CN116342666B - Three-dimensional point cloud registration method based on multi-form optimization and electronic equipment - Google Patents
Three-dimensional point cloud registration method based on multi-form optimization and electronic equipment Download PDFInfo
- Publication number
- CN116342666B CN116342666B CN202310101088.5A CN202310101088A CN116342666B CN 116342666 B CN116342666 B CN 116342666B CN 202310101088 A CN202310101088 A CN 202310101088A CN 116342666 B CN116342666 B CN 116342666B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- registration
- task
- registration task
- robust
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 88
- 238000005457 optimization Methods 0.000 title claims abstract description 30
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 34
- 230000008569 process Effects 0.000 claims abstract description 32
- 238000012546 transfer Methods 0.000 claims abstract description 22
- 230000008859 change Effects 0.000 claims abstract description 20
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 15
- 238000011156 evaluation Methods 0.000 claims description 44
- 230000006870 function Effects 0.000 claims description 43
- 238000005070 sampling Methods 0.000 claims description 25
- 238000004891 communication Methods 0.000 claims description 15
- 238000013519 translation Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 5
- 230000000875 corresponding effect Effects 0.000 description 25
- 238000010586 diagram Methods 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 9
- 239000013598 vector Substances 0.000 description 7
- 230000009466 transformation Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000013011 mating Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 238000012854 evaluation process Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000002079 cooperative effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 241000384062 Armadillo Species 0.000 description 1
- 102000016904 Armadillo Domain Proteins Human genes 0.000 description 1
- 108010014223 Armadillo Domain Proteins Proteins 0.000 description 1
- 241000283973 Oryctolagus cuniculus Species 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 102000004169 proteins and genes Human genes 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Genetics & Genomics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a three-dimensional point cloud registration method and electronic equipment based on multi-form optimization, wherein the method comprises the following steps: acquiring a point cloud pair to be registered of a scene to obtain a corresponding farthest point set; dividing the three-dimensional point cloud registration problem into a robust registration task and an accurate registration task, solving the three-dimensional point cloud registration problem based on the robust registration task and the accurate registration task by utilizing a multi-task evolution algorithm, and taking the solution of the accurate registration task as an optimal scene point cloud change parameter; the information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy; and registering each point in the furthest point set according to the optimal scene point cloud change parameters so as to realize three-dimensional point cloud registration. The registration method can ensure robustness and accuracy at the same time.
Description
Technical Field
The invention belongs to the technical fields of computer vision, computer graphics and point cloud processing, and particularly relates to a three-dimensional point cloud registration method based on multi-form optimization and electronic equipment.
Background
With the development of three-dimensional scanning technology, three-dimensional point clouds have been widely used in the fields of automatic driving, architectural design, digital animation production, bioinformatics, medical treatment, and the like. These point clouds may be acquired from different perspectives, at different times, using different platforms, or by multi-mode sensors. Since the sensor can only capture scans over its limited field of view, a registration algorithm is required to generate a complete scene or object. The goal of the registration is to find the correspondence between two point clouds or to establish a transformation matrix from one point cloud to the other. By applying the transformation matrix, partial scans of the same scene or object can be merged into one complete point cloud. The point cloud registration provides basic technical support for point cloud positioning, city planning, change detection, automatic driving and the like.
Current research on point cloud registration problems generally calculates transformation parameters by finding correspondence between points, lines, planes, etc. They can be further subdivided into feature-based and evolutionary algorithm-based registration methods: and the feature-based algorithm obtains the matched feature correspondence between the two point clouds through the local descriptor. The feature descriptors are used to encode the spatial information of the local patches into compact feature vectors for the purpose of mutually identifying the keypoints, and various matching strategies are used to calculate the corresponding geometric features. Such as: (1) Article "Fast descriptors and correspondence propagation for robust global point cloud registration (Lei H, jiang G, quan l.fast descriptors and correspondence propagation for robust global point cloud registration.ieee Transactions on Image processing.2017may3;26 (8): 3614-23.)" introduced a mechanism named corresponding propagation: matching and aggregating each seed characteristic into a group of large number of matches; computing a plurality of transitions between the point clouds by the matching sets; the quality function formulated according to the distance error is used for identifying the optimal transformation and completing the alignment of the point cloud; (2) "Shi Fengbo, cao Qin, wei Jun. Feature point based curved surface point cloud registration method [ J ]. Beijing mapping, 2022,36 (10): 1345-1349. Evolutionary algorithm based alignment method uses global search strategy to find optimal transformations in solution space, and does not require accurate estimation of initial information of the point cloud, obtains initial matching relationship by computing hausdorff distance of feature point curvature, and then extracts accurate feature point pairs by utilizing shape features of the point cloud local surface curve; the article "A point cloud registration algorithm based on normal vector and particle swarm optimization (Zhan X, cai Y, li H, li Y, he p.a point cloud registration algorithm based on normal vector and particle swarm optimization.measurement and control.2020mar;53 (3-4): 265-75.)" proposes a new method for point cloud registration using genetic algorithms but which searches in a seven parameter space consisting of three translation parameters, three rotation parameters and one surface overlap parameter, assuming that the overlap information is sufficient to pre-align the 3D oriented surfaces.
However, a disadvantage of the above proposed feature-based registration method is that the registration performance is sensitive to the quality of the features; noise and defective parts in the point cloud inevitably reduce the accuracy and robustness of the feature; for point clouds with larger volumes, the huge calculation amount of feature extraction also affects the practicability of the method; incomplete data caused by shielding in a complex urban environment can cause changes in the details of the point cloud, so that corresponding feature pairs cannot be found.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a three-dimensional point cloud registration method based on multi-form optimization and electronic equipment. The technical problems to be solved by the invention are realized by the following technical scheme:
in a first aspect, an embodiment of the present invention provides a three-dimensional point cloud registration method based on multi-form optimization, including:
acquiring a point cloud pair to be registered of a scene to obtain a corresponding farthest point set;
dividing the three-dimensional point cloud registration problem into a robust registration task and an accurate registration task, solving the three-dimensional point cloud registration problem based on the robust registration task and the accurate registration task by utilizing a multi-task evolution algorithm, and taking the solution of the accurate registration task as an optimal scene point cloud change parameter; the information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy;
and registering each point in the furthest point set according to the optimal scene point cloud change parameters so as to realize three-dimensional point cloud registration.
In one embodiment of the present invention, acquiring a pair of point clouds to be registered of a scene to obtain a corresponding farthest point set, includes, for each point cloud to be registered:
randomly sampling a first sampling point in the point cloud;
the Euclidean distance between the first sampling point and other points in the point cloud is calculated, and the Euclidean distance is stored in a distance array;
selecting the maximum point in all Euclidean distances as the next sampling point;
and continuously calculating Euclidean distance between the sampling point and other points in the point cloud, judging whether the new Euclidean distance is smaller than the Euclidean distance in the corresponding distance array, and if so, updating the distance array until the sampling point with the expected sampling number is traversed to obtain the farthest point set.
In one embodiment of the invention, solving a three-dimensional point cloud registration problem based on a robust registration task and an accurate registration task using a multitasking evolutionary algorithm comprises:
setting parameters of a population in a multitask evolution algorithm, and initializing the population;
encoding each individual in the initialized population into a unified rotation and translation space and assigning a skill factor representative of its task attributes of skill;
in the whole evolution process, information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy so as to obtain a new generation population;
decoding each individual in the new generation population to obtain a corresponding scene point cloud change parameter and a corresponding skill factor thereof;
judging task attributes of each individual in the new generation of population according to the decoded skill factors, and selecting a corresponding fitness evaluation function according to the task attributes of each individual in the new generation of population for evaluation so as to solve the three-dimensional point cloud registration problem.
In one embodiment of the invention, each individual in the initialized population is a vector of randomly generated six-dimensional random variables R 1 ,R 2 ,R 3 ,t 1 ,t 2 ,t 3 ]Wherein R is 1 、R 2 、R 3 The rotation parameters of the individual in x, y and z coordinate axes, t 1 、t 2 、t 3 The translation parameters of the individual in the x, y and z coordinate axes are respectively shown.
In one embodiment of the invention, information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy, which comprises the following steps:
the first stage, the information flow is from the robust registration task to the accurate registration task, and the robust registration task occupies a large proportion of evaluation processing at the first stage;
the second stage, the information flow is from the accurate registration task to the robust registration task, and the accurate registration task occupies a large proportion of evaluation processing at the stage.
In one embodiment of the present invention, selecting a corresponding fitness evaluation function according to the task attribute of each individual in the new generation population for evaluation to solve the three-dimensional point cloud registration problem includes:
if the task attribute is a robust registration task, constructing a scale self-adaptive cauchy weighted estimator, and reconstructing an adaptability evaluation function of the robust registration task according to the constructed scale self-adaptive cauchy weighted estimator;
if the task attribute is an accurate registration task, constructing an adaptability evaluation function of the accurate registration task;
and respectively evaluating according to the constructed fitness evaluation function to solve the three-dimensional point cloud registration problem.
In one embodiment of the invention, the constructed scale-adaptive cauchy-weight estimator is expressed as:
wherein ρ (·) represents a scale-adaptive cauchy-weighted estimator, e i ＝||Rx i +t-y i || 2 Representation point (x) i ,y i ) K represents the scale parameter, r= [ R ] 1 ,R 2 ,R 3 ]，t＝[t 1 ,t 2 ,t 3 ]，||·|| 2 Representation calculation 2 Norms.
In one embodiment of the invention, the fitness evaluation function of the robust registration task is expressed as:
wherein f 1 (. Cndot.) represents the fitness evaluation function of the robust registration task, τ represents a fixed parameter, N represents the number of sampling points in the point cloud, and M represents the number of inliers in N.
In one embodiment of the invention, the fitness evaluation function of the exact registration task is expressed as:
wherein f 2 (-) represents the fitness evaluation function of the exact registration task, alpha represents an adjustment parameter and is used to adjust the adjustment parameter, I.I p Representation calculation p Norms.
In a second aspect, an embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete communication with each other through the communication bus;
the memory is used for storing a computer program;
and the processor is used for realizing the steps of any of the three-dimensional point cloud registration methods based on the multi-form optimization when executing the program stored on the memory.
The invention has the beneficial effects that:
the three-dimensional point cloud registration method based on the multi-form optimization, which is provided by the invention, can effectively solve the problems that the traditional point cloud registration method based on evolution is poor in robustness and easy to be trapped into local optimization, is a novel multi-form optimization registration strategy, and comprises the following configuration processes: acquiring a point cloud pair to be registered of a scene to obtain a corresponding farthest point set; dividing the three-dimensional point cloud registration problem into a robust registration task and an accurate registration task, solving the three-dimensional point cloud registration problem based on the robust registration task and the accurate registration task by utilizing a multi-task evolution algorithm, and taking the solution of the accurate registration task as an optimal scene point cloud change parameter; the information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy; and registering each point in the furthest point set according to the optimal scene point cloud change parameters so as to realize three-dimensional point cloud registration. Therefore, the point cloud registration problem is solved through a multi-task evolution algorithm, a two-stage bidirectional knowledge transfer strategy is constructed in the evolution process, robustness and accuracy are considered in two independent stages, so that the registration result can ensure both robustness and accuracy, and an adaptability evaluation function is newly designed in the realization of a robust registration task, so that the registration robustness is greatly improved; meanwhile, the situation of local optimization of the algorithm is reduced through the cooperative action between a robust registration task and an accurate registration task in the multi-task evolution algorithm in the whole process, so that the accuracy of registration is greatly improved. The registration method provided by the invention has robustness to noise, abnormal values and partial overlapping conditions, namely the sensitivity to the characteristic quality is reduced in the registration process, and the registration method can be applied to a plurality of real registration scenes, such as object registration, scene reconstruction, simultaneous positioning and mapping, and the like.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a schematic flow chart of a three-dimensional point cloud registration method based on multi-form optimization provided by an embodiment of the invention;
FIG. 2 is a schematic diagram of a result of acquiring the farthest point in the scene point cloud according to the embodiment of the present invention;
FIG. 3 is a schematic flow chart for solving the three-dimensional point cloud registration problem by using a multi-task evolutionary algorithm according to an embodiment of the present invention;
FIGS. 4 (a) -4 (b) are schematic diagrams of a two-stage bi-directional knowledge transfer strategy provided by embodiments of the present invention;
FIG. 5 is a schematic diagram of a complete three-dimensional point cloud registration method based on multi-form optimization provided by an embodiment of the present invention;
FIGS. 6 (a) -6 (c) are diagrams of registration results under different partial overlaps provided by embodiments of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but embodiments of the present invention are not limited thereto.
The inventor researches find that the existing registration method based on the evolution algorithm adopts different evolution algorithms to solve the point cloud registration problem, and the limitations of the method mainly comprise three aspects: firstly, the important role of robustness is ignored, and local optimum is easily trapped due to interference of shielding, abnormal values and noise; secondly, there is no reliable way to determine if the solution falls into a local minimum; thirdly, considering that no perfect loss function exists, the robustness and the precision of the algorithm cannot be realized at the same time. Therefore, the potential of evolutionary algorithms to solve the point cloud registration problem is still to be explored further to achieve satisfactory results in different registration scenarios.
Based on the above analysis, please refer to fig. 1, the embodiment of the invention provides a three-dimensional point cloud registration method based on multi-form optimization, which specifically includes the following steps:
s10, acquiring a point cloud pair to be registered of the scene to obtain a corresponding farthest point set.
Since the furthest point can sample, the structure of the point cloud can be reserved after downsampling, and the structure has better coverage rate on the whole point set, the embodiment of the invention provides an alternative scheme for acquiring the point cloud pairs to be registered of a scene to obtain the corresponding furthest point set, and the method comprises the following steps of, for each point cloud to be registered:
randomly sampling a first sampling point in the point cloud; the Euclidean distance between the first sampling point and other points in the point cloud is calculated, and the Euclidean distance is stored in a distance array; selecting the maximum point in all Euclidean distances as the next sampling point; and continuously calculating Euclidean distance between the sampling point and other points in the point cloud, judging whether the new Euclidean distance is smaller than the Euclidean distance in the corresponding distance array, if so, updating the distance array, and if so, maintaining the original distance array until the sampling point of the expected sampling number is traversed to obtain the farthest point set. As shown in fig. 2, the acquisition situation of the furthest point in the scene point cloud is given when the expected sampling numbers are 1, 2, 4, 10, 50 and 200 respectively. For example, there is a point cloud pair waiting for registration (X ', Y') in the scene, and the acquisition of the farthest point set is performed on the point cloud X 'and the point cloud Y' in the point cloud pair (X ', Y'), so as to form a point cloud pair (X, Y), and the point cloud pair (X, Y) is registered.
Therefore, the embodiment of the invention converts the registration problem of the point cloud pairs in the original scene into the registration problem of solving the point cloud pairs (X, Y) after acquisition. The subsequent registration is based on the furthest point set, so that for point clouds with larger volumes, the data volume of the registration point clouds is reduced due to acquisition, and the calculated amount of feature extraction is further reduced, so that the proposed registration method is more practical.
S20, dividing the three-dimensional point cloud registration problem into a robust registration task and an accurate registration task, solving the three-dimensional point cloud registration problem based on the robust registration task and the accurate registration task by utilizing a multi-task evolution algorithm, and taking the solution of the accurate registration task as an optimal scene point cloud change parameter; the information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy.
Aiming at the problem that the accuracy and the robustness in the point cloud registration process cannot be considered in the existing algorithm, the embodiment of the invention provides the method for dividing the three-dimensional point cloud registration problem into two highly similar tasks, namely a robust registration task and an accurate registration task according to the three-dimensional point cloud registration problem to be solved. The robust registration task has a wide search space, and a large number of point-to-point relations need to be considered; the accurate registration task focuses on the point-to-point relationship within the more promising overlap domain.
The embodiment of the invention provides an alternative scheme, which solves a three-dimensional point cloud registration problem based on a robust registration task and an accurate registration task by utilizing a multi-task evolution algorithm, and please refer to fig. 3, and comprises the following steps:
s201, setting parameters of a population in a multi-task evolution algorithm, and initializing the population.
Parameters of the population in the multi-task evolution algorithm are set, including the population size Pop, the maximum iteration number MaxIter, the random mating probability Rmp, the simulated binary crossover index Mu, the standard deviation Sigma of the Gaussian variation model, and the upper threshold KMax of the registration loss function and the lower threshold KMin of the loss function.
Initializing a population, specifically:
vector R for each individual in the initialized population being a randomly generated six-dimensional random variable 1 ,R 2 ,R 3 ,t 1 ,t 2 ,t 3 ]Wherein R is 1 、R 2 、R 3 The rotation parameters of the individual in x, y and z coordinate axes, t 1 、t 2 、t 3 The translation parameters of the individual in the x, y and z coordinate axes are respectively shown. Wherein the rotation parameter R 1 、R 2 、R 3 Is within the range ofTranslation parameter t 1 、t 2 、t 3 The range of (2) is obtained by a point cloud X and a point cloud Y, and the variation range of each coordinate axis is [ -T 1 ,T 1 ]、[-T 2 ,T 2 ]、[-T 3 ,T 3 ]The method can be calculated as:
wherein Max T1 、Min T1 、Wax T2 、Min T2 、Max T3 、Min T3 The maximum and minimum coordinate values corresponding to the merging point cloud obtained by merging the point cloud X and the point cloud Y.
S202, each individual in the initialized population is encoded into a unified rotation and translation space, and skill factors representing the good task attributes of the individuals are allocated.
Each individual in the initialized population is coded into a unified rotation and translation space using existing methods and is randomly assigned a skill factor for its good task attributes. For example, each individual in the initialized population is randomly assigned a skill factor of 0, 1, wherein a skill factor of 1 represents that the individual is good at a robust registration task and a skill factor of 0 represents that the individual is good at a precise registration task.
And S203, in the whole evolution process, information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy so as to obtain a new generation population.
The robust registration task and the accurate registration task defined by the embodiment of the invention can solve the registration problem between point clouds, but because the robust registration task and the accurate registration task have different preferences and difficulties, in order to realize more efficient genetic material exchange between different tasks, the embodiment of the invention constructs a two-stage bidirectional knowledge transfer strategy, based on complementary tasks, the assumption of better optimization performance is realized through information sharing of different stages, and the multi-form registration process is divided into two stages through changing the direction of knowledge transfer. Specifically, information sharing in the evolution process is realized by using a two-stage bidirectional knowledge transfer strategy, wherein the two stages specifically comprise:
the first stage, the information flow is from the robust registration task to the accurate registration task, and the robust registration task occupies a large proportion of evaluation processing at the first stage; the second stage, the information flow is from the accurate registration task to the robust registration task, and the accurate registration task occupies a large proportion of evaluation processing at the stage. Specifically, the implementation of each stage is as follows:
in the first stage of knowledge transfer, the information flow is mainly from a robust registration task to an accurate registration task. Since the coarse-to-fine optimization strategy of the robust registration task has a large and wide search space, the probability of convergence to a local minimum can be greatly reduced, providing a more useful positioning solution for the accurate registration task. The accurate registration task focuses on point pairs in an overlapping domain, and the convergence is accelerated by means of self searching capability, so that a high-quality solution is realized. In the second stage of knowledge transfer, the information flows in the opposite direction, mainly from the exact registration task to the robust registration task. Registration achieves satisfactory results in the first stage, but the likelihood of a locally optimal solution is still small. For example, in fig. 4 (a), the robust registration task provides a more useful localization solution for the accurate registration task; when the optimal solution portion is not found in fig. 4 (b). By transferring the useful knowledge obtained in the exact registration task back to the robust task, directing the search to areas of high quality solution, the robust registration task continues to find potentially optimal values in a wider search space. The arrows in fig. 4 (a) to 4 (b) indicate the direction of information flow.
Meanwhile, in the first stage of knowledge transfer, the robust registration task occupies a large proportion of evaluation processing, namely, in the knowledge transfer process of the stage, a threshold value, such as 0.7, is preset, numbers between 0 and 1 are randomly generated for each individual in the initialized population, the generated random numbers are compared with the threshold value, if the random numbers are smaller than or equal to the threshold value, the robust registration task evaluation process is executed, skill factors corresponding to task attributes are correspondingly distributed, if the random numbers are larger than the threshold value, the accurate registration task evaluation process is executed, skill factors corresponding to the task attributes are correspondingly distributed, and the specific adoption of the fitness evaluation function is detailed later.
Similarly, in the second stage of knowledge transfer, the accurate registration task occupies a large proportion of evaluation processing, that is, in the knowledge transfer process in this stage, a process similar to that in the first stage is adopted, and details are not repeated here.
For the first stage and the second stage execution times, the maximum iteration times MaxIter may be designed, for example, the first stage may execute MaxIter/2 times in the maximum iteration times MaxIter, and the second stage may execute the remaining MaxIter/2 times in the maximum iteration times MaxIter, but not limited to such design.
It can be seen that each task can produce unique search capabilities, with different tasks remaining robust and accurate during the search process.
S204, decoding each individual in the new generation population to obtain corresponding scene point cloud change parameters and corresponding skill factors thereof.
By adopting the existing method, each individual in the new generation population is decoded to obtain a corresponding scene point cloud change parameter, namely a rotation parameter R and a translation parameter t, and a corresponding skill factor, wherein the skill factor is used for judging the task attribute of each individual in the new generation population in the subsequent evaluation process.
S205, judging task attributes of each individual in the new generation of population according to the decoded skill factors, and selecting a corresponding fitness evaluation function according to the task attributes of each individual in the new generation of population for evaluation so as to solve the three-dimensional point cloud registration problem.
The task attribute of each individual in the new generation population can be obtained through decoding, and two different tasks are evaluated by using different fitness evaluation functions. The embodiment of the invention provides an alternative scheme, which selects a corresponding fitness evaluation function according to the task attribute of each individual in a new generation population to evaluate so as to solve a three-dimensional point cloud registration problem, and comprises the following steps:
if the task attribute is a robust registration task, constructing a scale self-adaptive cauchy weighted estimator, and reconstructing an adaptability evaluation function of the robust registration task according to the constructed scale self-adaptive cauchy weighted estimator; if the task attribute is an accurate registration task, constructing an adaptability evaluation function of the accurate registration task; and respectively evaluating according to the constructed fitness evaluation function to solve the three-dimensional point cloud registration problem. Specifically:
for the robust registration task, the embodiment of the invention provides a new fitness evaluation function to be robust to noise, outliers and partial overlap. First, the embodiment of the invention constructs a scale self-adaptive cauchy weight estimator, which can be expressed as:
wherein ρ (·) represents a scale-adaptive cauchy-weighted estimator, e i ＝||Rx i +t-y i || 2 Representation point (x) i ,y i ) K represents the scale parameter, r= [ R ] 1 ,R 2 ,R 3 ]，t＝[t 1 ,t 2 ,t 3 ]，||·|| 2 Representation calculation 2 Norms. Different scale parameters k allow different residuals to participate in the observation, thus having different effects on the total energy. The larger the scale parameter k value, the wider the cost curve and therefore the more residual values are involved in the observation. This largely avoids disabling interior points due to incorrect scene point cloud change parameters, i.e., rotation parameter R and translation parameter t. In contrast, smaller values of the scale parameter k result in smaller widths of the cost curve, which makes scene point cloud variation parameter estimation more accurate.
Based on formula (2), constructing an fitness evaluation function of the robust registration task, expressed as:
wherein f 1 (. Cndot.) represents the fitness evaluation function of the robust registration task, τ represents a fixed parameter, N represents the number of sampling points in the point cloud, and M represents the number of inliers in N.
In order to obtain more accurate results, it is desirable to maximize the cost contribution of only the inlier residuals. l (L) p Curve ratio of norms l 1 The curve of the norm is much flatter, which can greatly reduce the effects of outliers. But there is no distinction between the interior points and the outliers. The Huber function may divide the corresponding points into interior points and outliers by a threshold. The new Huber-p estimator emphasizes the inliers rather than outliers, almost completely ignoring the effects of outliers with higher accuracy. Thus, the invention is practicalThe fitness evaluation function of the precise registration task constructed by the embodiment can be expressed as:
wherein f 2 (-) represents the fitness evaluation function of the exact registration task, alpha represents an adjustment parameter and is used to adjust the adjustment parameter, I.I p Representation calculation p Norms.
According to the definition of the robust registration task and the accurate registration task, the fitness value of each individual in the population is calculated by using two residual error calculation methods with different properties.
In the population iteration process, judging whether the maximum iteration times MaxIter is met, and if the current iteration times are smaller than or equal to the maximum iteration times MaxIter, returning to the two-stage bidirectional knowledge migration strategy to continue the evolutionary algorithm iteration process; otherwise, the loop is terminated, and a solution of the accurate registration task is output as an optimal scene point cloud change parameter. In the population iteration process, a traditional evolutionary algorithm is adopted, and detailed description is not developed here.
And S30, registering each point in the farthest point set according to the optimal scene point cloud change parameters so as to realize three-dimensional point cloud registration.
By adopting the existing method, according to the optimal scene point cloud change parameters output by the step S20, the furthest point set acquired by the step S10 is registered to obtain configured point cloud data so as to realize three-dimensional point cloud registration.
Referring to fig. 5, a schematic diagram of a complete three-dimensional point cloud registration method frame based on multi-form optimization is provided, and the whole process of S10 to S30 is simply illustrated.
In order to verify the effectiveness of the three-dimensional point cloud registration method based on multi-form optimization provided by the embodiment of the invention, the following experiment is performed for verification.
Experiment 1: the method solves the point cloud registration problem under different outlier rates.
Three different levels of anomaly rate by adding random anomaly valuesRegistration experiments were 10%, 50% and 90%, respectively. The point cloud pair (X, Y) is generated as follows: first, the rabbit point cloud in the Stanford dataset is fixed as the source point cloud, rotated to another coordinate system, passed through y i ＝Rx i +t yields the reference point cloud. The rotation matrix R is at an angle [ -pi/2, pi/2]A 3 x 3 matrix randomly generated. The translation vector T is at [ -T, T]A 3 x 1 vector randomly generated within the range, where T is the length of the point cloud bounding box. Finally, to obtain the point cloud pairs (X, Y), random points can be generated within the largest bounding box.
In the experiment 1 simulation, the population size pop=100, the maximum iteration number maxiter=60, the random mating probability rmp=0.5, the simulated binary crossover index mu=10, the gaussian variation model standard deviation sigma=0.02, the upper threshold KMax of the registration loss function is set to the maximum bounding box length of the point cloud, and the lower threshold KMin of the loss function is set to twice the resolution of the point cloud. Finally, registration results of the method provided by the invention under different outlier ratios are shown in table 1.
Table 1 registration results at different outliers
| External point rate | Rotational error | Translational error |
| 10％ | 0.01 | 1.24E-05 |
| 50％ | 0.47 | 1.88E-04 |
| 90％ | 0.72 | 1.51E-04 |
As can be seen from table 1, the registration method provided by the embodiment of the invention can still obtain good registration results under different outliers.
Experiment 2: the method solves the point cloud registration problem under different noise.
To obtain data of different gaussian noise levels, the original point clouds are all normalized to [ -1,1], and then gaussian noise with zero mean and standard deviation σ is added to them. Finally they return to their original size. In experiment 2, sigma is 0.01, 0.02, 0.03, 0.04 and 0.05, and noise point clouds with different deviations are generated; in the experiment 2 simulation, the population size pop=100, the maximum iteration number maxiter=60, the random mating probability rmp=0.5, the simulated binary intersection index mu=10, the standard deviation sigma=0.02 of the gaussian variation model, the upper threshold KMax of the registration loss function is set to the maximum bounding box length of the point cloud, and the lower threshold KMin of the loss function is set to twice the resolution of the point cloud. Finally, the registration results of the method provided by the invention under different noise are shown in table 2.
Table 2 registration results at different noise
As can be seen from table 2, the registration method provided by the embodiment of the invention can still obtain good registration results under the influence of different noises.
Experiment 3, the method provided by the invention is used for solving the problem of partially overlapped point cloud registration.
Evaluated on three pairs of scan point clouds, their data sources were Armadillo, dragon and Buddha, respectively. In the experiment 3 simulation, the population size pop=100, the maximum iteration number maxiter=60, the random mating probability rmp=0.5, the simulated binary crossover index mu=10, the gaussian variation model standard deviation sigma=0.02, the upper threshold KMax of the registration loss function is set to the maximum bounding box length of the point cloud, and the lower threshold KMin of the loss function is set to twice the resolution of the point cloud. Finally, the registration results of the registration method provided by the invention under different partial overlapping are shown in fig. 6 (a) to 6 (c), wherein fig. 6 (a) is a schematic diagram of three partial overlapping images input by a scene, fig. 6 (b) is a schematic diagram of registration results of three overlapping images, and fig. 6 (c) is a schematic diagram of a cross section of configuration results of the three overlapping images. Therefore, the registration method provided by the embodiment of the invention can still obtain a good registration result under the condition of partial overlapping images.
In summary, the three-dimensional point cloud registration method based on multi-form optimization provided by the embodiment of the invention can effectively solve the problems that the traditional point cloud registration method based on evolution is poor in robustness and easy to fall into local optimization, and is a novel multi-form optimization registration strategy, and the configuration process comprises the following steps: acquiring a point cloud pair to be registered of a scene to obtain a corresponding farthest point set; dividing the three-dimensional point cloud registration problem into a robust registration task and an accurate registration task, solving the three-dimensional point cloud registration problem based on the robust registration task and the accurate registration task by utilizing a multi-task evolution algorithm, and taking the solution of the accurate registration task as an optimal scene point cloud change parameter; the information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy; and registering each point in the furthest point set according to the optimal scene point cloud change parameters so as to realize three-dimensional point cloud registration. Therefore, the point cloud registration problem is solved through the multi-task evolution algorithm, a two-stage bidirectional knowledge transfer strategy is constructed in the evolution process, robustness and accuracy are considered in two independent stages, so that the registration result can ensure both robustness and accuracy, and an adaptability evaluation function is newly designed in the realization of a robust registration task, so that the registration robustness is greatly improved; meanwhile, the situation of local optimization of the algorithm is reduced through the cooperative action between a robust registration task and an accurate registration task in the multi-task evolution algorithm in the whole process, so that the accuracy of registration is greatly improved. The registration method provided by the embodiment of the invention has robustness to noise, abnormal values and partial overlapping conditions, namely the sensitivity to the characteristic quality is reduced in the registration process, and the method can be applied to a plurality of real registration scenes, such as object registration, scene reconstruction, simultaneous positioning and mapping and the like.
Referring to fig. 7, an embodiment of the present invention provides an electronic device, which includes a processor 701, a communication interface 702, a memory 703 and a communication bus 704, wherein the processor 701, the communication interface 702 and the memory 703 complete communication with each other through the communication bus 704;
a memory 703 for storing a computer program;
the processor 701 is configured to implement the steps of the three-dimensional point cloud registration method based on multi-form optimization when executing the program stored in the memory 703.
The embodiment of the invention provides a computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and the computer program realizes the steps of the three-dimensional point cloud registration method based on multi-form optimization when being executed by a processor.
For the electronic device/storage medium embodiments, the description is relatively simple as it is substantially similar to the method embodiments, as relevant points are found in the partial description of the method embodiments.
In the description of the present invention, it should be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
Although the invention is described herein in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the specification and the drawings. In the description, the word "comprising" does not exclude other elements or steps, and the "a" or "an" does not exclude a plurality. Some measures are described in mutually different embodiments, but this does not mean that these measures cannot be combined to produce a good effect.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.
Claims (4)
1. The three-dimensional point cloud registration method based on multi-form optimization is characterized by comprising the following steps of:
acquiring a point cloud pair to be registered of a scene to obtain a corresponding farthest point set;
dividing the three-dimensional point cloud registration problem into a robust registration task and an accurate registration task, solving the three-dimensional point cloud registration problem based on the robust registration task and the accurate registration task by utilizing a multi-task evolution algorithm, and taking the solution of the accurate registration task as an optimal scene point cloud change parameter;
registering each point in the furthest point set according to the optimal scene point cloud change parameters so as to realize three-dimensional point cloud registration; wherein,
acquiring a point cloud pair to be registered of a scene to obtain a corresponding farthest point set, and aiming at each point cloud to be registered, the method comprises the following steps:
randomly sampling a first sampling point in the point cloud; the Euclidean distance between the first sampling point and other points in the point cloud is calculated, and the Euclidean distance is stored in a distance array; selecting the maximum point in all Euclidean distances as the next sampling point; continuously calculating Euclidean distance between the sampling point and other points in the point cloud, judging whether the new Euclidean distance is smaller than the Euclidean distance in the corresponding distance array, if so, updating the distance array until the sampling point of the expected sampling number is traversed to obtain the farthest point set;
solving a three-dimensional point cloud registration problem based on a robust registration task and an accurate registration task by using a multi-task evolutionary algorithm, comprising:
setting parameters of a population in a multitask evolution algorithm, and initializing the population; encoding each individual in the initialized population into a unified rotation and translation space and assigning a skill factor representative of its task attributes of skill; in the whole evolution process, information sharing in the evolution process is realized by utilizing a two-stage bidirectional knowledge transfer strategy so as to obtain a new generation population; decoding each individual in the new generation population to obtain a corresponding scene point cloud change parameter and a corresponding skill factor thereof; judging task attributes of each individual in the new generation of population according to the decoded skill factors, and selecting a corresponding fitness evaluation function according to the task attributes of each individual in the new generation of population for evaluation so as to solve the three-dimensional point cloud registration problem; wherein,
selecting a corresponding fitness evaluation function according to the task attribute of each individual in the new generation population to evaluate so as to solve the three-dimensional point cloud registration problem, wherein the method comprises the following steps:
if the task attribute is a robust registration task, constructing a scale self-adaptive cauchy weighted estimator, and reconstructing an adaptability evaluation function of the robust registration task according to the constructed scale self-adaptive cauchy weighted estimator; if the task attribute is an accurate registration task, constructing an adaptability evaluation function of the accurate registration task; evaluating according to the constructed fitness evaluation function to solve the three-dimensional point cloud registration problem; wherein, the constructed scale self-adaptive cauchy weighted estimator is expressed as:
；
wherein,representing a scale-adaptive cauchy-weighted estimator, < >>Representation dot->Is used for the residual error of (c),krepresenting scale parameters->，/>Wherein->、/>、/>Rotation parameters of the individual in x, y, z coordinate axes, respectively->、/>、/>Translation parameters of the individual in x, y, z coordinate axes, respectively->Representation calculationl 2 A norm; correspondingly, the fitness evaluation function of the robust registration task is expressed as:
；
wherein,fitness evaluation function representing a robust registration task, < ->The fixed parameter is indicated as such,Nrepresenting the number of sampling points in the point cloud,Mrepresenting the number of inliers in N.
2. The three-dimensional point cloud registration method based on multi-form optimization according to claim 1, wherein the information sharing in the evolution process is realized by using a two-stage bidirectional knowledge transfer strategy, comprising:
the first stage, the information flow is from the robust registration task to the accurate registration task, and the robust registration task occupies a large proportion of evaluation processing at the first stage;
the second stage, the information flow is from the accurate registration task to the robust registration task, and the accurate registration task occupies a large proportion of evaluation processing at the stage.
3. The three-dimensional point cloud registration method based on multi-form optimization according to claim 1, wherein the fitness evaluation function of the accurate registration task is expressed as:
；
wherein,fitness evaluation function representing an accurate registration task, +.>Representing adjustment parameters->Representation calculationl p Norms.
4. An electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus;
the memory is used for storing a computer program;
the processor is configured to implement the three-dimensional point cloud registration method based on multi-form optimization according to any one of claims 1 to 3 when executing the program stored in the memory.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310101088.5A CN116342666B (en) | 2023-02-10 | 2023-02-10 | Three-dimensional point cloud registration method based on multi-form optimization and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310101088.5A CN116342666B (en) | 2023-02-10 | 2023-02-10 | Three-dimensional point cloud registration method based on multi-form optimization and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116342666A CN116342666A (en) | 2023-06-27 |
| CN116342666B true CN116342666B (en) | 2024-03-19 |
Family
ID=86881273
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310101088.5A Active CN116342666B (en) | 2023-02-10 | 2023-02-10 | Three-dimensional point cloud registration method based on multi-form optimization and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116342666B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108470358A (en) * | 2018-02-13 | 2018-08-31 | 南京旅游职业学院 | A kind of point cloud registration method for shaking artificial bee colony algorithm based on second order |
| CN108564605A (en) * | 2018-04-09 | 2018-09-21 | 大连理工大学 | A kind of three-dimensional measurement spots cloud optimization method for registering |
| CN109949350A (en) * | 2019-03-11 | 2019-06-28 | 中国矿业大学(北京) | A kind of multidate point cloud autoegistration method based on form invariant features |
| CN111415379A (en) * | 2020-03-23 | 2020-07-14 | 南京大学 | Three-dimensional point cloud data registration method based on cuckoo optimization |
| CN112669356A (en) * | 2020-12-14 | 2021-04-16 | 哈尔滨理工大学 | Three-dimensional point cloud registration method based on self-adaptive firework and KD tree algorithm |
| CN112991402A (en) * | 2021-03-19 | 2021-06-18 | 西北大学 | Cultural relic point cloud registration method and system based on improved differential evolution algorithm |
| CN115147833A (en) * | 2022-07-25 | 2022-10-04 | 温州大学乐清工业研究院 | Part pose identification method and system |
| CN115601408A (en) * | 2022-10-11 | 2023-01-13 | 西安电子科技大学（Cn） | Point cloud registration method based on particle swarm optimization and topological graph |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013173465A1 (en) * | 2012-05-15 | 2013-11-21 | Board Of Regents, The University Of Texas System | Imaging device capable of producing three dimensional representations and methods of use |
| GB2526342A (en) * | 2014-05-22 | 2015-11-25 | Nokia Technologies Oy | Point cloud matching method |
| US10229533B2 (en) * | 2016-11-03 | 2019-03-12 | Mitsubishi Electric Research Laboratories, Inc. | Methods and systems for fast resampling method and apparatus for point cloud data |
| US11506502B2 (en) * | 2019-07-12 | 2022-11-22 | Honda Motor Co., Ltd. | Robust localization |
-
2023
- 2023-02-10 CN CN202310101088.5A patent/CN116342666B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108470358A (en) * | 2018-02-13 | 2018-08-31 | 南京旅游职业学院 | A kind of point cloud registration method for shaking artificial bee colony algorithm based on second order |
| CN108564605A (en) * | 2018-04-09 | 2018-09-21 | 大连理工大学 | A kind of three-dimensional measurement spots cloud optimization method for registering |
| CN109949350A (en) * | 2019-03-11 | 2019-06-28 | 中国矿业大学(北京) | A kind of multidate point cloud autoegistration method based on form invariant features |
| CN111415379A (en) * | 2020-03-23 | 2020-07-14 | 南京大学 | Three-dimensional point cloud data registration method based on cuckoo optimization |
| CN112669356A (en) * | 2020-12-14 | 2021-04-16 | 哈尔滨理工大学 | Three-dimensional point cloud registration method based on self-adaptive firework and KD tree algorithm |
| CN112991402A (en) * | 2021-03-19 | 2021-06-18 | 西北大学 | Cultural relic point cloud registration method and system based on improved differential evolution algorithm |
| CN115147833A (en) * | 2022-07-25 | 2022-10-04 | 温州大学乐清工业研究院 | Part pose identification method and system |
| CN115601408A (en) * | 2022-10-11 | 2023-01-13 | 西安电子科技大学（Cn） | Point cloud registration method based on particle swarm optimization and topological graph |
Non-Patent Citations (5)
| Title |
|---|
| Evolutionary Multitasking for Coarse-to-Fine Point Cloud Registration with Chaotic Opposition Search Strategy;Yue Wu et al;《Second CAAI International Conference》;第293-304页 * |
| Evolutionary Multitasking with Solution Space Cutting for Point Cloud Registration;Yue Wu et al;《arXiv》;第1-16页 * |
| Multi-view Point Cloud Registration based on Evolutionary Multitasking with Bi-Channel Knowledge Sharing Mechanism;Yue Wu et al;《arXiv》;第1-19页 * |
| Point Cloud Registration Based on Global and Local Feature Fusion;Wenping Ma et al;《5th IFIP TC 12 International Conference》;第310-317页 * |
| 基于ＲＯＩ－ＲＳＩＣＰ算法的车轮廓形动态检测;易倩 等;《中国激光》;第47卷(第11期);第1-12页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116342666A (en) | 2023-06-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11880927B2 (en) | Three-dimensional object reconstruction from a video | |
| CN110287873B (en) | Non-cooperative target pose measurement method and system based on deep neural network and terminal equipment | |
| CN111627065A (en) | Visual positioning method and device and storage medium | |
| CN111524168B (en) | Point cloud data registration method, system and device and computer storage medium | |
| Yu et al. | Robust robot pose estimation for challenging scenes with an RGB-D camera | |
| CN113160287B (en) | Complex component point cloud splicing method and system based on feature fusion | |
| CN112084849A (en) | Image recognition method and device | |
| CN111768447A (en) | Monocular camera object pose estimation method and system based on template matching | |
| CN116662600B (en) | Visual positioning method based on lightweight structured line map | |
| CN114332125A (en) | Point cloud reconstruction method and device, electronic equipment and storage medium | |
| Xu et al. | GLORN: Strong generalization fully convolutional network for low-overlap point cloud registration | |
| CN116402976A (en) | Training method and device for three-dimensional target detection model | |
| Wu et al. | Sc-wls: Towards interpretable feed-forward camera re-localization | |
| CN116310098A (en) | Multi-view three-dimensional reconstruction method based on attention mechanism and variable convolution depth network | |
| Lichy et al. | Fast light-weight near-field photometric stereo | |
| CN112364881A (en) | Advanced sampling consistency image matching algorithm | |
| CN114663880A (en) | Three-dimensional target detection method based on multi-level cross-modal self-attention mechanism | |
| Potje et al. | Learning geodesic-aware local features from RGB-D images | |
| Zhao et al. | Learning probabilistic coordinate fields for robust correspondences | |
| CN116342666B (en) | Three-dimensional point cloud registration method based on multi-form optimization and electronic equipment | |
| CN115661218B (en) | Virtual super-point-based laser point cloud registration method and system | |
| Slimani et al. | RoCNet++: Triangle-based descriptor for accurate and robust point cloud registration | |
| Cheng et al. | An augmented reality image registration method based on improved ORB | |
| Tan et al. | Using multi-level consistency learning for partial-to-partial point cloud registration | |
| Lyra et al. | Development of an efficient 3D reconstruction solution from permissive open-source code |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |