WO2022052582A1 - 一种图像配准方法、装置及电子设备和存储介质 - Google Patents
一种图像配准方法、装置及电子设备和存储介质 Download PDFInfo
- Publication number
- WO2022052582A1 WO2022052582A1 PCT/CN2021/103359 CN2021103359W WO2022052582A1 WO 2022052582 A1 WO2022052582 A1 WO 2022052582A1 CN 2021103359 W CN2021103359 W CN 2021103359W WO 2022052582 A1 WO2022052582 A1 WO 2022052582A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- parallax
- point
- pixel
- value
- confidence
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
- G06T7/55—Depth or shape recovery from multiple images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
- G06T7/55—Depth or shape recovery from multiple images
- G06T7/564—Depth or shape recovery from multiple images from contours
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
- G06T7/55—Depth or shape recovery from multiple images
- G06T7/593—Depth or shape recovery from multiple images from stereo images
-
- 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
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
-
- 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
- G06T2207/20228—Disparity calculation for image-based rendering
Definitions
- the present application relates to the technical field of image matching, and more particularly, to an image registration method, an apparatus, an electronic device, and a computer-readable storage medium.
- Binocular stereo vision uses the principle of bionics to imitate the human eye to form the basis of 3D vision. It uses two cameras separated by a certain distance to capture images, and calculates the pixel disparity between the matching points by finding the corresponding matching points in the two images. , and then the depth of the scene point from the camera can be calculated through the principle of triangulation.
- Image matching can be divided into two types: global method and local method.
- the parallax accuracy obtained by the global method is very high, but the global optimization process involves the design of a large number of parameters, the solution process faces complex and unstable problems, and the computational efficiency cannot meet the needs of online applications.
- the local method assumes that all pixels in the neighborhood window have the same disparity value, which will cause smooth and blurred disparity results, and there are problems such as boundary overflow, and it is difficult to obtain accurate depth boundaries.
- the local method is not as accurate as the global method, but it is very computationally efficient. Combining the advantages of the global method and the local method, the semi-global matching method SGM (semi-global matching) has emerged.
- This method calculates the global matching cost in several sampling directions while calculating the matching cost based on the window. Combined with the local cost and the global cost, the accuracy of the disparity calculation result is improved, and the calculation efficiency is between the local method and the global method. Due to the sampling optimization in the global direction, the semi-global method has problems such as fringe aliasing and boundary burrs in the parallax results, resulting in low image matching accuracy.
- the purpose of the present application is to provide an image registration method and apparatus, an electronic device and a computer-readable storage medium, which improve the matching accuracy of binocular images.
- an image registration method including:
- the disparity of the pixels between the points is assigned as a target value; wherein, the target value is a value close to the disparity value of the depth boundary point among the first disparity value and the second disparity value, and the first disparity value is the parallax value of the first pixel, the second parallax value is the parallax value of the second pixel, the first pixel and the second pixel are adjacent pixels, and the parallax boundary A point is located between the first pixel point and the second pixel point.
- the extracting the edge pixels of the binocular image includes:
- performing a filtering operation on the gradient image includes:
- the position gradient values in the preset neighborhood range with the same magnitude and different directions as the gradient values are set to 0.
- performing non-maximum suppression on the gradient image after the filtering operation in order to determine the edge pixels includes:
- a pixel corresponding to a target gradient value greater than all gradient values within the preset neighborhood range is determined as the edge pixel.
- determining the high-confidence parallax point and the parallax of each of the high-confidence parallax points in the edge pixels includes:
- An edge pixel whose matching cost meets a preset condition is determined as the high-confidence parallax point; wherein, the preset condition includes that the minimum matching cost is less than the first preset value and the ratio of the minimum matching cost to the next smallest matching cost is less than the first matching cost.
- the preset condition includes that the minimum matching cost is less than the first preset value and the ratio of the minimum matching cost to the next smallest matching cost is less than the first matching cost.
- the disparity of the minimum matching cost is determined as the disparity of the high-confidence disparity point.
- determining the depth boundary points in the edge pixels according to the distribution of the high-confidence parallax points includes:
- An edge pixel corresponding to a statistical value greater than the third preset value is determined as a depth boundary point.
- the determining the parallax boundary point of each depth boundary point in the target direction includes:
- a second preset search range is searched for each of the described depth boundary points along the target direction, and a parallax difference value between every two adjacent pixel points is determined;
- the middle position of the first pixel point and the second pixel point is determined as the parallax boundary point.
- an image registration device including:
- a first determination module configured to acquire a binocular image, extract edge pixels of the binocular image, and determine a high-confidence parallax point and the parallax of each of the high-confidence parallax points in the edge pixels;
- the projection module is used for triangulating all the high-confidence parallax points based on the three-dimensional space formed by the pixel coordinates of each high-confidence parallax point and the parallax to obtain a triangular mesh, and for each non-high-confidence parallax point The parallax point is projected to the triangular mesh along a direction parallel to the parallax dimension to obtain a triangular surface;
- the second determination module is configured to determine the interval formed by the maximum and minimum values of the parallax of the three vertices in each of the triangular faces as the parallax search range of the parallax of each of the non-high-confidence parallax points. Calculate the matching costs corresponding to all the parallaxes within the aforementioned parallax search range, and determine the parallax with the smallest matching cost as the parallax of the corresponding non-high-confidence parallax point;
- the adjustment module is configured to determine depth boundary points in the edge pixels according to the distribution of the high-confidence parallax points, determine the parallax boundary points of each of the depth boundary points in the target direction, and assign the depth boundary points to the edge pixels.
- the parallax to the pixels between the parallax boundary points is adjusted as a target value; wherein the target value is a value close to the parallax value of the depth boundary point among the first parallax value and the second parallax value, and the The first disparity value is the disparity value of the first pixel point, the second disparity value is the disparity value of the second pixel point, and the first pixel point and the second pixel point are adjacent pixel points , the parallax boundary point is located between the first pixel point and the second pixel point.
- an electronic device including:
- the processor is configured to implement the steps of the above-mentioned image registration method when executing the computer program.
- the present application provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above image registration method are implemented.
- an image registration method includes: acquiring a binocular image, extracting edge pixels of the binocular image, and determining high-confidence parallax points and each of the edge pixels in the edge pixels.
- the parallax of the high-confidence parallax point triangulate all the high-confidence parallax points based on the three-dimensional space formed by the pixel coordinates of each high-confidence parallax point and the parallax to obtain a triangular mesh, and for each of the high-confidence parallax points
- the non-high-confidence parallax point is projected to the triangular mesh along the direction parallel to the parallax dimension to obtain a triangular surface; the interval formed by the maximum and minimum parallax values of the three vertices in each of the triangular surfaces is determined as each
- the matching costs corresponding to all parallaxes are calculated
- the image matching method provided in this application proposes the idea of step-by-step matching.
- the high-confidence parallax points with obvious features are matched to obtain a relatively accurate parallax value, and then the high-confidence parallax points are used as a guide.
- Confidence disparity points are quickly matched, that is, the disparity values of non-high-confidence disparity points are determined.
- the depth boundary that can accurately correspond to the edge position of the object is extracted, and then the proposed parallax boundary optimization method based on the depth boundary is used.
- the boundary position in the dense parallax result is then filled with pixels to align the parallax boundary with the depth boundary, and finally the binocular image matching calculation result with accurate parallax boundary can be obtained. It can be seen that the image matching method provided by the present application improves the matching accuracy of binocular images.
- the present application also discloses an image registration device, an electronic device, and a computer-readable storage medium, which can also achieve the above technical effects.
- FIG. 1 is a flowchart of an image registration method according to an exemplary embodiment
- FIG. 2 is a flowchart of another image registration method according to an exemplary embodiment
- FIG. 3 is a flow chart of yet another image registration method according to an exemplary embodiment
- FIG. 4 is a structural diagram of an image registration apparatus according to an exemplary embodiment
- FIG. 5 is a structural diagram of an electronic device according to an exemplary embodiment.
- the embodiment of the present application discloses an image registration method, which improves the matching accuracy of binocular images.
- FIG. 1 a flowchart of an image registration method according to an exemplary embodiment, as shown in FIG. 1 , includes:
- S101 Acquire a binocular image, extract edge pixels of the binocular image, and determine a high-confidence parallax point and the parallax of each of the high-confidence parallax points in the edge pixels;
- the binocular images are two images captured by two cameras separated by a certain distance.
- edge pixels of the binocular image are extracted, and high-confidence disparity points are determined in the edge pixels. It should be noted that all pixels in the binocular image are divided into high-confidence parallax points and non-high-confidence parallax points. In this step, the parallax of the high-confidence parallax points is first determined according to a local matching algorithm.
- S102 Triangulate all the high-confidence parallax points based on the three-dimensional space formed by the pixel coordinates of each high-confidence parallax point and the parallax to obtain a triangular mesh, and perform triangulation for each non-high-confidence parallax point along the Projecting the triangular mesh in a direction parallel to the parallax dimension to obtain a triangular surface;
- S103 Determine the interval formed by the parallax maximum value and the minimum value of the three vertices in each of the triangular faces as a parallax search range of the parallax of each of the non-high-confidence parallax points, within each parallax search range Calculate the matching cost corresponding to all parallaxes, and determine the parallax with the smallest matching cost as the parallax of the corresponding non-high-confidence parallax point;
- Delaunay triangulation is performed in the three-dimensional space formed by its pixel coordinates and the corresponding parallax to obtain a triangular mesh.
- a pixel that is not a high-confidence parallax point project it to the triangular mesh along the direction parallel to the parallax dimension to obtain a specific triangular surface corresponding to it.
- the matching costs corresponding to all the parallaxes are calculated within the parallax search range, and the parallax corresponding to the minimum cost is selected as the parallax value of the current pixel.
- S104 Determine a depth boundary point in the edge pixels according to the distribution of the high-confidence parallax points, determine the parallax boundary point of each depth boundary point in the target direction, and assign the depth boundary point to the The disparity of the pixels between the disparity boundary points is assigned as the target value;
- the target value is a value close to the disparity value of the depth boundary point among the first disparity value and the second disparity value
- the first disparity value is the disparity value of the first pixel point
- the The second parallax value is the parallax value of the second pixel
- the first pixel and the second pixel are adjacent pixels
- the parallax boundary point is located between the first pixel and the first pixel. between two pixels.
- the determining a depth boundary point in the edge pixels according to the distribution of the high-confidence parallax points includes: determining a high-resolution edge pixel within a first preset search range for each edge pixel. Whether the number of confidence parallax points is greater than the third preset value; if so, calculate the statistical value of the parallax of all high-confidence parallax points within the first preset search range; wherein, the statistical value includes the mean square error; The edge pixel corresponding to the statistical value greater than the third preset value is determined as the depth boundary point.
- each edge pixel within a search window of a given size, it is searched whether there is a high-confidence disparity point. If it does not exist or exists less than a given threshold, the texture edge pixel is not a depth boundary; if it exists, go to high-confidence disparity point disparity distribution evaluation.
- the mean square error of all the confidence disparities within the search window range is calculated, and other metrics such as the absolute value of the difference between the maximum and minimum disparity can also be selected, which are not specifically limited here.
- the mean square error is greater than a given threshold, the current texture edge pixel is determined as a depth boundary, otherwise the current texture edge pixel is not a depth boundary.
- the parallax boundary point of each depth boundary point in the target direction is determined, and the target direction here includes horizontal positive direction, horizontal negative direction, vertical positive direction and vertical negative direction.
- the determining of the parallax boundary point of each depth boundary point in the target direction includes: searching a second preset search range for each depth boundary point along the target direction, and determining each depth boundary point in a second preset search range.
- the parallax difference value between two adjacent pixel points; the adjacent pixel points corresponding to the parallax difference value greater than the fourth preset value are determined as the first pixel point and the second pixel point; the first pixel point and the second pixel point are determined; The intermediate position between a pixel point and the second pixel point is determined as the parallax boundary point.
- a search is performed in a range of a given size along the horizontal positive (or negative) direction, and the parallax difference between two adjacent pixels is judged. , if the absolute value of the parallax difference between adjacent pixels A and B is greater than a given threshold, the position between the two pixels is regarded as the parallax boundary, and the search process in the negative (or positive) direction is not performed. Assign the disparity of all pixels between the current depth boundary point and the disparity boundary as the disparity of the pixel in A and B that is close to D.
- a search in the negative (or positive) direction is performed to adjust the parallax boundary point and parallax.
- the search in the vertical direction is similar to the search in the horizontal direction, and will not be repeated here.
- the present application designs a high-fidelity and efficient binocular image matching method for the problem of accurate parallax boundary preservation and high-resolution image-based calculation efficiency in the process of binocular image matching.
- an efficient matching method that can be used for high-resolution images is designed.
- the robust feature points with obvious features are used for high-confidence matching to obtain relatively accurate sparse disparity.
- the sparse disparity is used as a guide to quickly match the remaining pixels to obtain dense disparity results.
- a method for image depth boundary recognition is designed. First, the fine texture boundary of the image is extracted, combined with the sparse disparity value of the image, the depth boundary is identified in the texture boundary.
- the image matching method provided by the embodiment of the present application proposes the idea of step-by-step matching.
- the high-confidence parallax points with obvious features are used for matching to obtain a relatively accurate parallax value, and then the high-confidence parallax points are used as a guide.
- the non-high-confidence parallax points are quickly matched, that is, the parallax values of the non-high-confidence parallax points are determined.
- the depth boundary that can accurately correspond to the edge position of the object is extracted, and then the proposed parallax boundary optimization method based on the depth boundary is used.
- the boundary position in the dense parallax result is then filled with pixels to align the parallax boundary with the depth boundary, and finally the binocular image matching calculation result with accurate parallax boundary can be obtained. It can be seen that the image matching method provided by the embodiment of the present application improves the matching accuracy of the binocular image.
- FIG. 2 a flowchart of another image registration method according to an exemplary embodiment, as shown in FIG. 2 , includes:
- S201 Calculate the gradient value of each pixel in the binocular image, so as to obtain a gradient image corresponding to the binocular image, and perform a filtering operation on the gradient image;
- this step firstly, for each pixel in the binocular image, calculate the absolute value of the grayscale difference between it and the pixels in the surrounding 8 neighborhoods, take the maximum value as the gradient value of the current pixel, and record its direction to obtain the gradient image.
- the gradient image is filtered to merge non-isotropic repeated gradient values.
- performing a filtering operation on the gradient image includes: comparing each of the gradient values with all gradient values within a preset neighborhood range; comparing the preset neighborhood range with the The gradient values of the positions with the same gradient value and different directions are set to 0.
- for each gradient value compare whether the magnitude and direction of the gradient values in the surrounding 8 neighborhoods are the same as it, and set all the position gradient values with the same size but different directions to 0 to obtain the gradient filtered image.
- S202 Perform non-maximum suppression on the gradient image after the filtering operation, so as to determine edge pixels;
- non-maximum suppression is performed on the gradient filtered image.
- performing non-maximum suppression on the gradient image after the filtering operation to determine the edge pixels includes: comparing each target gradient value after the filtering operation with all gradient values within a preset neighborhood range ; Determine the pixel corresponding to the target gradient value greater than all gradient values within the preset neighborhood range as the edge pixel.
- the pixel for each pixel of the gradient filtered image, compare its gradient value with that of all pixels in the surrounding 8 neighborhoods, and if the pixel gradient is greater than or equal to all the pixel gradients in the surrounding 8 neighborhoods, the pixel is regarded as an edge pixel.
- S203 Perform a search for the neighborhood of a preset pixel range in the horizontal direction and the vertical direction for all the edge pixels. If the edge pixel exists in the neighborhood of the preset pixel range and the directly adjacent position is a non-edge pixel, then The directly adjacent positions are set as edge pixels in order to determine all edge pixels corresponding to the binocular image.
- edge pixels are searched for 2-pixel range neighborhoods in the horizontal and vertical directions respectively. If there are edge pixels in the 2-pixel range neighborhood in a certain direction and the immediate adjacent position is not an edge, the phase Neighboring pixels are set as edges, resulting in the full edge of the image.
- FIG. 3 a flowchart of another image registration method according to an exemplary embodiment, as shown in FIG. 3 , includes:
- S301 Extract the feature vector of each pixel in the binocular image, and perform matching cost calculation for all the edge pixels in the preset parallax search range with the feature vector of all pixels in the neighborhood window;
- the Sobel feature responses in the horizontal and vertical directions of all pixels in a certain neighborhood range around each pixel in the binocular image are calculated.
- several feature response values are selected from the calculated Sobel feature responses of all pixels in its surrounding neighborhood according to a certain specified rule to form a feature vector, which is used as the feature description of the pixel.
- the matching cost is calculated based on the features of all pixels in the surrounding neighborhood window.
- S302 Determine the edge pixels whose matching cost meets a preset condition as the high-confidence disparity point; wherein, the preset condition includes that the minimum matching cost is less than a first preset value and the ratio of the minimum matching cost to the next smallest matching cost less than the second preset value;
- S303 Determine the disparity of the minimum matching cost as the disparity of the high-confidence disparity point.
- the pixel if the minimum value of all the calculated disparity matching costs is greater than a given threshold, the pixel is not a high-confidence match; otherwise, select all the calculated disparity matching costs.
- the minimum value and the second minimum value of calculate the ratio of the minimum value to the second minimum value. If the ratio is less than a given threshold, the matching cost of the corresponding minimum value is the matching cost of the corresponding pixel, and the corresponding disparity value is used as the pixel. disparity; otherwise, the pixel is not a high-confidence match.
- An image registration apparatus provided by an embodiment of the present application is introduced below.
- An image registration apparatus described below and an image registration method described above can be referred to each other.
- FIG. 4 a structural diagram of an image registration apparatus according to an exemplary embodiment, as shown in FIG. 4 , includes:
- a first determination module 401 configured to acquire a binocular image, extract edge pixels of the binocular image, and determine a high-confidence parallax point and the parallax of each of the high-confidence parallax points in the edge pixels;
- the projection module 402 is configured to triangulate all the high-confidence parallax points based on the three-dimensional space formed by the pixel coordinates of each of the high-confidence parallax points and the parallax to obtain a triangular mesh, and for each non-high-confidence parallax point The parallax point is projected to the triangular mesh along a direction parallel to the parallax dimension to obtain a triangular surface;
- the second determining module 403 is configured to determine the interval formed by the parallax maximum value and the minimum value of the three vertices in each of the triangular faces as the parallax search range of the parallax of each of the non-high-confidence parallax points. Calculate the matching costs corresponding to all the parallaxes within the parallax search range, and determine the parallax with the smallest matching cost as the parallax of the corresponding non-high-confidence parallax point;
- the adjustment module 404 is configured to determine a depth boundary point in the edge pixels according to the distribution of the high-confidence parallax points, determine the parallax boundary point of each of the depth boundary points in the target direction, and assign the depth boundary point.
- the parallax of the pixel between the point and the parallax boundary point is adjusted as a target value; wherein, the target value is a value close to the parallax value of the depth boundary point among the first parallax value and the second parallax value, so
- the first disparity value is the disparity value of the first pixel point
- the second disparity value is the disparity value of the second pixel point
- the first pixel point and the second pixel point are adjacent pixels point
- the parallax boundary point is located between the first pixel point and the second pixel point.
- the image matching device provided in the embodiment of the present application proposes the idea of step-by-step matching.
- the high-confidence parallax points with obvious features are matched to obtain a relatively accurate parallax value, and then the high-confidence parallax points are used as a guide.
- the non-high-confidence parallax points are quickly matched, that is, the parallax values of the non-high-confidence parallax points are determined.
- the depth boundary that can accurately correspond to the edge position of the object is extracted, and then the proposed parallax boundary optimization method based on the depth boundary is used.
- the boundary position in the dense parallax result is then filled with pixels to align the parallax boundary with the depth boundary, and finally the binocular image matching calculation result with accurate parallax boundary can be obtained. It can be seen that the image matching device provided by the embodiment of the present application improves the matching accuracy of the binocular image.
- the first determination module 401 includes:
- a first calculation unit configured to calculate the gradient value of each pixel in the binocular image, so as to obtain a gradient image corresponding to the binocular image, and perform a filtering operation on the gradient image;
- a first determination unit configured to perform non-maximum suppression on the gradient image after the filtering operation, so as to determine edge pixels
- the second determination unit is configured to perform a search for the neighborhood of preset pixel ranges in the horizontal direction and the vertical direction for all the edge pixels. If the edge pixel exists in the neighborhood of the preset pixel range and the directly adjacent position is If it is not an edge pixel, the directly adjacent position is set as an edge pixel, so as to determine all edge pixels corresponding to the binocular image.
- the first computing unit includes:
- a calculation subunit used to calculate the gradient value of each pixel in the binocular image, so as to obtain a gradient image corresponding to the binocular image
- a first comparison subunit for comparing each of the gradient values with all gradient values within a preset neighborhood range
- a subunit is set, configured to set the gradient value of the position within the preset neighborhood range with the same magnitude and different direction as the gradient value to 0.
- the first determining unit includes:
- the second comparison subunit is used to compare each target gradient value after the filtering operation with all gradient values within the preset neighborhood range
- a determination subunit configured to determine a pixel corresponding to a target gradient value greater than all gradient values within the preset neighborhood range as the edge pixel.
- the first determination module 401 includes:
- an extraction unit for extracting edge pixels of the binocular image
- the second calculation unit is used to extract the feature vector of each pixel in the binocular image, and perform matching cost calculation for all the edge pixels within the preset parallax search range with the feature vectors of all pixels in the neighborhood window;
- a third determining unit configured to determine the edge pixels whose matching cost meets a preset condition as the high-confidence disparity point; wherein the preset condition includes that the minimum matching cost is less than the first preset value and the minimum matching cost is the same as the second The ratio of the small matching cost is less than the second preset value;
- the fourth determining unit is configured to determine the parallax of the minimum matching cost as the parallax of the high-confidence parallax point.
- the adjustment module 404 includes:
- a fifth determination unit configured to determine whether the number of high-confidence parallax points is greater than a third preset value within the first preset search range for each of the edge pixels; if so, calculate the first preset search range Statistical values of parallax of all high-confidence parallax points within; wherein, the statistic values include mean square error; the edge pixels corresponding to the statistic values greater than the third preset value are determined as depth boundary points;
- a first adjustment unit configured to determine the parallax boundary point of each of the depth boundary points in the target direction, and adjust the parallax between the depth boundary point and the pixel between the parallax boundary points to a target value; wherein,
- the target value is a value close to the disparity value of the depth boundary point among the first disparity value and the second disparity value, the first disparity value is the disparity value of the first pixel point, and the second disparity value is the disparity value of the first pixel.
- the parallax value is the parallax value of the second pixel, the first pixel and the second pixel are adjacent pixels, and the parallax boundary point is located between the first pixel and the second pixel between points.
- the adjustment module 404 includes:
- a sixth determining unit configured to determine depth boundary points in the edge pixels according to the distribution of the high-confidence parallax points
- the seventh determination unit is used to search the second preset search range for each of the depth boundary points along the target direction, and determine the parallax difference value between every two adjacent pixel points; it will be greater than the fourth preset value
- the adjacent pixel points corresponding to the parallax difference value are determined as the first pixel point and the second pixel point; the middle position of the first pixel point and the second pixel point is determined as the parallax boundary point ;
- the second adjustment unit is configured to adjust the disparity of the pixel between the depth boundary point and the disparity boundary point to a target value.
- FIG. 5 is a structural diagram of an electronic device according to an exemplary embodiment. As shown in Figure 5, the electronic equipment includes:
- Communication interface 1 which can exchange information with other devices such as network devices;
- the processor 2 is connected to the communication interface 1 to realize information interaction with other devices, and is used to execute the access method of the application provided by one or more of the above technical solutions when running the computer program.
- the computer program is instead stored on the memory 3 .
- bus system 4 is used to realize the connection communication between these components.
- the bus system 4 also includes a power bus, a control bus and a status signal bus.
- the various buses are designated as bus system 4 in FIG. 5 .
- the memory 3 in the embodiment of the present application is used to store various types of data to support the operation of the electronic device. Examples of such data include: any computer program used to operate on an electronic device.
- the memory 3 may be a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memory.
- the non-volatile memory can be a read-only memory (ROM, Read Only Memory), a programmable read-only memory (PROM, Programmable Read-Only Memory), an erasable programmable read-only memory (EPROM, Erasable Programmable Read-only memory) Only Memory), Electrically Erasable Programmable Read-Only Memory (EEPROM, Electrically Erasable Programmable Read-Only Memory), Magnetic Random Access Memory (FRAM, ferromagnetic random access memory), Flash Memory (Flash Memory), Magnetic Surface Memory , CD-ROM, or CD-ROM (Compact Disc Read-Only Memory); magnetic surface memory can be disk memory or tape memory.
- RAM Random Access Memory
- SRAM Static Random Access Memory
- SSRAM Synchronous Static Random Access Memory
- DRAM Dynamic Random Access Memory
- SDRAM Synchronous Dynamic Random Access Memory
- DDRSDRAM Double Data Rate Synchronous Dynamic Random Access Memory
- ESDRAM Enhanced Type Synchronous Dynamic Random Access Memory
- SLDRAM Synchronous Link Dynamic Random Access Memory
- DRRAM Direct Rambus Random Access Memory
- the memory 2 described in the embodiments of the present application is intended to include but not limited to these and any other suitable types of memory.
- the methods disclosed in the above embodiments of the present application may be applied to the processor 2 or implemented by the processor 2 .
- the processor 2 may be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above-mentioned method can be completed by a hardware integrated logic circuit in the processor 2 or an instruction in the form of software.
- the above-mentioned processor 2 may be a general-purpose processor, a DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the like.
- the processor 2 may implement or execute the methods, steps, and logical block diagrams disclosed in the embodiments of this application.
- a general purpose processor may be a microprocessor or any conventional processor or the like.
- the steps of the methods disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor.
- the software module may be located in a storage medium, the storage medium is located in the memory 3, and the processor 2 reads the program in the memory 3, and completes the steps of the foregoing method in combination with its hardware.
- the embodiment of the present application further provides a storage medium, that is, a computer storage medium, specifically a computer-readable storage medium, for example, including a memory 3 storing a computer program, and the above-mentioned computer program can be executed by the processor 2, to complete the steps described in the preceding method.
- the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disk, or CD-ROM.
- the aforementioned program may be stored in a computer-readable storage medium, and when the program is executed, execute It includes the steps of the above method embodiments; and the aforementioned storage medium includes: a removable storage device, a ROM, a RAM, a magnetic disk or an optical disk and other media that can store program codes.
- the above-mentioned integrated units of the present application are implemented in the form of software function modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.
- the computer software products are stored in a storage medium and include several instructions for An electronic device (which may be a personal computer, a server, or a network device, etc.) is caused to execute all or part of the methods described in the various embodiments of the present application.
- the aforementioned storage medium includes: a removable storage device, a ROM, a RAM, a magnetic disk or an optical disk and other mediums that can store program codes.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
一种图像配准方法、装置及一种电子设备和计算机可读存储介质,包括:获取双目图像,提取双目图像的边缘像素,并在边缘像素中确定高置信度视差点和视差;对所有高置信度视差点进行三角剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向三角网格进行投影得到三角面;确定每个非高置信度视差点视差的视差搜索范围,在每个视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;在边缘像素中确定深度边界点,确定每个深度边界点在目标方向上的视差边界点,将深度边界点至视差边界点之间的像素的视差赋值为目标值,提高了双目图像的匹配精度。
Description
本申请要求于2020年9月8日提交中国专利局、申请号为202010934767.7、发明名称为“一种图像配准方法、装置及电子设备和存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及图像匹配技术领域,更具体地说,涉及一种图像配准方法、装置及一种电子设备和一种计算机可读存储介质。
双目立体视觉是利用仿生原理,模仿人眼形成3D视觉的基理,利用相隔一定距离的两个相机拍摄图像,通过在两幅图像中寻找对应的匹配点,计算匹配点之间的像素视差,再通过三角测量原理就可以计算出该场景点距离相机的深度。
图像匹配整体上可以分为全局方法和局部方法两种。全局方法得到的视差精度非常高,但是全局优化过程涉及大量参数的设计,求解过程面临复杂和不稳定的问题,计算效率满足不了在线应用需求。局部方法假设邻域窗口内所有像素具有相同的视差值,会造成视差结果的平滑模糊,存在边界溢出等问题,难以得到精确的深度边界。局部方法精度不如全局方法,但计算效率非常高。结合全局方法和局部方法的优点出现了半全局匹配方法SGM(semi-global matching)等,这种方法在基于窗口进行匹配代价计算的同时,在若干个采样方向上进行全局匹配代价的计算,通过结合局部代价和全局代价,提高了视差计算结果的精确度,计算效率介于局部方法和全局方法之间。半全局方法由于在全局方向上的采样优化,视差结果中存在条纹走样及边界毛刺等问题,导致图像匹配精度较低。
因此,如何提高双目图像的匹配精度是本领域技术人员需要解决的技术问题。
发明内容
本申请的目的在于提供一种图像配准方法、装置及一种电子设备和一种计算机可读存储介质,提高了双目图像的匹配精度。
为实现上述目的,本申请提供了一种图像配准方法,包括:
获取双目图像,提取所述双目图像的边缘像素,并在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差;
基于每个所述高置信度视差点的像素坐标和视差构成的三维空间对所有所述高置信度视差点进行三角剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向所述三角网格进行投影得到三角面;
将每个所述三角面中三个顶点的视差最大值和最小值形成的区间确定为每个所述非高置信度视差点视差的视差搜索范围,在每个所述视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;
根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,确定每个所述深度边界点在目标方向上的视差边界点,并将所述深度边界点至所述视差边界点之间的像素的视差赋值为目标值;其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
其中,所述提取所述双目图像的边缘像素,包括:
计算所述双目图像中每个像素的梯度值,以便得到所述双目图像对应的梯度图像,并对所述梯度图像进行滤波操作;
对滤波操作后的梯度图像进行非最大抑制,以便确定边缘像素;
对所有所述边缘像素进行水平方向和竖直方向预设像素范围邻域的搜索,若在所述预设像素范围邻域内存在所述边缘像素且直接相邻位置为非边缘像素,则将所述直接相邻位置设置为边缘像素,以便确定所述双目图像对应的所有边缘像素。
其中,对所述梯度图像进行滤波操作,包括:
比较每个所述梯度值与预设邻域范围内的所有梯度值;
将所述预设邻域范围内与所述梯度值大小相同方向不同的位置梯度值设置为0。
其中,所述对滤波操作后的梯度图像进行非最大抑制,以便确定边缘像素,包括:
比较滤波操作后的每个目标梯度值与预设邻域范围内的所有梯度值;
将大于所述预设邻域范围内的所有梯度值的目标梯度值对应的像素确定为所述边缘像素。
其中,在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差,包括:
提取所述双目图像中每个像素的特征向量,对所有所述边缘像素在预设视差搜索范围内以邻域窗口内所有像素的特征向量进行匹配代价计算;
将匹配代价满足预设条件的边缘像素确定为所述高置信度视差点;其中,所述预设条件包括最小匹配代价小于第一预设值且最小匹配代价与次小匹配代价的比值小于第二预设值;
将最小匹配代价的视差确定为所述高置信度视差点的视差。
其中,所述根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,包括:
对每个所述边缘像素在第一预设搜索范围内判断高置信度视差点的数量是否大于第三预设值;
若是,则计算所述第一预设搜索范围内所有高置信度视差点的视差的统计值;其中,所述统计值包括均方差;
将大于第三预设值的统计值对应的边缘像素确定为深度边界点。
其中,所述确定每个深度边界点在目标方向上的视差边界点,包括:
对每个所述深度边界点沿目标方向进行第二预设搜索范围的搜索,确定每两个相邻像素点之间的视差差值;
将大于第四预设值的视差差值对应的相邻像素点确定为所述第一像素点和所述第二像素点;
将所述第一像素点和所述第二像素点的中间位置确定为所述视差边界点。
为实现上述目的,本申请提供了一种图像配准装置,包括:
第一确定模块,用于获取双目图像,提取所述双目图像的边缘像素,并在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差;
投影模块,用于基于每个所述高置信度视差点的像素坐标和视差构成的三维空间对所有所述高置信度视差点进行三角剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向所述三角网格进行投影得到三角面;
第二确定模块,用于将每个所述三角面中三个顶点的视差最大值和最小值形成的区间确定为每个所述非高置信度视差点视差的视差搜索范围,在每个所述视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;
调整模块,用于根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,确定每个所述深度边界点在目标方向上的视差边界点,并将所述深度边界点至所述视差边界点之间的像素的视差调整为目标值;其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
为实现上述目的,本申请提供了一种电子设备,包括:
存储器,用于存储计算机程序;
处理器,用于执行所述计算机程序时实现如上述图像配准方法的步骤。
为实现上述目的,本申请提供了一种计算机可读存储介质,所述计算机可读存储介质上存储有计算机程序,所述计算机程序被处理器执行时实现如上述图像配准方法的步骤。
通过以上方案可知,本申请提供的一种图像配准方法,包括:获取双目图像,提取所述双目图像的边缘像素,并在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差;基于每个所述高置信度视差点的像素坐标和视差构成的三维空间对所有所述高置信度视差点进行三角 剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向所述三角网格进行投影得到三角面;将每个所述三角面中三个顶点的视差最大值和最小值形成的区间确定为每个所述非高置信度视差点视差的视差搜索范围,在每个所述视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,确定每个所述深度边界点在目标方向上的视差边界点,并将所述深度边界点至所述视差边界点之间的像素的视差赋值为目标值;其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
本申请提供的图像匹配方法,提出了分步匹配的思想,首先通过特征明显的高置信度视差点进行匹配,得到比较精确的视差值,然后以高置信度视差点作为引导,对非高置信度视差点进行快速匹配,即确定非高置信度视差点的视差值。其次,通过提取图像深度边界,提取出能精确对应物体边缘位置的深度边界,然后利用提出的基于深度边界的视差边界优化方法,以深度边界为基准,在一定的邻域范围通过特定方向搜索确定稠密视差结果中的边界位置,再通过像素填充使得视差边界与深度边界对齐,最后可以得到具有精确视差边界的双目图像匹配计算结果。由此可见,本申请提供的图像匹配方法,提高了双目图像的匹配精度。本申请还公开了一种图像配准装置及一种电子设备和一种计算机可读存储介质,同样能实现上述技术效果。
应当理解的是,以上的一般描述和后文的细节描述仅是示例性的,并不能限制本申请。
为了更清楚地说明本申请实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地, 下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。附图是用来提供对本公开的进一步理解,并且构成说明书的一部分,与下面的具体实施方式一起用于解释本公开,但并不构成对本公开的限制。在附图中:
图1为根据一示例性实施例示出的一种图像配准方法的流程图;
图2为根据一示例性实施例示出的另一种图像配准方法的流程图;
图3为根据一示例性实施例示出的又一种图像配准方法的流程图;
图4为根据一示例性实施例示出的一种图像配准装置的结构图;
图5为根据一示例性实施例示出的一种电子设备的结构图。
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述。显然,所描述的实施例仅仅是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。另外,在本申请实施例中,“第一”、“第二”等是用于区别类似的对象,而不必用于描述特定的顺序或先后次序。
本申请实施例公开了一种图像配准方法,提高了双目图像的匹配精度。
参见图1,根据一示例性实施例示出的一种图像配准方法的流程图,如图1所示,包括:
S101:获取双目图像,提取所述双目图像的边缘像素,并在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差;
在本实施例中,双目图像为相隔一定距离的两个相机拍摄的两幅图像。在具体实施中,提取双目图像的边缘像素,并在边缘像素中确定高置信度视差点。需要说明的是,双目图像中的所有像素被划分为高置信度视差点和非高置信度视差点,在本步骤首先按照局部匹配算法确定高置信度视差点的视差。
S102:基于每个所述高置信度视差点的像素坐标和视差构成的三维空间对所有所述高置信度视差点进行三角剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向所述三角网格进行投影得到三角面;
S103:将每个所述三角面中三个顶点的视差最大值和最小值形成的区间确定为每个所述非高置信度视差点视差的视差搜索范围,在每个所述视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;
在具体实施中,对于每个高置信度视差点,在其像素坐标和对应视差构成的三维空间内进行Delaunay三角剖分,得到三角网格。对任一非高置信度视差点的像素,沿着平行于视差维度的方向向三角网格做投影,得到与其对应的某个具体三角面。计算该三角面三个顶点的视差最小值和最大值,形成一个视差区间,把该区间作为当前像素匹配计算的视差搜索范围。按照局部匹配算法的计算过程,在视差搜索范围内计算所有视差对应的匹配代价,选择其中最小代价对应的视差作为当前像素的视差值。
S104:根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,确定每个所述深度边界点在目标方向上的视差边界点,并将所述深度边界点至所述视差边界点之间的像素的视差赋值为目标值;
其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
本步骤的目的在于确定双目图像的深度边界点,从而基于深度边界进行视差边界的对齐。作为一种可行的实施方式,所述根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,包括:对每个所述边缘像素在第一预设搜索范围内判断高置信度视差点的数量是否大于第三预设值;若是,则计算所述第一预设搜索范围内所有高置信度视差点的视差的统计值;其中,所述统计值包括均方差;将大于第三预设值的统计值对应的边缘像素确定为深度边界点。
在具体实施中,对每个边缘像素,在一个给定大小的搜索窗口范围内,搜索是否存在高置信度视差点。如果不存在或存在个数小于一个给定阈值,则该纹理边缘像素不是深度边界;如果存在,转到进行高置信度视差点视差分布评估。作为一种可行的实施方式,计算搜索窗口范围内所有该置信度视差的均方差,也可以选用其他度量标准如最大和最小视差之间差值的绝对值,在此不进行具体限定。对于深度边界判定,若均方差大于一个给定的阈值,则当前纹理边缘像素判定为深度边界,否则当前纹理边缘像素不是深度边界。
确定深度边界点后,确定每个深度边界点在目标方向上的视差边界点,此处的目标方向包括水平正向、水平负向、竖直正向和竖直负向。作为一种可行的实施方式,所述确定每个深度边界点在目标方向上的视差边界点,包括:对每个所述深度边界点沿目标方向进行第二预设搜索范围的搜索,确定每两个相邻像素点之间的视差差值;将大于第四预设值的视差差值对应的相邻像素点确定为所述第一像素点和所述第二像素点;将所述第一像素点和所述第二像素点的中间位置确定为所述视差边界点。
在具体实施中,以水平方向搜索为例,对任一深度边界点D,沿水平正(或负)方向一个给定大小的范围内进行搜索,判断相邻两个像素之间的视差差值,若出现相邻像素A和B的视差差值绝对值大于一个给定阈值,则将这两个像素之间的位置认为视差边界,同时不再进行负(或正)方向的搜索过程。将当前深度边界点和视差边界之间的所有像素视差赋值为A和B中靠近D的那个像素的视差。若未出现相邻像素视差差值绝对值大于一个给定阈值,则进行负(或正)方向的搜索,从而视差边界点以及视差调整。竖直方向搜索与水平方向的搜索类似,在此不再赘述。
可见,本申请针对双目图像匹配过程中的精确视差边界保持和基于高分辨率图像的计算效率问题,设计了一种高保真度高效双目图像匹配方法。针对匹配计算过程耗时的问题,设计了一种可用于高分辨率图像的高效匹配方法,首先通过特征明显的鲁棒特征点进行高置信度的匹配,得到比较精确的稀疏视差,然后以这些稀疏视差作为引导,对剩余像素进行快速匹配,得到稠密视差结果;设计了一种图像深度边界识别方法,首先提取图 像的精细纹理边界,结合图像的稀疏视差值,在纹理边界中识别深度边界,提取出能精确对应物体边缘位置的深度边界;设计了一种基于深度边界的视差边界优化方法,以深度边界为基准,在一定的邻域范围通过特定方向搜索确定稠密视差结果中的边界位置,再通过像素填充使得视差边界与深度边界对齐。
本申请实施例提供的图像匹配方法,提出了分步匹配的思想,首先通过特征明显的高置信度视差点进行匹配,得到比较精确的视差值,然后以高置信度视差点作为引导,对非高置信度视差点进行快速匹配,即确定非高置信度视差点的视差值。其次,通过提取图像深度边界,提取出能精确对应物体边缘位置的深度边界,然后利用提出的基于深度边界的视差边界优化方法,以深度边界为基准,在一定的邻域范围通过特定方向搜索确定稠密视差结果中的边界位置,再通过像素填充使得视差边界与深度边界对齐,最后可以得到具有精确视差边界的双目图像匹配计算结果。由此可见,本申请实施例提供的图像匹配方法,提高了双目图像的匹配精度。
下面介绍一种具体的确定边缘像素的方法,具体的:
参见图2,根据一示例性实施例示出的另一种图像配准方法的流程图,如图2所示,包括:
S201:计算所述双目图像中每个像素的梯度值,以便得到所述双目图像对应的梯度图像,并对所述梯度图像进行滤波操作;
在本步骤中,首先对双目图像中每个像素,分别计算其与周围8邻域内像素灰度差值绝对值,取其中的最大值作为当前像素的梯度值,并记录其方向,得到梯度图像。
其次,对梯度图像进行滤波,合并非同向重复梯度值。作为一种可行的实施方式,对所述梯度图像进行滤波操作,包括:比较每个所述梯度值与预设邻域范围内的所有梯度值;将所述预设邻域范围内与所述梯度值大小相同方向不同的位置梯度值设置为0。在具体实施中,对每个梯度值,比较其周围8邻域内梯度值大小和方向是否与其相同,将所有与其大小相同但方向不同的位置梯度值设置为0,得到梯度滤波后的图像。
S202:对滤波操作后的梯度图像进行非最大抑制,以便确定边缘像素;
在本步骤,对梯度滤波后图像进行非最大抑制。作为一种可行的实施方式,所述对滤波操作后的梯度图像进行非最大抑制,以便确定边缘像素,包括:比较滤波操作后的每个目标梯度值与预设邻域范围内的所有梯度值;将大于所述预设邻域范围内的所有梯度值的目标梯度值对应的像素确定为所述边缘像素。
在具体实施中,对梯度滤波后图像的每一个像素,比较其与周围8邻域内所有像素梯度值大小,如果该像素梯度大于等于周围8邻域内所有像素梯度,则将该像素作为边缘像素。
S203:对所有所述边缘像素进行水平方向和竖直方向预设像素范围邻域的搜索,若在所述预设像素范围邻域内存在所述边缘像素且直接相邻位置为非边缘像素,则将所述直接相邻位置设置为边缘像素,以便确定所述双目图像对应的所有边缘像素。
在本步骤中,对所有边缘像素分别进行水平和竖直方向2像素范围邻域的搜索,若某个方向上的2像素范围邻域内存在边缘像素且直接相邻位置不是边缘,则把该相邻位置像素设置为边缘,最终得到图像的完整边缘。
下面介绍一种具体的确定高置信度视差点的视差的方法,具体的:
参见图3,根据一示例性实施例示出的又一种图像配准方法的流程图,如图3所示,包括:
S301:提取所述双目图像中每个像素的特征向量,对所有所述边缘像素在预设视差搜索范围内以邻域窗口内所有像素的特征向量进行匹配代价计算;
在本步骤中,计算双目图像中每个像素周围一定邻域范围内所有像素在水平和竖直方向的Sobel特征响应。对每个像素,从计算得到的其周围邻域内所有像素Sobel特征响应中按照某种指定规则选取若干特征响应值组成特征向量,作为该像素的特征描述。对所有图像边缘像素,按照局部匹配算法,在给定的一个视差搜索范围内,以周围邻域窗口内所有像素的特征进行匹配代价计算。
S302:将匹配代价满足预设条件的边缘像素确定为所述高置信度视差点;其中,所述预设条件包括最小匹配代价小于第一预设值且最小匹配代价与次小匹配代价的比值小于第二预设值;
S303:将最小匹配代价的视差确定为所述高置信度视差点的视差。
在具体实施中,对于某个像素,若计算得到的所有视差匹配代价中的最小值大于某个给定的阈值,则该像素不是高置信度匹配;否则,选取计算得到的所有视差匹配代价中的最小值和次小值,计算最小值与次小值的比值,如果该比值小于某个给定阈值,将对应最小值的匹配代价最为对应像素的匹配代价,对应的视差值作为该像素的视差;否则,该像素不是高置信度匹配。
下面对本申请实施例提供的一种图像配准装置进行介绍,下文描述的一种图像配准装置与上文描述的一种图像配准方法可以相互参照。
参见图4,根据一示例性实施例示出的一种图像配准装置的结构图,如图4所示,包括:
第一确定模块401,用于获取双目图像,提取所述双目图像的边缘像素,并在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差;
投影模块402,用于基于每个所述高置信度视差点的像素坐标和视差构成的三维空间对所有所述高置信度视差点进行三角剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向所述三角网格进行投影得到三角面;
第二确定模块403,用于将每个所述三角面中三个顶点的视差最大值和最小值形成的区间确定为每个所述非高置信度视差点视差的视差搜索范围,在每个所述视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;
调整模块404,用于根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,确定每个所述深度边界点在目标方向上的视差边界点,并将所述深度边界点至所述视差边界点之间的像素的视差调整为目标值; 其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
本申请实施例提供的图像匹配装置,提出了分步匹配的思想,首先通过特征明显的高置信度视差点进行匹配,得到比较精确的视差值,然后以高置信度视差点作为引导,对非高置信度视差点进行快速匹配,即确定非高置信度视差点的视差值。其次,通过提取图像深度边界,提取出能精确对应物体边缘位置的深度边界,然后利用提出的基于深度边界的视差边界优化方法,以深度边界为基准,在一定的邻域范围通过特定方向搜索确定稠密视差结果中的边界位置,再通过像素填充使得视差边界与深度边界对齐,最后可以得到具有精确视差边界的双目图像匹配计算结果。由此可见,本申请实施例提供的图像匹配装置,提高了双目图像的匹配精度。
在上述实施例的基础上,作为一种优选实施方式,所述第一确定模块401包括:
获取单元,用于获取双目图像;
第一计算单元,用于计算所述双目图像中每个像素的梯度值,以便得到所述双目图像对应的梯度图像,并对所述梯度图像进行滤波操作;
第一确定单元,用于对滤波操作后的梯度图像进行非最大抑制,以便确定边缘像素;
第二确定单元,用于对所有所述边缘像素进行水平方向和竖直方向预设像素范围邻域的搜索,若在所述预设像素范围邻域内存在所述边缘像素且直接相邻位置为非边缘像素,则将所述直接相邻位置设置为边缘像素,以便确定所述双目图像对应的所有边缘像素。
在上述实施例的基础上,作为一种优选实施方式,所述第一计算单元包括:
计算子单元,用于计算所述双目图像中每个像素的梯度值,以便得到所述双目图像对应的梯度图像;
第一比较子单元,用于比较每个所述梯度值与预设邻域范围内的所有梯度值;
设置子单元,用于将所述预设邻域范围内与所述梯度值大小相同方向不同的位置梯度值设置为0。
在上述实施例的基础上,作为一种优选实施方式,所述第一确定单元包括:
第二比较子单元,用于比较滤波操作后的每个目标梯度值与预设邻域范围内的所有梯度值;
确定子单元,用于将大于所述预设邻域范围内的所有梯度值的目标梯度值对应的像素确定为所述边缘像素。
在上述实施例的基础上,作为一种优选实施方式,所述第一确定模块401包括:
获取单元,用于获取双目图像;
提取单元,用于提取所述双目图像的边缘像素;
第二计算单元,用于提取所述双目图像中每个像素的特征向量,对所有所述边缘像素在预设视差搜索范围内以邻域窗口内所有像素的特征向量进行匹配代价计算;
第三确定单元,用于将匹配代价满足预设条件的边缘像素确定为所述高置信度视差点;其中,所述预设条件包括最小匹配代价小于第一预设值且最小匹配代价与次小匹配代价的比值小于第二预设值;
第四确定单元,用于将最小匹配代价的视差确定为所述高置信度视差点的视差。
在上述实施例的基础上,作为一种优选实施方式,所述调整模块404包括:
第五确定单元,用于对每个所述边缘像素在第一预设搜索范围内判断高置信度视差点的数量是否大于第三预设值;若是,则计算所述第一预设搜索范围内所有高置信度视差点的视差的统计值;其中,所述统计值包括均方差;将大于第三预设值的统计值对应的边缘像素确定为深度边界点;
第一调整单元,用于确定每个所述深度边界点在目标方向上的视差边 界点,并将所述深度边界点至所述视差边界点之间的像素的视差调整为目标值;其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
在上述实施例的基础上,作为一种优选实施方式,所述调整模块404包括:
第六确定单元,用于根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点;
第七确定单元,用于对每个所述深度边界点沿目标方向进行第二预设搜索范围的搜索,确定每两个相邻像素点之间的视差差值;将大于第四预设值的视差差值对应的相邻像素点确定为所述第一像素点和所述第二像素点;将所述第一像素点和所述第二像素点的中间位置确定为所述视差边界点;
第二调整单元,用于将所述深度边界点至所述视差边界点之间的像素的视差调整为目标值。
关于上述实施例中的装置,其中各个模块执行操作的具体方式已经在有关该方法的实施例中进行了详细描述,此处将不做详细阐述说明。
基于上述程序模块的硬件实现,且为了实现本申请实施例的方法,本申请实施例还提供了一种电子设备,图5为根据一示例性实施例示出的一种电子设备的结构图,如图5所示,电子设备包括:
通信接口1,能够与其它设备比如网络设备等进行信息交互;
处理器2,与通信接口1连接,以实现与其它设备进行信息交互,用于运行计算机程序时,执行上述一个或多个技术方案提供的应用的访问方法。而所述计算机程序存储在存储器3上。
当然,实际应用时,电子设备中的各个组件通过总线***4耦合在一起。可理解,总线***4用于实现这些组件之间的连接通信。总线***4除包括数据总线之外,还包括电源总线、控制总线和状态信号总线。但是 为了清楚说明起见,在图5中将各种总线都标为总线***4。
本申请实施例中的存储器3用于存储各种类型的数据以支持电子设备的操作。这些数据的示例包括:用于在电子设备上操作的任何计算机程序。
可以理解,存储器3可以是易失性存储器或非易失性存储器,也可包括易失性和非易失性存储器两者。其中,非易失性存储器可以是只读存储器(ROM,Read Only Memory)、可编程只读存储器(PROM,Programmable Read-Only Memory)、可擦除可编程只读存储器(EPROM,Erasable Programmable Read-Only Memory)、电可擦除可编程只读存储器(EEPROM,Electrically Erasable Programmable Read-Only Memory)、磁性随机存取存储器(FRAM,ferromagnetic random access memory)、快闪存储器(Flash Memory)、磁表面存储器、光盘、或只读光盘(CD-ROM,Compact Disc Read-Only Memory);磁表面存储器可以是磁盘存储器或磁带存储器。易失性存储器可以是随机存取存储器(RAM,Random Access Memory),其用作外部高速缓存。通过示例性但不是限制性说明,许多形式的RAM可用,例如静态随机存取存储器(SRAM,Static Random Access Memory)、同步静态随机存取存储器(SSRAM,Synchronous Static Random Access Memory)、动态随机存取存储器(DRAM,Dynamic Random Access Memory)、同步动态随机存取存储器(SDRAM,Synchronous Dynamic Random Access Memory)、双倍数据速率同步动态随机存取存储器(DDRSDRAM,Double Data Rate Synchronous Dynamic Random Access Memory)、增强型同步动态随机存取存储器(ESDRAM,Enhanced Synchronous Dynamic Random Access Memory)、同步连接动态随机存取存储器(SLDRAM,SyncLink Dynamic Random Access Memory)、直接内存总线随机存取存储器(DRRAM,Direct Rambus Random Access Memory)。本申请实施例描述的存储器2旨在包括但不限于这些和任意其它适合类型的存储器。
上述本申请实施例揭示的方法可以应用于处理器2中,或者由处理器2实现。处理器2可能是一种集成电路芯片,具有信号的处理能力。在实现过程中,上述方法的各步骤可以通过处理器2中的硬件的集成逻辑电路 或者软件形式的指令完成。上述的处理器2可以是通用处理器、DSP,或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。处理器2可以实现或者执行本申请实施例中的公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者任何常规的处理器等。结合本申请实施例所公开的方法的步骤,可以直接体现为硬件译码处理器执行完成,或者用译码处理器中的硬件及软件模块组合执行完成。软件模块可以位于存储介质中,该存储介质位于存储器3,处理器2读取存储器3中的程序,结合其硬件完成前述方法的步骤。
处理器2执行所述程序时实现本申请实施例的各个方法中的相应流程,为了简洁,在此不再赘述。
在示例性实施例中,本申请实施例还提供了一种存储介质,即计算机存储介质,具体为计算机可读存储介质,例如包括存储计算机程序的存储器3,上述计算机程序可由处理器2执行,以完成前述方法所述步骤。计算机可读存储介质可以是FRAM、ROM、PROM、EPROM、EEPROM、Flash Memory、磁表面存储器、光盘、或CD-ROM等存储器。
本领域普通技术人员可以理解:实现上述方法实施例的全部或部分步骤可以通过程序指令相关的硬件来完成,前述的程序可以存储于一计算机可读取存储介质中,该程序在执行时,执行包括上述方法实施例的步骤;而前述的存储介质包括:移动存储设备、ROM、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
或者,本申请上述集成的单元如果以软件功能模块的形式实现并作为独立的产品销售或使用时,也可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请实施例的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台电子设备(可以是个人计算机、服务器、或者网络设备等)执行本申请各个实施例所述方法的全部或部分。而前述的存储介质包括:移动存储设备、ROM、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局 限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到变化或替换,都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以所述权利要求的保护范围为准。
Claims (10)
- 一种图像匹配方法,其特征在于,包括:获取双目图像,提取所述双目图像的边缘像素,并在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差;基于每个所述高置信度视差点的像素坐标和视差构成的三维空间对所有所述高置信度视差点进行三角剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向所述三角网格进行投影得到三角面;将每个所述三角面中三个顶点的视差最大值和最小值形成的区间确定为每个所述非高置信度视差点视差的视差搜索范围,在每个所述视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,确定每个所述深度边界点在目标方向上的视差边界点,并将所述深度边界点至所述视差边界点之间的像素的视差赋值为目标值;其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
- 根据权利要求1所述图像匹配方法,其特征在于,所述提取所述双目图像的边缘像素,包括:计算所述双目图像中每个像素的梯度值,以便得到所述双目图像对应的梯度图像,并对所述梯度图像进行滤波操作;对滤波操作后的梯度图像进行非最大抑制,以便确定边缘像素;对所有所述边缘像素进行水平方向和竖直方向预设像素范围邻域的搜索,若在所述预设像素范围邻域内存在所述边缘像素且直接相邻位置为非边缘像素,则将所述直接相邻位置设置为边缘像素,以便确定所述双目图像对应的所有边缘像素。
- 根据权利要求2所述图像匹配方法,其特征在于,对所述梯度图像进行滤波操作,包括:比较每个所述梯度值与预设邻域范围内的所有梯度值;将所述预设邻域范围内与所述梯度值大小相同方向不同的位置梯度值设置为0。
- 根据权利要求2所述图像匹配方法,其特征在于,所述对滤波操作后的梯度图像进行非最大抑制,以便确定边缘像素,包括:比较滤波操作后的每个目标梯度值与预设邻域范围内的所有梯度值;将大于所述预设邻域范围内的所有梯度值的目标梯度值对应的像素确定为所述边缘像素。
- 根据权利要求1所述图像匹配方法,其特征在于,在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差,包括:提取所述双目图像中每个像素的特征向量,对所有所述边缘像素在预设视差搜索范围内以邻域窗口内所有像素的特征向量进行匹配代价计算;将匹配代价满足预设条件的边缘像素确定为所述高置信度视差点;其中,所述预设条件包括最小匹配代价小于第一预设值且最小匹配代价与次小匹配代价的比值小于第二预设值;将最小匹配代价的视差确定为所述高置信度视差点的视差。
- 根据权利要求1所述图像匹配方法,其特征在于,所述根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,包括:对每个所述边缘像素在第一预设搜索范围内判断高置信度视差点的数量是否大于第三预设值;若是,则计算所述第一预设搜索范围内所有高置信度视差点的视差的统计值;其中,所述统计值包括均方差;将大于第三预设值的统计值对应的边缘像素确定为深度边界点。
- 根据权利要求1至6中任一项所述图像匹配方法,其特征在于,所述确定每个深度边界点在目标方向上的视差边界点,包括:对每个所述深度边界点沿目标方向进行第二预设搜索范围的搜索,确定每两个相邻像素点之间的视差差值;将大于第四预设值的视差差值对应的相邻像素点确定为所述第一像素点和所述第二像素点;将所述第一像素点和所述第二像素点的中间位置确定为所述视差边界点。
- 一种图像匹配装置,其特征在于,包括:第一确定模块,用于获取双目图像,提取所述双目图像的边缘像素,并在所述边缘像素中确定高置信度视差点和每个所述高置信度视差点的视差;投影模块,用于基于每个所述高置信度视差点的像素坐标和视差构成的三维空间对所有所述高置信度视差点进行三角剖分得到三角网格,并对每个非高置信度视差点沿平行于视差维度的方向向所述三角网格进行投影得到三角面;第二确定模块,用于将每个所述三角面中三个顶点的视差最大值和最小值形成的区间确定为每个所述非高置信度视差点视差的视差搜索范围,在每个所述视差搜索范围内计算所有视差对应的匹配代价,并将最小匹配代价的视差确定为对应的非高置信度视差点的视差;调整模块,用于根据所述高置信度视差点的分布在所述边缘像素中确定深度边界点,确定每个所述深度边界点在目标方向上的视差边界点,并将所述深度边界点至所述视差边界点之间的像素的视差调整为目标值;其中,所述目标值为第一视差值与第二视差值中接近所述深度边界点视差值的值,所述第一视差值为第一像素点的视差值,所述第二视差值为第二像素点的视差值,所述第一像素点与所述第二像素点为相邻像素点,所述视差边界点位于所述第一像素点与所述第二像素点之间。
- 一种电子设备,其特征在于,包括:存储器,用于存储计算机程序;处理器,用于执行所述计算机程序时实现如权利要求1至7任一项所述图像匹配方法的步骤。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有计算机程序,所述计算机程序被处理器执行时实现如权利要求1至7任一项所述图像匹配方法的步骤。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US18/012,584 US20230252664A1 (en) | 2020-09-08 | 2021-06-30 | Image Registration Method and Apparatus, Electronic Apparatus, and Storage Medium |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010934767.7A CN112116639B (zh) | 2020-09-08 | 2020-09-08 | 一种图像配准方法、装置及电子设备和存储介质 |
CN202010934767.7 | 2020-09-08 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022052582A1 true WO2022052582A1 (zh) | 2022-03-17 |
Family
ID=73802340
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/103359 WO2022052582A1 (zh) | 2020-09-08 | 2021-06-30 | 一种图像配准方法、装置及电子设备和存储介质 |
Country Status (3)
Country | Link |
---|---|
US (1) | US20230252664A1 (zh) |
CN (1) | CN112116639B (zh) |
WO (1) | WO2022052582A1 (zh) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115908170A (zh) * | 2022-11-04 | 2023-04-04 | 浙江华诺康科技有限公司 | 双目图像的降噪方法、装置、电子装置和存储介质 |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112116639B (zh) * | 2020-09-08 | 2022-06-07 | 苏州浪潮智能科技有限公司 | 一种图像配准方法、装置及电子设备和存储介质 |
CN114693546B (zh) * | 2022-02-25 | 2022-12-16 | 合肥的卢深视科技有限公司 | 图像去噪方法、装置、电子设备和计算机可读存储介质 |
CN115063467B (zh) * | 2022-08-08 | 2022-11-15 | 煤炭科学研究总院有限公司 | 煤矿井下高分辨率图像视差估计方法及装置 |
CN117808861A (zh) * | 2022-09-26 | 2024-04-02 | 神顶科技(南京)有限公司 | 一种双目视觉***的运行方法 |
CN117290537B (zh) * | 2023-09-28 | 2024-06-07 | 腾讯科技(深圳)有限公司 | 图像搜索方法、装置、设备及存储介质 |
CN117372647A (zh) * | 2023-10-26 | 2024-01-09 | 天宫开物(深圳)科技有限公司 | 一种建筑用三维模型的快速构建方法及*** |
CN117437563B (zh) * | 2023-12-13 | 2024-03-15 | 黑龙江惠达科技股份有限公司 | 一种基于双目视觉的植保无人机打点方法、装置及设备 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130083994A1 (en) * | 2011-09-29 | 2013-04-04 | Texas Instruments Incorporated | Semi-Global Stereo Correspondence Processing With Lossless Image Decomposition |
CN105160702A (zh) * | 2015-08-20 | 2015-12-16 | 武汉大学 | 基于LiDAR点云辅助的立体影像密集匹配方法及*** |
CN108122250A (zh) * | 2016-11-29 | 2018-06-05 | 三星电子株式会社 | 用于立体匹配的方法和设备 |
CN108564604A (zh) * | 2018-03-26 | 2018-09-21 | 北京邮电大学 | 基于平面约束和三角剖分的双目视觉立体匹配方法及装置 |
CN112116639A (zh) * | 2020-09-08 | 2020-12-22 | 苏州浪潮智能科技有限公司 | 一种图像配准方法、装置及电子设备和存储介质 |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2014096062A (ja) * | 2012-11-09 | 2014-05-22 | Yamaguchi Univ | 画像処理方法及び画像処理装置 |
CN106529495B (zh) * | 2016-11-24 | 2020-02-07 | 腾讯科技(深圳)有限公司 | 一种飞行器的障碍物检测方法和装置 |
CN109191512B (zh) * | 2018-07-27 | 2020-10-30 | 深圳市商汤科技有限公司 | 双目图像的深度估计方法及装置、设备、程序及介质 |
CN111210468B (zh) * | 2018-11-22 | 2023-07-11 | 中移(杭州)信息技术有限公司 | 一种图像深度信息获取方法及装置 |
-
2020
- 2020-09-08 CN CN202010934767.7A patent/CN112116639B/zh active Active
-
2021
- 2021-06-30 US US18/012,584 patent/US20230252664A1/en active Pending
- 2021-06-30 WO PCT/CN2021/103359 patent/WO2022052582A1/zh active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130083994A1 (en) * | 2011-09-29 | 2013-04-04 | Texas Instruments Incorporated | Semi-Global Stereo Correspondence Processing With Lossless Image Decomposition |
CN105160702A (zh) * | 2015-08-20 | 2015-12-16 | 武汉大学 | 基于LiDAR点云辅助的立体影像密集匹配方法及*** |
CN108122250A (zh) * | 2016-11-29 | 2018-06-05 | 三星电子株式会社 | 用于立体匹配的方法和设备 |
CN108564604A (zh) * | 2018-03-26 | 2018-09-21 | 北京邮电大学 | 基于平面约束和三角剖分的双目视觉立体匹配方法及装置 |
CN112116639A (zh) * | 2020-09-08 | 2020-12-22 | 苏州浪潮智能科技有限公司 | 一种图像配准方法、装置及电子设备和存储介质 |
Non-Patent Citations (2)
Title |
---|
JELLAL RADOUANE AIT; LANGE MANUEL; WASSERMANN BENJAMIN; SCHILLING ANDREAS; ZELL ANDREAS: "LS-ELAS: Line segment based efficient large scale stereo matching", 2017 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA), IEEE, 29 May 2017 (2017-05-29), pages 146 - 152, XP033126770, DOI: 10.1109/ICRA.2017.7989019 * |
LEE, SEONG-WHAN ; LI, STAN Z: " Advances in biometrics : international conference, ICB 2007, Seoul, Korea, August 27 - 29, 2007 ; proceedings", vol. 6492 Chap.3, 8 November 2010, SPRINGER , Berlin, Heidelberg , ISBN: 3540745491, article GEIGER ANDREAS; ROSER MARTIN; URTASUN RAQUEL: "Efficient Large-Scale Stereo Matching", pages: 25 - 38, XP047371684, 032548, DOI: 10.1007/978-3-642-19315-6_3 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115908170A (zh) * | 2022-11-04 | 2023-04-04 | 浙江华诺康科技有限公司 | 双目图像的降噪方法、装置、电子装置和存储介质 |
CN115908170B (zh) * | 2022-11-04 | 2023-11-21 | 浙江华诺康科技有限公司 | 双目图像的降噪方法、装置、电子装置和存储介质 |
Also Published As
Publication number | Publication date |
---|---|
US20230252664A1 (en) | 2023-08-10 |
CN112116639A (zh) | 2020-12-22 |
CN112116639B (zh) | 2022-06-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022052582A1 (zh) | 一种图像配准方法、装置及电子设备和存储介质 | |
US8447099B2 (en) | Forming 3D models using two images | |
US8452081B2 (en) | Forming 3D models using multiple images | |
US9774837B2 (en) | System for performing distortion correction and calibration using pattern projection, and method using the same | |
KR101532864B1 (ko) | 모바일 디바이스들에 대한 평면 맵핑 및 트래킹 | |
US9338437B2 (en) | Apparatus and method for reconstructing high density three-dimensional image | |
JP2019075082A (ja) | 深度値推定を用いた映像処理方法及び装置 | |
JP2018515825A (ja) | Lidarステレオ融合実写3dモデルバーチャルリアリティビデオ | |
US20160050372A1 (en) | Systems and methods for depth enhanced and content aware video stabilization | |
JP2015527764A (ja) | マルチ・フレーム画像キャリブレータ | |
KR20190030456A (ko) | 스테레오 매칭 방법 및 장치 | |
CN106952247B (zh) | 一种双摄像头终端及其图像处理方法和*** | |
CN111882655B (zh) | 三维重建的方法、装置、***、计算机设备和存储介质 | |
WO2021195939A1 (zh) | 一种双目拍摄装置的外参的标定方法、可移动平台及*** | |
CN109902675B (zh) | 物体的位姿获取方法、场景重构的方法和装置 | |
CN110825079A (zh) | 一种地图构建方法及装置 | |
JP2023021087A (ja) | 両眼画像マッチング方法、装置、デバイス、及び記憶媒体 | |
CN110443228B (zh) | 一种行人匹配方法、装置、电子设备及存储介质 | |
CN112270748B (zh) | 基于图像的三维重建方法及装置 | |
CN110148086B (zh) | 稀疏深度图的深度补齐方法、装置及三维重建方法、装置 | |
US20230419605A1 (en) | Map generation apparatus, map generation method, and non-transitory computer-readable medium storing program | |
US11120572B2 (en) | Method, system and apparatus for associating a target object in images | |
CN115661258A (zh) | 标定方法及装置、畸变校正方法及装置、存储介质、终端 | |
KR102288237B1 (ko) | 이미지 연관 방법, 시스템 및 장치 | |
Kim et al. | A high quality depth map upsampling method robust to misalignment of depth and color boundaries |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21865638 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21865638 Country of ref document: EP Kind code of ref document: A1 |