CN112884884A - Candidate region generation method and system - Google Patents
Candidate region generation method and system Download PDFInfo
- Publication number
- CN112884884A CN112884884A CN202110165577.8A CN202110165577A CN112884884A CN 112884884 A CN112884884 A CN 112884884A CN 202110165577 A CN202110165577 A CN 202110165577A CN 112884884 A CN112884884 A CN 112884884A
- Authority
- CN
- China
- Prior art keywords
- data
- similarity
- point cloud
- region
- candidate
- 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 60
- 238000000605 extraction Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 abstract description 9
- 238000001914 filtration Methods 0.000 description 12
- 230000011218 segmentation Effects 0.000 description 11
- 238000004364 calculation method Methods 0.000 description 10
- 238000005457 optimization Methods 0.000 description 8
- 238000001514 detection method Methods 0.000 description 7
- 239000013078 crystal Substances 0.000 description 6
- 230000001575 pathological effect Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 239000013598 vector Substances 0.000 description 6
- 238000013136 deep learning model Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000009499 grossing Methods 0.000 description 4
- 230000000877 morphologic effect Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000004044 response Effects 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000005034 decoration Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000009792 diffusion process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000000178 monomer Substances 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000004576 sand Substances 0.000 description 2
- 238000010187 selection method Methods 0.000 description 2
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000003287 optical 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
- G06T17/005—Tree description, e.g. octree, quadtree
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration using histogram techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a candidate region generation method and a candidate region generation system. Wherein, the method comprises the following steps: acquiring point cloud input data, and performing voxelization on the point cloud input data to obtain voxel data; extracting feature data according to the voxel data; generating region similarity according to the feature data; and combining the discrete regions according to the region similarity to obtain a candidate region. The invention solves the technical problem that the method based on the sliding cube in the prior art usually assumes that the ground objects are framed in the cube, and has poor treatment effect on the adhesion phenomenon of the ground objects.
Description
Technical Field
The invention relates to the field of target identification, in particular to a candidate region generation method and a candidate region generation system.
Background
Three-dimensional Target Recognition Technology (3D Target Recognition Technology) can rapidly, automatically, and accurately acquire a large amount of three-dimensional Point data of a Target object surface in Virtual Reality (VR) and Augmented Reality (AR), i.e., Point Clouds (Point Clouds). According to the method, ground point cloud segmentation is carried out on an input large scene according to a trained database, and finally, objects in the large scene are identified and modeled.
The three-dimensional target recognition technology is widely applied to the aspects of computer vision and virtual reality, and comprises the aspects of establishing an indoor three-dimensional space, quickly modeling an outdoor building, constructing a three-dimensional virtual sand table, establishing a three-dimensional virtual city and the like. Compared with the conventional two-dimensional recognition, the three-dimensional recognition can more intuitively and accurately present the result, and has higher observability and higher ornamental value.
However, after segmentation of the ground point cloud, non-ground points tend to have blocking phenomena, especially overlapping between different ground objects (such as buildings and trees). The result of the final presentation is that each individual cannot be singulated and two or more objects are considered to be one object. In the traditional method, a sliding cube is constructed by directly using prior information or a geometric feature matching method, local features or feature packets are extracted from point clouds in the cube, and finally the point clouds are classified by a classifier. The method almost performs exhaustive search on point cloud, and also needs to consider the influence of the scale and the rotation angle, so that the overall calculation is very time-consuming. In addition, the method based on the sliding cube generally assumes that the ground objects are framed inside the cube, the treatment effect on the adhesion phenomenon of the ground objects is poor, and the final presented result is not good.
After the traditional candidate target rapid detection method is used for ground point cloud segmentation, non-ground points often have adhesion phenomena, and particularly overlap among different ground objects (such as buildings and trees). If the prior information or the geometric feature matching method is directly used, a sliding cube is constructed, local features or feature packets are extracted from the point cloud in the cube, and finally the point cloud is classified through a classifier. The method almost performs exhaustive search on point clouds, and also needs to consider the influence of the scale and the rotation angle, so that the calculation is time-consuming. In addition, the method based on the sliding cube generally assumes that the ground objects are framed inside the cube, and the treatment effect on the adhesion phenomenon of the ground objects is poor.
In view of the above problems, no effective solution has been proposed.
Disclosure of Invention
The embodiment of the invention provides a candidate area generation method and a candidate area generation system, which are used for at least solving the technical problem that the method based on a sliding cube in the prior art usually assumes that a ground object is framed in the cube, and the processing effect of the method on the adhesion phenomenon of the ground object is poor.
According to an aspect of the embodiments of the present invention, there is provided a candidate region generation method, including: acquiring point cloud input data, and performing voxelization on the point cloud input data to obtain voxel data; extracting feature data according to the voxel data; generating region similarity according to the feature data; and combining the discrete regions according to the region similarity to obtain a candidate region.
Optionally, the extracting feature data according to the voxel data includes: generating a color similarity from the voxel data; and extracting the feature data according to the color similarity.
Optionally, before generating the region similarity according to the feature data, the method further includes: obtaining characteristic frequency according to the characteristic data; and obtaining texture similarity and form similarity according to the characteristic frequency.
Optionally, the candidate region includes: and (5) dividing the point cloud scene.
According to another aspect of the embodiments of the present invention, there is also provided a candidate region generation system, including: the acquisition module is used for acquiring point cloud input data and carrying out voxelization on the point cloud input data to obtain voxelization data; the extraction module is used for extracting characteristic data according to the voxel data; the generating module is used for generating the region similarity according to the characteristic data; and the merging module is used for merging the discrete regions according to the region similarity to obtain the candidate regions.
Optionally, the extracting module includes: a generating unit for generating a color similarity from the voxel data; and the extraction unit is used for extracting the feature data according to the color similarity.
Optionally, the system further includes: the frequency module is used for obtaining characteristic frequency according to the characteristic data; and the similarity module is used for obtaining texture similarity and form similarity according to the characteristic frequency.
Optionally, the candidate region includes: and (5) dividing the point cloud scene.
According to another aspect of the embodiments of the present invention, there is also provided a nonvolatile storage medium including a stored program, wherein the program controls a device in which the nonvolatile storage medium is located to execute a candidate region generation method when running.
According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including a processor and a memory; the memory has stored therein computer readable instructions for execution by the processor, wherein the computer readable instructions when executed perform a method of candidate region generation.
In the embodiment of the invention, point cloud input data is obtained and subjected to voxelization to obtain voxel data; extracting feature data according to the voxel data; generating region similarity according to the feature data; the method for combining the discrete regions according to the region similarity to obtain the candidate regions solves the technical problem that the method based on the sliding cube in the prior art usually assumes that the ground objects are framed in the cube, and has poor treatment effect on the adhesion phenomenon of the ground objects.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:
FIG. 1 is a flow chart of a candidate region generation method according to an embodiment of the present invention;
fig. 2 is a block diagram of a candidate area generation system according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
In accordance with an embodiment of the present invention, there is provided a method embodiment of a candidate area generation method, it should be noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer executable instructions, and that while a logical order is illustrated in the flowchart, in some cases the steps illustrated or described may be performed in an order different than here.
Example one
Fig. 1 is a flowchart of a candidate region generation method according to an embodiment of the present invention, as shown in fig. 1, the method includes the following steps:
step S102, point cloud input data are obtained, and voxel is carried out on the point cloud input data to obtain voxel data.
And step S104, extracting characteristic data according to the voxel data.
And step S106, generating the region similarity according to the feature data.
And step S108, merging the discrete regions according to the region similarity to obtain candidate regions.
Optionally, the extracting feature data according to the voxel data includes: generating a color similarity from the voxel data; and extracting the feature data according to the color similarity.
Optionally, before generating the region similarity according to the feature data, the method further includes: obtaining characteristic frequency according to the characteristic data; and obtaining texture similarity and form similarity according to the characteristic frequency.
Optionally, the candidate region includes: and (5) dividing the point cloud scene.
Specifically, the embodiment of the present invention belongs to the field of image processing, and a research on rapid detection of candidate targets has been greatly advanced in recent years, and particularly, a Selective Search Strategy (SS) can greatly reduce the Search range of candidate targets by sufficiently utilizing texture distribution of targets and by extracting and analyzing regional rapid features. Therefore, on the basis of non-ground point cloud segmentation, the invention designs a rapid detection method of candidate ground objects aiming at the characteristics of point cloud distribution, and the method is used as the input of a deep learning model. The method can make the selected objects of different ground features more robust, and has consistency and high efficiency.
In addition, the algorithm for quickly generating the ground feature candidate region based on the hierarchical merged tree provided by the embodiment of the invention comprises the following steps:
step 101: the input point cloud is divided by using an octree to realize voxelization, and then regional growth crystal nuclei are uniformly distributed in a voxel space.
Step 102: and (4) bringing the square with the minimum distance function value of the point cloud superpixel into the crystal nucleus, and finally finishing the growth of all the crystal nuclei almost at the same time.
Step 103: the original point cloud may be divided into a plurality of different regions based on step 102, and the nuclei in each region are guaranteed to be similar.
Step 104: r, G, B color histograms of all corresponding positions of every two adjacent point cloud regions are respectively calculated, wherein the horizontal axis is the color intensity value of R, G, B, and the vertical axis is the proportion of the intensity value in the whole body.
Step 105: calculating the L2 norm of the corresponding color intensity value of every two adjacent point cloud areas in the R, G, B color histogram to obtain the color similarity, namely the color similarity
Step 106: and (4) carrying out FPFH (field programmable gate hydrographic) feature extraction on the large-scale point cloud. For the characteristics of each point and 8 neighboring points, the maximum expectation is obtained by calculation using a Gaussian mixture model. And using the obtained 9 characteristic sequences as the characteristic codebook of the point.
Step 107: and counting the frequency of each feature in the codebook to obtain the morphological strength. And then calculating the average value of the vectors with smaller intensity difference of the same type features of the two regions, thereby obtaining the morphological similarity.
Step 108: and extracting LBP texture characteristics of each point of the candidate area. The nearest k neighbor points are computed for each point. If the pixel of the central point is higher than the neighbor point, the neighbor point is assigned 1, otherwise, the neighbor point is assigned 0.
Step 109: each point has a string of binary features of length k. And then can count each pointIntensity information. The texture similarity S can be obtained by calculating the difference value of the sum of the intensity information of the candidate regionst(Ri，Rj)。
Step 110: based on step 104 and 109, the color similarity S of the two regions is determinedcMorphological similarity SsAnd the texture similarity StMultiplying the weight coefficients alpha, beta and lambda respectively to obtain the regional comprehensive similarity: s (R)i，Rj)＝αSc(Ri，Rj)+βSs(Ri，Rj)+γSt(Ri,Rj)。
Step 111: and searching two most similar areas according to the comprehensive similarity of the areas, combining the two areas, and calculating a histogram of the combined area to obtain a new vector. The calculation method comprises the following steps: respectively solving norms of two old regions, which are expressed as | | xi||2And | | xj||2Then multiplied by the vectors respectively
And
finally, the result is obtained by adding and dividing the two regions by the value of integral norm of the two regions
Step 112: deleting the old region and the corresponding similarity, and recalculating the similarity of the newly generated region and the adjacent region after merging.
Step 113: and respectively calculating the hyper-voxel distance value and the reciprocal of the comprehensive similarity of the regions for each region, and performing weighted summation on the hyper-voxel distance value and the reciprocal of the comprehensive similarity of the regions to obtain the response value of each region.
Step 114: and starting from the whole situation based on all the response values, constructing a plurality of minimum spanning trees by adopting a greedy strategy so as to obtain a plurality of minimum weight paths.
Step 115: and merging the smallest two paths, namely merging the two most similar area blocks.
Step 116: the step 104 and the step 115 are repeatedly executed for all the areas to be merged, so that the merging of the discrete areas can be realized, and finally the point cloud scene after segmentation is formed.
In order to better realize the monomer segmentation of objects in a large scene, the project designs a rapid detection method of candidate ground objects aiming at the characteristics of point cloud distribution on the basis of non-ground point cloud segmentation, and the method is used as the input of a deep learning model to classify targets in virtual reality. Compared with a two-dimensional image, the three-dimensional point cloud has large difference in feature forms, and the point cloud distribution has irregular characteristics, so that the candidate feature selection method has the following three characteristics: (1) and (4) dimension adaptability. Namely, the device is robust to ground objects with different sizes; (2) and (5) consistency. The differences of the color, the texture and the shape of the interior of the candidate ground object point cloud cannot be too large; (3) high efficiency. The point cloud data volume is large, and the candidate set is only a basis for subsequent classification and should not consume too much computing resources.
It should be noted that curvature filtering is an optimization algorithm in image processing, and it first appeared in the sixth chapter of doctor's paper from Gong Yuan Hao doctor (ETH E-Collection: Spectraily regulated surfaces). Whether denoising and smoothing problems in two-dimensional images or in three-dimensional point clouds, are usually pathological, and the pathological problems require regularization terms. Curvature regularization is a commonly used regularization term for pathological problems, and the obtained models are generally good, but the models are also difficult to solve. The traditional solving method has two types: one based on the gradient descent method (diffusion equation) and the other based on the Euler Lagrange equation. The latter solution is generally more efficient than the former, but how to obtain the equation is generally very complex, and the resulting equation is difficult to see its corresponding physical meaning. The curvature filtering considers the optimization problem from another point of view, which is a kind of filtering but optimizes a certain regular term, and the known surface of differential geometry is implicitly used in the filtering process, so that the gaussian curvature or the mean curvature does not need to be calculated, and the complexity of calculation is reduced. The advantages of curvature filtering are: high efficiency, one hundred to one thousand times faster than the traditional method; generality, an arbitrarily complex noise model can be solved; the theoretical guarantee is based on the classical differential set theory; easy to implement and parallel.
The invention relates to a terrain candidate region fast generation algorithm based on a hierarchical merged tree, which is innovated and popularized based on the idea of curvature filtering. According to the point cloud distortion correction method, optimization of a regular term or optimization based on a Euler Lagrange equation is not needed, each point of the point cloud data is updated according to a specified rule, the calculation complexity can be reduced, and the effect and efficiency of smoothing the point cloud data are improved.
Through the embodiment, the technical problem that the method based on the sliding cube in the prior art usually assumes that the ground objects are framed inside the cube, and the treatment effect on the adhesion phenomenon of the ground objects is poor is solved.
Example two
Fig. 2 is a block diagram of a candidate area generation system according to an embodiment of the present invention, and as shown in fig. 2, the system includes:
the acquisition module 20 is configured to acquire point cloud input data, and perform voxelization on the point cloud input data to obtain voxelization data.
And an extraction module 22, configured to extract feature data according to the voxel data.
And a generating module 24, configured to generate the region similarity according to the feature data.
And a merging module 26, configured to merge the discrete regions according to the region similarity to obtain a candidate region.
Optionally, the extracting module includes: a generating unit for generating a color similarity from the voxel data; and the extraction unit is used for extracting the feature data according to the color similarity.
Optionally, the system further includes: the frequency module is used for obtaining characteristic frequency according to the characteristic data; and the similarity module is used for obtaining texture similarity and form similarity according to the characteristic frequency.
Optionally, the candidate region includes: and (5) dividing the point cloud scene.
Specifically, the embodiment of the present invention belongs to the field of image processing, and a research on rapid detection of candidate targets has been greatly advanced in recent years, and particularly, a Selective Search Strategy (SS) can greatly reduce the Search range of candidate targets by sufficiently utilizing texture distribution of targets and by extracting and analyzing regional rapid features. Therefore, on the basis of non-ground point cloud segmentation, the invention designs a rapid detection method of candidate ground objects aiming at the characteristics of point cloud distribution, and the method is used as the input of a deep learning model. The method can make the selected objects of different ground features more robust, and has consistency and high efficiency.
In addition, the algorithm for quickly generating the ground feature candidate region based on the hierarchical merged tree provided by the embodiment of the invention comprises the following steps:
step 101: the input point cloud is divided by using an octree to realize voxelization, and then regional growth crystal nuclei are uniformly distributed in a voxel space.
Step 102: and (4) bringing the square with the minimum distance function value of the point cloud superpixel into the crystal nucleus, and finally finishing the growth of all the crystal nuclei almost at the same time.
Step 103: the original point cloud may be divided into a plurality of different regions based on step 102, and the nuclei in each region are guaranteed to be similar.
Step 104: r, G, B color histograms of all corresponding positions of every two adjacent point cloud regions are respectively calculated, wherein the horizontal axis is the color intensity value of R, G, B, and the vertical axis is the proportion of the intensity value in the whole body.
Step 105: calculating the L2 norm of the corresponding color intensity value of every two adjacent point cloud areas in the R, G, B color histogram to obtain the color similarity, namely the color similarity
Step 106: and (4) carrying out FPFH (field programmable gate hydrographic) feature extraction on the large-scale point cloud. For the characteristics of each point and 8 neighboring points, the maximum expectation is obtained by calculation using a Gaussian mixture model. And using the obtained 9 characteristic sequences as the characteristic codebook of the point.
Step 107: and counting the frequency of each feature in the codebook to obtain the morphological strength. And then calculating the average value of the vectors with smaller intensity difference of the same type features of the two regions, thereby obtaining the morphological similarity.
Step 108: and extracting LBP texture characteristics of each point of the candidate area. The nearest k neighbor points are computed for each point. If the pixel of the central point is higher than the neighbor point, the neighbor point is assigned 1, otherwise, the neighbor point is assigned 0.
Step 109: each point has a string of binary features of length k. And then the intensity information of each point can be counted. The texture similarity S can be obtained by calculating the difference value of the sum of the intensity information of the candidate regionst(Ri，Rj)。
Step 110: based on step 104 and 109, the color similarity S of the two regions is determinedcMorphological similarity SsAnd the texture similarity StMultiplying the weight coefficients alpha, beta and lambda respectively to obtain the regional comprehensive similarity: s (R)i，Rj)＝αSc(Ri，Rj)+βSs(Ri，Rj)+γSt(Ri，Rj)。
Step 111: and searching two most similar areas according to the comprehensive similarity of the areas, combining the two areas, and calculating a histogram of the combined area to obtain a new vector. The calculation method comprises the following steps: respectively solving norms of two old regions, which are expressed as | | xi||2And | | xj||2Then multiplied by the vectors respectively
And
finally, the result is obtained by adding and dividing the two regions by the value of integral norm of the two regions
Step 112: deleting the old region and the corresponding similarity, and recalculating the similarity of the newly generated region and the adjacent region after merging.
Step 113: and respectively calculating the hyper-voxel distance value and the reciprocal of the comprehensive similarity of the regions for each region, and performing weighted summation on the hyper-voxel distance value and the reciprocal of the comprehensive similarity of the regions to obtain the response value of each region.
Step 114: and starting from the whole situation based on all the response values, constructing a plurality of minimum spanning trees by adopting a greedy strategy so as to obtain a plurality of minimum weight paths.
Step 115: and merging the smallest two paths, namely merging the two most similar area blocks.
Step 116: the step 104 and the step 115 are repeatedly executed for all the areas to be merged, so that the merging of the discrete areas can be realized, and finally the point cloud scene after segmentation is formed.
In order to better realize the monomer segmentation of objects in a large scene, the project designs a rapid detection method of candidate ground objects aiming at the characteristics of point cloud distribution on the basis of non-ground point cloud segmentation, and the method is used as the input of a deep learning model to classify targets in virtual reality. Compared with a two-dimensional image, the three-dimensional point cloud has large difference in feature forms, and the point cloud distribution has irregular characteristics, so that the candidate feature selection method has the following three characteristics: (1) and (4) dimension adaptability. Namely, the device is robust to ground objects with different sizes; (2) and (5) consistency. The differences of the color, the texture and the shape of the interior of the candidate ground object point cloud cannot be too large; (3) high efficiency. The point cloud data volume is large, and the candidate set is only a basis for subsequent classification and should not consume too much computing resources.
It should be noted that curvature filtering is an optimization algorithm in image processing, and it first appeared in the sixth chapter of doctor's paper from Gong Yuan Hao doctor (ETH E-Collection: Spectraily regulated surfaces). Whether denoising and smoothing problems in two-dimensional images or in three-dimensional point clouds, are usually pathological, and the pathological problems require regularization terms. Curvature regularization is a commonly used regularization term for pathological problems, and the obtained models are generally good, but the models are also difficult to solve. The traditional solving method has two types: one based on the gradient descent method (diffusion equation) and the other based on the Euler Lagrange equation. The latter solution is generally more efficient than the former, but how to obtain the equation is generally very complex, and the resulting equation is difficult to see its corresponding physical meaning. The curvature filtering considers the optimization problem from another point of view, which is a kind of filtering but optimizes a certain regular term, and the known surface of differential geometry is implicitly used in the filtering process, so that the gaussian curvature or the mean curvature does not need to be calculated, and the complexity of calculation is reduced. The advantages of curvature filtering are: high efficiency, one hundred to one thousand times faster than the traditional method; generality, an arbitrarily complex noise model can be solved; the theoretical guarantee is based on the classical differential set theory; easy to implement and parallel.
The invention relates to a terrain candidate region fast generation algorithm based on a hierarchical merged tree, which is innovated and popularized based on the idea of curvature filtering. According to the point cloud distortion correction method, optimization of a regular term or optimization based on a Euler Lagrange equation is not needed, each point of the point cloud data is updated according to a specified rule, the calculation complexity can be reduced, and the effect and efficiency of smoothing the point cloud data are improved.
According to another aspect of the embodiments of the present invention, there is also provided a nonvolatile storage medium including a stored program, wherein the program controls a device in which the nonvolatile storage medium is located to execute a candidate region generation method when running.
According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including a processor and a memory; the memory has stored therein computer readable instructions for execution by the processor, wherein the computer readable instructions when executed perform a method of candidate region generation.
Through the embodiment, the technical problem that the method based on the sliding cube in the prior art usually assumes that the ground objects are framed inside the cube, and the treatment effect on the adhesion phenomenon of the ground objects is poor is solved.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.
Claims (10)
1. A candidate region generation method, comprising:
acquiring point cloud input data, and performing voxelization on the point cloud input data to obtain voxel data;
extracting feature data according to the voxel data;
generating region similarity according to the feature data;
and combining the discrete regions according to the region similarity to obtain a candidate region.
2. The method of claim 1, wherein said extracting feature data from said voxel data comprises:
generating a color similarity from the voxel data;
and extracting the feature data according to the color similarity.
3. The method of claim 2, wherein prior to said generating region similarities from said feature data, said method further comprises:
obtaining characteristic frequency according to the characteristic data;
and obtaining texture similarity and form similarity according to the characteristic frequency.
4. The method of claim 1, wherein the candidate regions comprise: and (5) dividing the point cloud scene.
5. A candidate region generation system, comprising:
the acquisition module is used for acquiring point cloud input data and carrying out voxelization on the point cloud input data to obtain voxelization data;
the extraction module is used for extracting characteristic data according to the voxel data;
the generating module is used for generating the region similarity according to the characteristic data;
and the merging module is used for merging the discrete regions according to the region similarity to obtain the candidate regions.
6. The system of claim 5, wherein the extraction module comprises:
a generating unit for generating a color similarity from the voxel data;
and the extraction unit is used for extracting the feature data according to the color similarity.
7. The method of claim 6, wherein the system further comprises:
the frequency module is used for obtaining characteristic frequency according to the characteristic data;
and the similarity module is used for obtaining texture similarity and form similarity according to the characteristic frequency.
8. The system of claim 5, wherein the candidate regions comprise: and (5) dividing the point cloud scene.
9. A non-volatile storage medium, comprising a stored program, wherein the program, when executed, controls an apparatus in which the non-volatile storage medium is located to perform the method of any one of claims 1 to 4.
10. An electronic device comprising a processor and a memory; the memory has stored therein computer readable instructions for execution by the processor, wherein the computer readable instructions when executed perform the method of any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110165577.8A CN112884884A (en) | 2021-02-06 | 2021-02-06 | Candidate region generation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110165577.8A CN112884884A (en) | 2021-02-06 | 2021-02-06 | Candidate region generation method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112884884A true CN112884884A (en) | 2021-06-01 |
Family
ID=76055955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110165577.8A Pending CN112884884A (en) | 2021-02-06 | 2021-02-06 | Candidate region generation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112884884A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113826598A (en) * | 2021-09-10 | 2021-12-24 | 宁波权智科技有限公司 | Unmanned aerial vehicle pesticide spreading method and device based on neural network |
-
2021
- 2021-02-06 CN CN202110165577.8A patent/CN112884884A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113826598A (en) * | 2021-09-10 | 2021-12-24 | 宁波权智科技有限公司 | Unmanned aerial vehicle pesticide spreading method and device based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shen et al. | Real-time superpixel segmentation by DBSCAN clustering algorithm | |
| CN107862698B (en) | Light field foreground segmentation method and device based on K mean cluster | |
| US9483709B2 (en) | Visual saliency estimation for images and video | |
| CN108537239B (en) | Method for detecting image saliency target | |
| CN105761238B (en) | A method of passing through gray-scale statistical data depth information extraction well-marked target | |
| WO2019071976A1 (en) | Panoramic image saliency detection method based on regional growth and eye movement model | |
| CN108596919B (en) | Automatic image segmentation method based on depth map | |
| Arbelot et al. | Local texture-based color transfer and colorization | |
| CN110188763B (en) | Image significance detection method based on improved graph model | |
| Pound et al. | A patch-based approach to 3D plant shoot phenotyping | |
| Pan et al. | Automatic segmentation of point clouds from multi-view reconstruction using graph-cut | |
| CN103971377A (en) | Building extraction method based on prior shape level set segmentation | |
| Zhao et al. | A saliency detection based method for 3d surface simplification | |
| CN110930503A (en) | Method and system for establishing three-dimensional model of clothing, storage medium and electronic equipment | |
| Li et al. | Unsupervised road extraction via a Gaussian mixture model with object-based features | |
| Yun et al. | Supervoxel-based saliency detection for large-scale colored 3D point clouds | |
| Kumar et al. | Automatic image segmentation using wavelets | |
| CN113409332B (en) | Building plane segmentation method based on three-dimensional point cloud | |
| CN116993947B (en) | Visual display method and system for three-dimensional scene | |
| Zhao et al. | Extended non-local means filter for surface saliency detection | |
| CN112884884A (en) | Candidate region generation method and system | |
| Fayolle et al. | Segmentation of discrete point clouds using an extensible set of templates | |
| Elashry et al. | Feature matching enhancement using the graph neural network (gnn-ransac) | |
| CN107085725B (en) | Method for clustering image areas through LLC based on self-adaptive codebook | |
| Kim et al. | Data-driven single image depth estimation using weighted median statistics |
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 |