading an image sequence modulated by the structural light, and copying the image sequence in the memory of the computer to a public memory at a GPU (graphics processing unit) device end in parallel;
s3: the GPU is used for carrying out background segmentation on the images shot by the left camera and the right camera in parallel, and only three-dimensional reconstruction is carried out on the segmented foreground targets, so that redundant calculation is reduced, and the calculation efficiency is improved;
s4: using a GPU to solve the phase main value of the image sequence in a parallel acceleration mode and perform phase expansion to obtain an absolute phase value;
s5: carrying out distortion correction on the phase unwrapped map;
s6: performing stereo matching in a mode of combining serial and parallel, namely performing serial processing on local areas of the image, performing parallel processing among the areas, and performing stereo matching;
s7: preprocessing a disparity map;
s8: calculating three-dimensional point cloud according to the phase difference;
s9: and (5) point cloud noise processing.
Further, the S1 includes the following steps:
s1.1: calculating the required storage space, and opening up the storage space to avoid the time consumption caused by opening up the storage space in the algorithm operation process;
s1.2: and (3) loading all the camera parameters and the calibration parameters (camera internal parameters, external parameters and distortion parameters) to a specified storage position.
Further, the S2 includes the following steps:
s2.1: two threads are opened up and are processed synchronously; the CPU is used for transmitting the image sequence to the GPU in parallel;
s2.2: and copying the image sequences in the memory of the computer to a public memory at the GPU equipment side in parallel.
Further, after background segmentation is performed in the step S3 by using a dynamic threshold method, a binarized mask image is generated, where a value 1 in the mask image represents a foreground and a value 0 represents a background; in subsequent processing, only foreground data is processed, and the background segmentation can effectively reduce the computation amount, so that the computation speed is improved.
Further, S4 makes full use of the advantage that GPU supports vector operation, and the step utilizes vectorization operation to improve operation efficiency, namely, one work item processes data of 16 pixels; assigning 16 adjacent pixel points in the image to a vector variable, and uniformly performing phase and unwrapping operation; the method can effectively reduce the occupation of the GPU computing unit, and the GPU hardware resource is limited, so that the method can effectively improve the computing speed by verification, and the specific computing formula is as follows:
s4.1, firstly, solving the phase through a four-step phase shift formula, calculating phase main values theta (x, y) of different frequencies, wherein I '(x, y) is image average gray scale, I' (x, y) is image gray scale modulation,
as can be seen from the equation 1,
s4.2 according to the phase main values of different frequencies, unwrapping is carried out through the multi-frequency heterodyne principle, and the absolute phase value psi (x, y) is obtained, wherein fH，fLAre each thetaH(x,y)，θL(x, y) corresponding to the frequency,
ψ(x,y)＝N(x,y)×2π+θH(x, y) equation 5
Further, S5 performs distortion correction on the unwrapped phase unwrapped map in S4. Compared with the distortion correction of the original image sequence, the method only carries out the distortion correction on the left and right phase expansion images, and a large amount of calculation is reduced, wherein (x, y) is ideal imaging coordinates, (x ', y') is distorted coordinates, and r is reduced2＝x2+y2、ki、piIs a distortion parameter. Solving the undistorted coordinate from the distorted coordinate according to the corresponding relation before and after distortion to complete distortion correction,
further, S6 specifically includes the following steps:
s6.1: creating a matching container, taking the phase value and the limit slope as container indexes, creating a two-dimensional matching container, and putting the coordinate index value of the pixel point of the right phase diagram into the matching container according to the phase value and the limit slope value of the pixel point;
s6.2: and pixel-level stereo matching and sub-pixel-level interpolation, wherein the pixel-level stereo matching is to search a coordinate index value of a right phase diagram matched with a left phase diagram in a matching container according to a phase value and a limit slope of each pixel point of the left phase diagram, calculate a disparity diagram, and perform sub-pixel interpolation in a bilinear difference mode in order to improve the precision of three-dimensional reconstruction.
Further, S7 specifically includes the following steps:
s7.1, median filtering is carried out on the disparity map, and points with wrong matching in the matching process are removed through the median filtering;
s7.2 bilateral filtering of the disparity map, wherein the bilateral filtering plays a smoothing role.
Further, S8 specifically includes the following steps:
s8, calculating the three-dimensional point cloud according to the optimized disparity map, wherein the formula is as follows:
where f denotes the focal length of the camera, T denotes the length of the baseline, x, y denotes the coordinate values in the image coordinate system, and Δ x denotes the parallax value.
Further, S9 denoises the point cloud by using a radius filtering method that can perform parallel processing, and specifies a radius R to calculate the number of points falling in the center of the sphere with a certain point as the center of the sphere, and when the number of points is greater than a given value, the point is retained, and when the number is less than the given value, the point is considered as noise.
The invention has the beneficial effects that:
1. the three-dimensional reconstruction method based on GPU parallel acceleration can effectively utilize the parallel computing capability of the GPU, is suitable for various embedded platforms, and accelerates the productization of the algorithm. Compared with a PC (personal computer), the embedded platform has the advantages of low cost, low power consumption, small occupied space, wide application range and the like; the method is not only suitable for the industrial field, but also suitable for various consumer-grade products, such as access control systems and VR equipment.
2. The invention designs a three-dimensional reconstruction method based on GPU parallel acceleration, a binocular camera capable of projecting structured light shoots a modulated grating image, and the obtained image is stored in a computer memory; and then the stored image data is transferred in parallel to the GPU equipment storage area.
3. According to the three-dimensional reconstruction method based on the GPU parallel acceleration, the image sequence is usually composed of 20 to 40 images according to different algorithm precision requirements, distortion correction is directly carried out, a large amount of time is consumed, and in order to reduce redundant calculation, distortion correction is uniformly carried out on a left phase expanded image and a right phase expanded image after phase and unwrapping.
4. According to the three-dimensional reconstruction method based on the GPU parallel acceleration, because the image pixels are mutually independent and do not have mutual dependency relationship in the phase solution and unwrapping processes, the advantage that the GPU supports vectorization operation is fully utilized at the moment, 16 adjacent pixel points in the image are assigned to a vector variable, and the phase solution and unwrapping operations are performed in a unified mode; the method can effectively reduce the occupation of the GPU computing unit, and the GPU hardware resources are limited, so that the method can effectively improve the computing speed.
5. According to the three-dimensional reconstruction method based on the GPU parallel acceleration, when an image sequence is collected, a left camera and a right camera respectively collect a strip-free image for background segmentation. After background segmentation, subsequent operations such as distortion correction, creation of a matching container, stereo matching and the like are only performed on the segmented foreground target data. Usually, the shot target foreground occupies one third to two thirds of the whole image area, and the background segmentation can effectively reduce the calculation amount, thereby improving the calculation speed.
6. According to the three-dimensional reconstruction method based on the GPU parallel acceleration, the image is inevitably distorted in the image shooting process. The distortion correction is to restore the distorted image to a normal image, and is a key step for ensuring the three-dimensional reconstruction accuracy. And calculating a coordinate mapping relation between the image before distortion and the image after distortion according to the calibrated lens distortion parameters, wherein the distorted pixels are not integers, so that the solution is required to be carried out through interpolation, and a bilinear interpolation method is adopted in the algorithm.
7. The three-dimensional reconstruction method based on GPU parallel acceleration is designed, stereo matching is carried out, the process is a process of searching corresponding points of images shot by a left camera and a right camera, the process usually adopts a loop iteration method to search an optimal value in a local area of the images, so that correlation exists among local pixels, direct pixel-level parallel processing can damage the correlation among the pixels, and the loss of three-dimensional reconstruction precision is large; the invention adopts a mode of combining serial and parallel to carry out stereo matching, namely, the local areas of the image adopt serial processing and the areas are processed in parallel.
Drawings
FIG. 1 is an overall flowchart of a three-dimensional reconstruction method based on GPU parallel acceleration;
fig. 2 is a flowchart of S6.1 of a three-dimensional reconstruction method based on GPU parallel acceleration;
fig. 3 is a flowchart of S6.2 of a three-dimensional reconstruction method based on GPU parallel acceleration.
Detailed Description
In order to make the technical solutions of the present invention more clear and definite for those skilled in the art, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings, but the embodiments of the present invention are not limited thereto.
As shown in fig. 1-3, a three-dimensional reconstruction method based on GPU parallel acceleration includes the following steps
S1: initializing an algorithm;
s1.1: calculating the required storage space, and opening up the storage space to avoid the time consumption caused by opening up the storage space in the algorithm operation process;
s1.2: and loading all the camera parameters and the calibration parameters including camera internal parameters, external parameters and distortion parameters to a specified storage position.
S2: loading an image sequence modulated by the structural light, and copying the image sequence in the memory of the computer to a public memory at a GPU (graphics processing unit) device end in parallel;
s2.1: two threads are opened up and are processed synchronously; the CPU is used for transmitting the image sequence to the GPU in parallel;
s2.2: and copying the image sequences in the memory of the computer to a public memory at the GPU equipment side in parallel.
S3: the GPU is used for carrying out background segmentation on the images shot by the left camera and the right camera in parallel, and only three-dimensional reconstruction is carried out on the segmented foreground targets, so that redundant calculation is reduced, and the calculation efficiency is improved; after background segmentation is carried out by adopting a dynamic threshold value method in the S3, a binary mask image is generated, wherein the value 1 in the mask image represents the foreground, and the value 0 represents the background; in subsequent processing, only foreground data is processed, and the background segmentation can effectively reduce the calculation amount, thereby improving the calculation speed.
S4: using a GPU to solve the phase main value of the image sequence in a parallel acceleration mode and perform phase expansion to obtain an absolute phase value; s4, the advantage that GPU supports vector operation is fully utilized, vectorization operation is utilized in the step, operation efficiency is improved, and namely 16 pixel point data are processed by one work item; assigning 16 adjacent pixel points in the image to a vector variable, and uniformly performing phase and unwrapping operation; the method can effectively reduce the occupation of the GPU computing unit, and the method can effectively improve the computing speed by verification because the GPU hardware resources are limited, and the specific computing formula is as follows:
s4.1, firstly, solving the phase through a four-step phase shift formula, calculating phase main values theta (x, y) of different frequencies, wherein I '(x, y) is image average gray scale, I' (x, y) is image gray scale modulation,
as can be seen from the equation 1,
s4.2 according to the phase main values of different frequencies, unwrapping is carried out through the multi-frequency heterodyne principle, and the absolute phase value psi (x, y) is obtained, wherein fH，fLAre each thetaH(x,y)，θL(x, y) corresponding to the frequency,
ψ(x,y)＝N(x,y)×2π+θH(x, y) equation 5
S5: carrying out distortion correction on the phase unwrapped map;
s5 performs distortion correction on the unwrapped phase unwrapped map in S4. Compared with the distortion correction of the original image sequence, the method only carries out the distortion correction on the left and right phase expansion images, and a large amount of calculation is reduced, wherein (x, y) is ideal imaging coordinates, (x ', y') is distorted coordinates, and r is reduced2＝x2+y2、ki、piIs a distortion parameter. Solving undistorted coordinates from the distorted coordinates according to the corresponding relation before and after distortion to complete distortion correction,
according to different algorithm precision requirements, an image sequence is usually composed of 20 to 40 images, distortion correction is directly carried out, a large amount of time is consumed, and in order to reduce redundant calculation, after phase solution and unwrapping, distortion correction is uniformly carried out on left and right phase unwrapped pictures
S6: performing stereo matching in a mode of combining serial and parallel, namely performing serial processing on local areas of the image, performing parallel processing among the areas, and performing stereo matching;
s6.1: creating a matching container, taking the phase value and the limit slope as container indexes, creating a two-dimensional matching container, and putting the coordinate index value of the pixel point of the right phase diagram into the matching container according to the phase value and the limit slope value of the pixel point; the specific flow is shown in fig. 2.
S6.2: pixel-level stereo matching and sub-pixel-level interpolation, wherein the pixel-level stereo matching is to search a coordinate index value of a right phase map matched with a phase value and a limit slope of each pixel point of the left phase map in a matching container, and calculate a disparity map, and the specific flow is shown in fig. 3. In order to improve the precision of three-dimensional reconstruction, sub-pixel interpolation is carried out in a bilinear difference mode.
Stereo matching, namely a process of searching corresponding points of images shot by a left camera and a right camera, wherein an optimal value is searched for in a local area of the images by adopting a loop iteration method, so that correlation exists among local pixels, and the correlation among the pixels is damaged by direct pixel-level parallel processing, so that the loss of three-dimensional reconstruction precision is large; the invention adopts a mode of combining serial and parallel to carry out stereo matching, namely, the local area of the image adopts serial processing and the areas are processed in parallel.
S7: preprocessing a disparity map;
s7.1, median filtering is carried out on the disparity map, and points with wrong matching in the matching process are removed through the median filtering;
s7.2 bilateral filtering of the disparity map, wherein the bilateral filtering plays a smoothing role.
S8: calculating three-dimensional point cloud according to the phase difference;
s8, calculating the three-dimensional point cloud according to the optimized disparity map, wherein the formula is as follows:
where f represents the focal length of the camera, T represents the length of the baseline, x, y represent coordinate values in the image coordinate system, and Δ x represents the parallax value.
S9: and (5) point cloud noise processing.
S9 noise elimination is carried out to the point cloud by adopting a radius filtering mode capable of parallel processing, a certain point is taken as a sphere center, the radius R is appointed to calculate the number of the points falling in the sphere center, when the number of the points is larger than a given value, the points are reserved, and when the number of the points is smaller than the given value, the points are considered as noise points.
The three-dimensional reconstruction method based on GPU parallel acceleration can effectively utilize the parallel computing capability of the GPU, is suitable for various embedded platforms, and accelerates the productization of the algorithm. A computer memory for shooting the modulated grating image by a binocular camera capable of projecting structured light and storing the obtained image; and then the image data of the memory is transferred in parallel to the storage area of the GPU equipment.
While the invention has been described with respect to the preferred embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A three-dimensional reconstruction method based on GPU parallel acceleration is characterized in that: comprises the following steps
S1: initializing an algorithm;
s2: loading an image sequence modulated by the structural light, and copying the image sequence in the memory of the computer to a public memory at a GPU (graphics processing unit) device end in parallel;
s3: the GPU is used for carrying out background segmentation on the images shot by the left camera and the right camera in parallel, and only three-dimensional reconstruction is carried out on the segmented foreground targets, so that redundant calculation is reduced, and the calculation efficiency is improved;
s4: using a GPU to solve the phase main value of the image sequence in a parallel acceleration mode and perform phase expansion to obtain an absolute phase value;
s5: carrying out distortion correction on the phase unwrapped map;
s6: performing stereo matching in a mode of combining serial and parallel, namely performing serial processing on local areas of the image, performing parallel processing among the areas, and performing stereo matching;
s7: preprocessing a disparity map;
s8: calculating a three-dimensional point cloud according to the disparity map;
s9: and (5) point cloud noise processing.
2. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: the S1 includes the steps of:
s1.1: calculating the required storage space, and opening up the storage space to avoid the time consumption caused by opening up the storage space in the algorithm operation process;
s1.2: and loading all the camera parameters and the calibration parameters to a specified storage position.
3. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: the S2 includes the steps of:
s2.1: two threads are opened up and are processed synchronously; the CPU is used for transmitting the image sequence to the GPU in parallel;
s2.2: and copying the image sequences in the memory of the computer to a public memory at the GPU equipment side in parallel.
4. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: after the background segmentation is carried out by adopting a dynamic threshold value method in the S3, a binary mask image is generated, wherein the value 1 in the mask image represents the foreground, and the value 0 represents the background; in subsequent processing, only foreground data is processed, and the background segmentation can effectively reduce the computation amount, so that the computation speed is improved.
5. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: the S4 includes the steps of:
s4.1, firstly, the phase is solved through a four-step phase shift formula, phase main values theta (x, y) of different frequencies are calculated,
as can be seen from the equation 1,
wherein I' (x, y) is the image mean gray scale and I "(x, y) is the image gray scale modulation;
s4.2 according to the phase main values of different frequencies, unwrapping is carried out through a multi-frequency heterodyne principle, an absolute phase value psi (x, y) is obtained,
ψ(x,y)＝N(x,y)×2π+θH(x, y) equation 5
Wherein f isH，fLAre each thetaH(x,y)，θL(x, y) corresponding frequency.
6. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: s5, distortion correction is carried out on the unwrapped phase unwrapped images in S4, compared with the distortion correction of the original image sequence, the step only carries out the distortion correction on the left and right phase unwrapped images, a large amount of calculation is reduced, undistorted coordinates are solved according to the corresponding relation before and after distortion from the distorted coordinates, the distortion correction is completed,
wherein (x, y) is ideal imaging coordinate, (x ', y') is distorted coordinate, and r2＝x2+y2、ki、piIs a distortion parameter.
7. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: s6 specifically includes the following steps:
s6.1: creating a matching container, taking the phase value and the limit slope as container indexes, creating a two-dimensional matching container, and putting the coordinate index value of the pixel point of the right phase diagram into the matching container according to the phase value and the limit slope value of the pixel point;
s6.2: and pixel-level stereo matching and sub-pixel-level interpolation, wherein the pixel-level stereo matching is to search a coordinate index value of a right phase diagram matched with the pixel-level stereo matching in a matching container according to the phase value and the limit slope of each pixel point of the left phase diagram, and calculate the disparity map.
8. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: s7 specifically includes the following steps:
s7.1, median filtering is carried out on the disparity map, and points with wrong matching in the matching process are removed through the median filtering;
and S7.2 bilateral filtering of the disparity map.
9. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: s8 specifically includes the following steps:
s8, calculating the three-dimensional point cloud according to the optimized disparity map, wherein the formula is as follows:
where f represents the focal length of the camera, T represents the length of the baseline, x, y represent coordinate values in the image coordinate system, and Δ x represents the parallax value.
10. The three-dimensional reconstruction method based on GPU parallel acceleration as claimed in claim 1, characterized in that: s9 noise elimination is carried out to the point cloud by adopting radius filtering mode capable of parallel processing, a certain point is taken as the center of the sphere, the radius R is appointed to calculate the number of the points falling in the center of the sphere, when the number of the points is larger than a given value, the points are reserved, and when the number is smaller than the given value, the points are considered as noise points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010557165.4A CN111899326A (en) | 2020-06-18 | 2020-06-18 | Three-dimensional reconstruction method based on GPU parallel acceleration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010557165.4A CN111899326A (en) | 2020-06-18 | 2020-06-18 | Three-dimensional reconstruction method based on GPU parallel acceleration |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111899326A true CN111899326A (en) | 2020-11-06 |
Family
ID=73206810
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010557165.4A Pending CN111899326A (en) | 2020-06-18 | 2020-06-18 | Three-dimensional reconstruction method based on GPU parallel acceleration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111899326A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112802184A (en) * | 2021-01-25 | 2021-05-14 | 三一重工股份有限公司 | Three-dimensional point cloud reconstruction method and system, electronic device and storage medium |
| CN116320357A (en) * | 2023-05-17 | 2023-06-23 | 浙江视觉智能创新中心有限公司 | 3D structured light camera system, method, electronic device and readable storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102538709A (en) * | 2012-01-09 | 2012-07-04 | 黑龙江科技学院 | Method for utilizing GPU (Graphics Processing Unit) concurrent computation in three-dimensional measurement system based on structured light |
| CN103335611A (en) * | 2013-06-13 | 2013-10-02 | 华中科技大学 | Method for GPU-based object three-dimensional shape measurement |
| CN104220351A (en) * | 2012-02-10 | 2014-12-17 | 迪尔公司 | Method and stereo vision system for facilitating the unloading of agricultural material from a vehicle |
| KR20160149088A (en) * | 2015-06-17 | 2016-12-27 | 한국전자통신연구원 | Method and apparatus for detecting disparity by using hierarchical stereo matching |
| US20170032565A1 (en) * | 2015-07-13 | 2017-02-02 | Shenzhen University | Three-dimensional facial reconstruction method and system |
| CN107860312A (en) * | 2017-10-20 | 2018-03-30 | 武汉惟景三维科技有限公司 | A kind of the embedded of structural light three-dimensional Measurement Algorithm realizes system and method |
| CN107970026A (en) * | 2017-11-17 | 2018-05-01 | 西安交通大学 | Vola three-dimensional scanner and method based on four step grating phase shift method of single camera |
| CN109035224A (en) * | 2018-07-11 | 2018-12-18 | 哈尔滨工程大学 | A kind of Technique of Subsea Pipeline Inspection and three-dimensional rebuilding method based on multi-beam point cloud |
| CN109489585A (en) * | 2018-12-06 | 2019-03-19 | 广西师范大学 | Based on the method for three-dimensional measurement for improving multifrequency fringe structure light |
| CN111207693A (en) * | 2020-01-10 | 2020-05-29 | 西安交通大学 | Three-dimensional measurement method of turbine blade ceramic core based on binocular structured light |
-
2020
- 2020-06-18 CN CN202010557165.4A patent/CN111899326A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102538709A (en) * | 2012-01-09 | 2012-07-04 | 黑龙江科技学院 | Method for utilizing GPU (Graphics Processing Unit) concurrent computation in three-dimensional measurement system based on structured light |
| CN104220351A (en) * | 2012-02-10 | 2014-12-17 | 迪尔公司 | Method and stereo vision system for facilitating the unloading of agricultural material from a vehicle |
| CN103335611A (en) * | 2013-06-13 | 2013-10-02 | 华中科技大学 | Method for GPU-based object three-dimensional shape measurement |
| KR20160149088A (en) * | 2015-06-17 | 2016-12-27 | 한국전자통신연구원 | Method and apparatus for detecting disparity by using hierarchical stereo matching |
| US20170032565A1 (en) * | 2015-07-13 | 2017-02-02 | Shenzhen University | Three-dimensional facial reconstruction method and system |
| CN107860312A (en) * | 2017-10-20 | 2018-03-30 | 武汉惟景三维科技有限公司 | A kind of the embedded of structural light three-dimensional Measurement Algorithm realizes system and method |
| CN107970026A (en) * | 2017-11-17 | 2018-05-01 | 西安交通大学 | Vola three-dimensional scanner and method based on four step grating phase shift method of single camera |
| CN109035224A (en) * | 2018-07-11 | 2018-12-18 | 哈尔滨工程大学 | A kind of Technique of Subsea Pipeline Inspection and three-dimensional rebuilding method based on multi-beam point cloud |
| CN109489585A (en) * | 2018-12-06 | 2019-03-19 | 广西师范大学 | Based on the method for three-dimensional measurement for improving multifrequency fringe structure light |
| CN111207693A (en) * | 2020-01-10 | 2020-05-29 | 西安交通大学 | Three-dimensional measurement method of turbine blade ceramic core based on binocular structured light |
Non-Patent Citations (4)
| Title |
|---|
| 佟成;王士同;: "一种基于脉冲噪声检测的中值滤波方法", 计算机应用研究, no. 05 * |
| 刘煜等: "《阵列相机成像技术与应用》", 30 April 2018, pages: 48 - 51 * |
| 张迎果: ""基于相位映射表的外极斜率立体匹配方法"", 《研究与设计》, vol. 28, no. 9, pages 12 - 14 * |
| 林静;江开勇;林俊义;李莉君;: "基于立体校正的快速相位立体匹配", 贵州师范大学学报(自然科学版), no. 02 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112802184A (en) * | 2021-01-25 | 2021-05-14 | 三一重工股份有限公司 | Three-dimensional point cloud reconstruction method and system, electronic device and storage medium |
| CN112802184B (en) * | 2021-01-25 | 2024-04-30 | 盛景智能科技(嘉兴)有限公司 | Three-dimensional point cloud reconstruction method, three-dimensional point cloud reconstruction system, electronic equipment and storage medium |
| CN116320357A (en) * | 2023-05-17 | 2023-06-23 | 浙江视觉智能创新中心有限公司 | 3D structured light camera system, method, electronic device and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Point-based multi-view stereo network | |
| US10706622B2 (en) | Point cloud meshing method, apparatus, device and computer storage media | |
| CN108961390B (en) | Real-time three-dimensional reconstruction method based on depth map | |
| Long et al. | Multi-view depth estimation using epipolar spatio-temporal networks | |
| CN111243093B (en) | Three-dimensional face grid generation method, device, equipment and storage medium | |
| CN110135455A (en) | Image matching method, device and computer readable storage medium | |
| CN107025660B (en) | Method and device for determining image parallax of binocular dynamic vision sensor | |
| CN111612882B (en) | Image processing method, image processing device, computer storage medium and electronic equipment | |
| CN111275633A (en) | Point cloud denoising method, system and device based on image segmentation and storage medium | |
| CN111899326A (en) | Three-dimensional reconstruction method based on GPU parallel acceleration | |
| KR101593316B1 (en) | Method and apparatus for recontructing 3-dimension model using stereo camera | |
| CN115239861A (en) | Face data enhancement method and device, computer equipment and storage medium | |
| CN115329111B (en) | Image feature library construction method and system based on point cloud and image matching | |
| CN114677479A (en) | Natural landscape multi-view three-dimensional reconstruction method based on deep learning | |
| CN111881985A (en) | Stereo matching method, device, terminal and storage medium | |
| Meerits et al. | Real-time scene reconstruction and triangle mesh generation using multiple RGB-D cameras | |
| CN117132737B (en) | Three-dimensional building model construction method, system and equipment | |
| Blanchet et al. | Fattening free block matching | |
| CN116977596A (en) | Three-dimensional modeling system and method based on multi-view images | |
| Zhou et al. | Patch-based texture synthesis for image inpainting | |
| CN115239559A (en) | Depth map super-resolution method and system for fusion view synthesis | |
| CN115564639A (en) | Background blurring method and device, computer equipment and storage medium | |
| Uh et al. | Efficient multiview stereo by random-search and propagation | |
| Wang et al. | Accelerating image-domain-warping virtual view synthesis on GPGPU | |
| Zhao et al. | 3dfill: Reference-guided image inpainting by self-supervised 3d image alignment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |