Plant three-dimensional point cloud registration method and device, medium and product
By combining hierarchical non-convex optimization and L-BFGS optimization, and using color and geometric features for two-level optimization, the accuracy and efficiency problems of 3D point cloud registration for plants were solved, achieving efficient and stable registration results.
Patent Information
- Application Number
- CN202511075840.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-01
- Publication Date
- 2025-11-18
AI Technical Summary
Existing 3D point cloud registration techniques are highly dependent on initial transformations, easily get trapped in local optima, have high misregistration rates, low computational efficiency, do not fully utilize color information, and lack density adaptive mechanisms, resulting in insufficient accuracy and efficiency in plant point cloud registration.
By combining hierarchical non-convex optimization and L-BFGS optimization, the system preprocesses, extracts, and calculates features from the 3D point cloud of the plant, and performs two-level optimization using color and geometric features to obtain the full-resolution transformation value of the transformation matrix, thereby improving registration accuracy and efficiency.
It improves the accuracy and efficiency of 3D point cloud registration for plants, achieves a balance between detail preservation and global consistency, and enhances the stability and computational efficiency of registration.
Smart Images
Figure CN120976272A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of point cloud registration, and in particular to a plant three-dimensional point cloud registration method and device, medium and product. BACKGROUND
[0002] Three-dimensional point cloud registration technology plays a crucial role. Its main goal is to accurately align point cloud data from different perspectives, times or devices into the same coordinate system, thereby achieving high-precision three-dimensional reconstruction and analysis. The mainstream technical path of three-dimensional point cloud registration mainly includes: Iterative Closest Point (ICP) algorithm, Feature-based Registration method based on Fast Point Feature Histograms (FPFH) and Fast Global Registration (FGR) method.
[0003] The above methods mainly have the following technical problems:
[0004] 1. Dependence on initial transformation: ICP and its variants require good initial alignment, otherwise it is easy to fall into local optimum, affecting the registration accuracy.
[0005] 2. Misregistration problem: plant point cloud is sparse and local feature similarity is high, feature registration algorithm based on FPFH is prone to misregistration, reducing registration accuracy and stability.
[0006] 3. Low computational efficiency: traditional ICP and its improved methods have high computational complexity, which is difficult to meet the efficient processing needs of large-scale point cloud data.
[0007] 4. Ignoring color information: mainstream methods such as FGR only rely on geometric information and do not fully utilize color features, resulting in decreased registration accuracy in color complex plant point cloud scenes.
[0008] 5. Lack of density adaptive mechanism: unable to dynamically adjust the weight of registration point pairs according to the local density change of point cloud, resulting in uneven registration quality in high-density and low-density areas. SUMMARY
[0009] The purpose of the present application is to provide a plant three-dimensional point cloud registration method, device, medium and product to solve the problems of low efficiency and accuracy of plant three-dimensional point cloud registration.
[0010] To achieve the above purpose, the present application provides the following solutions:
[0011] In a first aspect, the present application provides a plant three-dimensional point cloud registration method, comprising:
[0012] obtaining a source point cloud and a target point cloud to be registered; the source point cloud and the target point cloud are both plant three-dimensional point clouds;
[0013] determining the source point cloud and the target point cloud as current point clouds respectively, pre-processing the current point clouds to obtain a full-resolution current point cloud and a down-sampled current point cloud;
[0014] extracting features from the down-sampled current point cloud to obtain geometric features, color features and fusion features of each point in the down-sampled current point cloud;
[0015] pairing each point in the down-sampled source point cloud with each point in the down-sampled target point cloud based on the fusion features of each point in the down-sampled source point cloud and the fusion features of each point in the down-sampled target point cloud, to obtain a first point pair set, and determining color similarity weights of each point pair in the first point pair set based on the color features of the two points of each point pair in the first point pair set respectively;
[0016] using hierarchical non-convex optimization and L-BFGS optimization, and based on the color similarity weights of all point pairs in the first point pair set, iteratively optimizing the transformation matrix to obtain a coarse transformation value of the transformation matrix, and using the coarse transformation value of the transformation matrix to transform the full-resolution source point cloud to obtain a coarsely transformed source point cloud; the first energy function is a function of the color similarity weights between the two points of all point pairs in the first point pair set, the geometric features of the two points of all point pairs in the first point pair set and the transformation matrix; the transformation matrix includes a rotation matrix and a translation vector;
[0017] extracting features from the coarsely transformed source point cloud to obtain geometric features, color features and fusion features of each point in the coarsely transformed source point cloud;
[0018] pairing each point in the coarsely transformed source point cloud with each point in the down-sampled target point cloud based on the fusion features of each point in the coarsely transformed source point cloud and the fusion features of each point in the down-sampled target point cloud, to obtain a second point pair set, and determining color similarity weights of each point pair in the second point pair set based on the color features of the two points of each point pair in the second point pair set respectively;
[0019] using hierarchical non-convex optimization and L-BFGS optimization, and based on the color similarity weights of all point pairs in the second point pair set, iteratively optimizing the transformation matrix to obtain a full-resolution transformation value of the transformation matrix; the second energy function is a function of the color similarity weights between the two points of all point pairs in the second point pair set, the geometric features of the two points of all point pairs in the second point pair set and the transformation matrix;
[0020] determine the registered point cloud based on the transform matrix-based full-resolution transform value, the full-resolution source point cloud and the full-resolution target point cloud.
[0021] In an embodiment, the preprocessing comprises: noise reduction processing and down-sampling; the preprocessing of the current point cloud comprises: obtaining the full-resolution current point cloud and the down-sampled current point cloud, comprising:
[0022] performing noise reduction processing on the current point cloud to obtain the full-resolution current point cloud;
[0023] performing down-sampling on the full-resolution current point cloud to obtain the down-sampled current point cloud.
[0024] In an embodiment, the feature extraction is performed on the down-sampled current point cloud to obtain the geometric feature, the color feature and the fusion feature of each point in the down-sampled current point cloud, comprising:
[0025] calculating the FPFH of each point in the down-sampled current point cloud respectively to obtain the geometric feature of each point in the down-sampled current point cloud;
[0026] respectively converting the RGB value of each point in the down-sampled current point cloud into the Lab value to obtain the color feature of each point in the down-sampled current point cloud;
[0027] determining the fusion feature of each point in the down-sampled current point cloud based on the first preset color weight, the geometric feature of each point in the down-sampled current point cloud and the color feature of each point in the down-sampled current point cloud.
[0028] In an embodiment, based on the fusion feature of each point in the down-sampled source point cloud and the fusion feature of each point in the down-sampled target point cloud, each point in the down-sampled source point cloud and each point in the down-sampled target point cloud are paired two by two to obtain a first point pair set, comprising:
[0029] pairing any point in the down-sampled source point cloud with any point in the down-sampled target point cloud to obtain a plurality of first initial point pairs;
[0030] respectively determining the norm of the difference of the fusion features of the two points in each first initial point pair, and determining all the first initial point pairs whose norm is less than a first similarity threshold as the point pairs in the first point pair set.
[0031] In an embodiment, the first energy function comprises:
[0032]
[0033] wherein E1(R,t) is the function value of the first energy function; C law is the first point pair set; w ij is pi a color similarity weight between p j and p i is a point in the first point pair set, p j is a point in the first point pair set, p μ is a robust kernel function with a control non-convexity of μ; R is a rotation matrix; FPFH(p i ) is a geometric feature of p i ; t is a translation vector; FPFH(p j ) is a geometric feature of p j ; ||·|| is a norm.
[0034] In an embodiment, feature extraction is performed on the coarsely transformed source point cloud to obtain geometric features, color features and fusion features of each point in the coarsely transformed source point cloud, including:
[0035] FPFH is calculated for each point in the coarsely transformed source point cloud to obtain geometric features of each point in the coarsely transformed source point cloud;
[0036] RGB values of each point in the coarsely transformed source point cloud are converted into Lab values to obtain color features of each point in the coarsely transformed source point cloud;
[0037] Based on the second preset color weight, the geometric features of each point in the coarsely transformed source point cloud and the color features of each point in the coarsely transformed source point cloud, fusion features of each point in the coarsely transformed source point cloud are determined.
[0038] In an embodiment, based on the fusion features of each point in the coarsely transformed source point cloud and the fusion features of each point in the down-sampled target point cloud, each point in the coarsely transformed source point cloud and each point in the down-sampled target point cloud are paired two by two to obtain a second point pair set, including:
[0039] Any point in the coarsely transformed source point cloud is paired with any point in the down-sampled target point cloud to obtain a plurality of second initial point pairs;
[0040] The norm of the difference between the fusion features of the two points in each second initial point pair is determined, and all second initial point pairs with a norm less than a second similarity threshold are determined as point pairs in the second point pair set.
[0041] In a second aspect, the present application provides a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the plant three-dimensional point cloud registration method described above.
[0042] In a third aspect, the present application provides a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the plant three-dimensional point cloud registration method described above.
[0043] In a fourth aspect, the present application provides a computer program product comprising a computer program, which, when executed by a processor, implements the plant three-dimensional point cloud registration method described above.
[0044] According to the specific embodiments provided by the present application, the following technical effects are disclosed:
[0045] The present application discloses a plant three-dimensional point cloud registration method, device, medium and product, which not only uses geometric features for point cloud alignment, but also introduces color features, and comprehensively obtains fusion features, so that the registration has a better balance in detail preservation and global consistency, thereby improving the registration accuracy; two-stage optimization is performed by using hierarchical non-convex optimization and L-BFGS optimization to obtain the full-resolution transformation value of the transformation matrix, effectively improving the registration efficiency and improving the efficiency and accuracy of plant three-dimensional point cloud registration. BRIEF DESCRIPTION OF DRAWINGS
[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0047] Figure 1 The plant three-dimensional point cloud registration method flowchart provided by an embodiment of the present application is shown in the figure.
[0048] Figure 2 The structural diagram of a computer device provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION
[0049] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0050] The purpose of the present application is to provide a plant three-dimensional point cloud registration method, device, medium and product, which aims to improve the efficiency and accuracy of plant three-dimensional point cloud registration.
[0051] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.
[0052] In an exemplary embodiment, as shown in Figure 1 a plant three-dimensional point cloud registration method is provided, comprising:
[0053] Step 1: obtaining a source point cloud and a target point cloud to be registered.
[0054] Among them, the source point cloud and the target point cloud are both plant three-dimensional point clouds.
[0055] Step 2: determining the source point cloud and the target point cloud as current point clouds respectively, pre-processing the current point clouds to obtain full-resolution current point clouds and down-sampled current point clouds.
[0056] As an optional implementation, the pre-processing includes noise reduction processing and down-sampling; and step 2 includes:
[0057] Step 21: performing noise reduction processing on the current point clouds to obtain full-resolution current point clouds.
[0058] Specifically, during noise reduction processing, statistical filtering and radius filtering are applied to remove outliers; and a pass-through filter is used to crop background and invalid areas.
[0059] Step 22: down-sampling the full-resolution current point clouds to obtain down-sampled current point clouds.
[0060] Specifically, during down-sampling, an appropriate voxel edge length is set so that the number of sampling points is 25% of the full-resolution current point clouds, thereby obtaining the down-sampled current point clouds.
[0061] Step 3: performing feature extraction on the down-sampled current point clouds to obtain geometric features, color features and fusion features of each point in the down-sampled current point clouds.
[0062] As an optional implementation, step 3 includes:
[0063] Step 31: calculating FPFH (Fast Point Feature Histograms) for each point in the down-sampled current point clouds respectively to obtain geometric features of each point in the down-sampled current point clouds.
[0064] Step 32: converting the RGB values of each point in the down-sampled current point clouds into Lab values respectively to obtain color features of each point in the down-sampled current point clouds.
[0065] Step 33: determining the fusion feature of each point in the down-sampled current point cloud based on the first preset color weight, the geometric feature of each point in the down-sampled current point cloud, and the color feature of each point in the down-sampled current point cloud.
[0066] Specifically, the calculation formula of the fusion feature of any point in the down-sampled current point cloud is:
[0067] fusion(p a )=[FPFH(p a ),λ1·Lab(p a )].
[0068] Wherein, fusion(p a ) is the fusion feature of p a , p a is a point a in the down-sampled current point cloud; FPFH(p a ) is the geometric feature of p a ; λ1 is the first preset color weight, λ1∈[0.5,2.0]; Lab(p a ) is the color feature of p a .
[0069] Step 4: pairing each point in the down-sampled source point cloud with each point in the down-sampled target point cloud based on the fusion feature of each point in the down-sampled source point cloud and the fusion feature of each point in the down-sampled target point cloud, to obtain a first point pair set, and determining the color similarity weight of each point pair in the first point pair set based on the color feature of the two points of each point pair in the first point pair set, respectively.
[0070] As an optional implementation, in step 4, pairing each point in the down-sampled source point cloud with each point in the down-sampled target point cloud based on the fusion feature of each point in the down-sampled source point cloud and the fusion feature of each point in the down-sampled target point cloud, to obtain a first point pair set, includes:
[0071] Step 41: pairing any point in the down-sampled source point cloud with any point in the down-sampled target point cloud to obtain a plurality of first initial point pairs.
[0072] Step 42: respectively determining the norm of the difference of the fusion features of the two points in each first initial point pair, and determining all first initial point pairs with a norm less than a first similarity threshold as the point pairs in the first point pair set.
[0073] Specifically, the first initial point pair is represented as:
[0074] C law ={(p i ,p j )|||fusion(pi ) fusion(p j )|| < τ1}.
[0075] where C law is the first point pair set; p i is a point in any point pair in the first point pair set belonging to the down-sampled source point cloud; p j is a point in any point pair in the first point pair set belonging to the down-sampled target point cloud; fusion(p i ) is the fusion feature of p i ; fusion(p j ) is the fusion feature of p j ; ||·|| is a norm; τ1 is the first similarity threshold.
[0076] The color similarity weight of each point pair in the first point pair set is calculated according to the following formula:
[0077] w ij = exp(- β · ||Lab(p i )- Lab(p j )||), (p 2 , p i ) ∈ C j . law .
[0078] where w ij is the color similarity weight between p i and p j ; exp(·) is an exponential function; β is a color similarity sensitivity adjustment parameter, usually in the range of [0.5, 2.0].
[0079] Step 5: Based on the color similarity weights of all point pairs in the first point pair set, the transformation matrix is iteratively optimized to obtain a coarse transformation value of the transformation matrix, and the full-resolution source point cloud is transformed using the coarse transformation value of the transformation matrix to obtain a coarse-transformed source point cloud, by using hierarchical non-convex optimization and L-BFGS optimization, and taking the minimum value of the first energy function as the target.
[0080] where the first energy function is a function of the color similarity weight between the two points of all point pairs in the first point pair set, the geometric features of the two points of all point pairs in the first point pair set, and the transformation matrix; the transformation matrix includes a rotation matrix and a translation vector.
[0081] As an optional implementation, in step 5, the first energy function includes:
[0082]
[0083] wherein E1(R, t) is a function value of the first energy function; C law is a first point pair set; w ij is a color similarity weight between p i and p j ; p i is a point in the first point pair set; p j is a point in the first point pair set; p μ is a robust kernel of a robust kernel function with a non-convex degree of μ; R is a rotation matrix; FPFH(p i ) is a geometric feature of p i ; t is a translation vector; FPFH(p j ) is a geometric feature of p j ; ||·|| is a norm.
[0084] In step 5, using Graduated Non-Convexity (GNC) and L-BFGS optimization, based on the color similarity weight between the two points of all point pairs in the first point pair set, the function value of the first energy function is minimized to obtain the coarse transformation value of the transformation matrix, including the following contents:
[0085] (1) Initialize the transformation matrix, initialize the robust kernel of the first energy function to make the robust kernel approximate linear, and the optimization problem is approximately convex.
[0086] (2) Input the color similarity weight between the two points of all point pairs in the first point pair set and the geometric feature of the two points of all point pairs in the first point pair set into the first energy function.
[0087] (3) Take GNC as the outer loop, and when iterating, the update formula of the robust kernel of the robust kernel function is: μ k+1 = γ·μ k , μ k+1 is the robust kernel of the k+1th iteration; γ is the update parameter of the robust kernel; μ k is the robust kernel of the kth iteration. Usually, 3 rounds-5 rounds of GNC can be set to converge.
[0088] (4) When the robust kernel of the first energy function is fixed in each round of GNC optimization, take L-BFGS as the inner loop, and use the L-BFGS optimizer to minimize the function value of the first energy function to obtain the coarse transformation value of the transformation matrix. Wherein, the transformation matrix is converted into Lie algebra form ξ = [ω T , t T ] T , ω is the rotation vector in Lie algebra form; T is the transpose; R = exp([ω] ×Orthogonality is maintained through exponential mapping; the gradient of the first energy function and the sparse Hessian matrix are calculated using an automatic differentiation tool; the inner iterations typically converge in 20–40 steps.
[0089] In step 5, the source point cloud at full resolution is transformed using the coarse transformation value of the transformation matrix to obtain the coarsely transformed source point cloud. The calculation formula is as follows:
[0090]
[0091] in, The source point cloud after coarse transformation; R low This represents the coarse transformation value of the rotation matrix; The source point cloud is at full resolution; t low This is the coarse transformation value of the translation vector.
[0092] Step 6: Extract features from the source point cloud after coarse transformation to obtain the geometric features, color features, and fusion features of each point in the source point cloud after coarse transformation.
[0093] As an optional implementation, step 6 includes:
[0094] Step 61: Calculate the FPFH for each point in the source point cloud after coarse transformation to obtain the geometric features of each point in the source point cloud after coarse transformation.
[0095] Step 62: Convert the RGB values of each point in the source point cloud after coarse transformation to Lab values to obtain the color features of each point in the source point cloud after coarse transformation.
[0096] Step 63: Based on the second preset color weight, the geometric features of each point in the coarsely transformed source point cloud, and the color features of each point in the coarsely transformed source point cloud, determine the fusion features of each point in the coarsely transformed source point cloud.
[0097] Specifically, the formula for calculating the fusion feature of any point in the source point cloud after coarse transformation is as follows:
[0098] fusion'(p b )=[FPFH(p b ), λ2·Lab(p b )).
[0099] Among them, fusion'(p b ) is p b fusion features, p b Point b is the source point cloud after coarse transformation; FPFH(p a ) is p b The geometric features; λ2 is the second preset color weight, λ2∈[0.5,2.0]; Lab(p a ) is pa Color characteristics.
[0100] Step 7: Based on the fusion features of each point in the source point cloud after coarse transformation and the fusion features of each point in the target point cloud after downsampling, pair each point in the source point cloud after coarse transformation and each point in the target point cloud after downsampling to obtain a second set of point pairs. Then, based on the color features of the two points in each pair in the second set of point pairs, determine the color similarity weight of each pair of point pairs in the second set of point pairs.
[0101] As an optional implementation, in step 7, based on the fusion features of each point in the coarsely transformed source point cloud and the fusion features of each point in the downsampled target point cloud, each point in the coarsely transformed source point cloud and each point in the downsampled target point cloud are paired up to obtain a second set of point pairs, including:
[0102] Step 71: Pair any point in the source point cloud after coarse transformation with any point in the target point cloud after downsampling to obtain multiple second initial point pairs.
[0103] Step 72: Determine the norm of the difference between the fusion features of the two points in each second initial point pair, and determine all second initial point pairs whose norm is less than the second similarity threshold as point pairs in the second point pair set.
[0104] Specifically, the second initial point pair is represented as:
[0105] C full ={(p m ,p n )|||fusion'(p m )-fusion(p n )||<τ2}.
[0106] Among them, C full For the second point pair set; p m p represents any point in the second set of point pairs that belongs to the source point cloud after the coarse transformation; n For any point pair in the second point pair set, the point belongs to the downsampled target point cloud; fusion'(p m ) is p m Fusion characteristics; fusion(p n ) is p n The fusion features; τ2 is the second similarity threshold.
[0107] The formula for calculating the color similarity weight of each pair of points in the second point pair set is as follows:
[0108] w mn =exp(-β·||Lab(p) m )-Lab(p n )||2 ),(p m ,p n )∈C full 。
[0109] where w mn is the color similarity weight between p m and p n .
[0110] Step 8: Based on the color similarity weights between all the point pairs in the second set of point pairs, the transformation matrix is iteratively optimized by using hierarchical non-convex optimization and L-BFGS optimization with the objective of minimizing the function value of the second energy function, to obtain the full-resolution transformation value of the transformation matrix.
[0111] where the second energy function is a function of the color similarity weights between the two points of all the point pairs in the second set of point pairs, the geometric features of the two points of all the point pairs in the second set of point pairs, and the transformation matrix.
[0112] In step 8, when the transformation matrix is iteratively optimized by using hierarchical non-convex optimization and L-BFGS optimization with the objective of minimizing the function value of the second energy function, based on the color similarity weights between all the point pairs in the second set of point pairs, to obtain the full-resolution transformation value of the transformation matrix, the following contents are included:
[0113] (1) The transformation matrix is initialized as the coarse transformation value of the transformation matrix, and the robust kernel of the second energy function is initialized to make the robust kernel approximately linear, and the optimization problem is approximately convex.
[0114] (2) The color similarity weights between the two points of all the point pairs in the second set of point pairs and the geometric features of the two points of all the point pairs in the second set of point pairs are input into the second energy function.
[0115] (3) GNC is used as the outer loop, and when iterating, the update formula of the robust kernel of the robust kernel function is: μ k+1 = γ·μ k , μ k+1 is the robust kernel of the k+1th iteration; γ is the update parameter of the robust kernel; μ k is the robust kernel of the kth iteration. Usually, the number of GNC rounds is set to 3-5 rounds, which can converge.
[0116] (4) When the robust kernel of the second energy function is fixed in each round of GNC optimization, L-BFGS is used as the inner loop, and the L-BFGS optimizer is used to minimize the function value of the second energy function to obtain the coarse transformation value of the transformation matrix. Where the transformation matrix is converted into Lie algebra form ξ = [ω T , t T ] T ; R = exp([ω] ×) by exponential mapping; the gradient and sparse Hessian matrix of the second energy function are calculated by automatic differentiation tools; the inner iteration usually converges in 20-40 steps.
[0117] Step 9: determining the registered point cloud based on the full resolution transform value of the transformation matrix, the full resolution source point cloud and the full resolution target point cloud.
[0118] Specifically, the calculation formula of the registered point cloud is:
[0119]
[0120] wherein P fused is the registered point cloud; is the full resolution target point cloud; R * is the full resolution transform value of the rotation matrix; is the full resolution source point cloud; t * is the full resolution transform value of the translation vector.
[0121] In an exemplary embodiment, a computer device is provided, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the computer program to implement the plant three-dimensional point cloud registration method.
[0122] In an exemplary embodiment, a computer readable storage medium is provided, having stored thereon a computer program, the computer program being executable by a processor to implement the plant three-dimensional point cloud registration method.
[0123] In an exemplary embodiment, a computer program product is provided, comprising a computer program, the computer program being executable by a processor to implement the plant three-dimensional point cloud registration method.
[0124] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal, and its internal structure diagram can be as shown in Figure 2As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with the terminal outside through the network connection. The computer program is executed by the processor to implement a plant three-dimensional point cloud registration method.
[0125] Those skilled in the art can understand that, Figure 2 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.
[0126] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments of each method. Any reference to memory, databases or other media used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc.
[0127] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.
[0128] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.
[0129] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, not all possible combinations of technical features in the above embodiments are described, but as long as the combination of technical features does not exist contradictory, it should be considered as the scope of the present application.
[0130] The principles and implementations of the present application are described in detail with specific examples in this paper, and the above examples are only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the present application.
Claims
1. A method for registering three-dimensional point clouds of plants, characterized in that, The plant three-dimensional point cloud registration method comprises: obtaining a source point cloud and a target point cloud to be registered; the source point cloud and the target point cloud are both plant three-dimensional point clouds; determining the source point cloud and the target point cloud as current point clouds respectively, pre-processing the current point clouds to obtain a full-resolution current point cloud and a down-sampled current point cloud; extracting features from the down-sampled current point cloud to obtain geometric features, color features and fusion features of each point in the down-sampled current point cloud; pairing each point in the down-sampled source point cloud with each point in the down-sampled target point cloud based on the fusion features of each point in the down-sampled source point cloud and the fusion features of each point in the down-sampled target point cloud, to obtain a first point pair set, and determining color similarity weights of each point pair in the first point pair set based on the color features of the two points of each point pair in the first point pair set respectively; using hierarchical non-convex optimization and L-BFGS optimization, and based on the color similarity weights of all point pairs in the first point pair set, iteratively optimizing the transformation matrix to obtain a coarse transformation value of the transformation matrix, and using the coarse transformation value of the transformation matrix to transform the full-resolution source point cloud to obtain a coarsely transformed source point cloud; the first energy function is a function of the color similarity weights between the two points of all point pairs in the first point pair set, the geometric features of the two points of all point pairs in the first point pair set and the transformation matrix; the transformation matrix comprises a rotation matrix and a translation vector; extracting features from the coarsely transformed source point cloud to obtain geometric features, color features and fusion features of each point in the coarsely transformed source point cloud; pairing each point in the coarsely transformed source point cloud with each point in the down-sampled target point cloud based on the fusion features of each point in the coarsely transformed source point cloud and the fusion features of each point in the down-sampled target point cloud, to obtain a second point pair set, and determining color similarity weights of each point pair in the second point pair set based on the color features of the two points of each point pair in the second point pair set respectively; using hierarchical non-convex optimization and L-BFGS optimization, and based on the color similarity weights of all point pairs in the second point pair set, iteratively optimizing the transformation matrix to obtain a full-resolution transformation value of the transformation matrix; the second energy function is a function of the color similarity weights between the two points of all point pairs in the second point pair set, the geometric features of the two points of all point pairs in the second point pair set and the transformation matrix; determining a registered point cloud based on the full-resolution transformation value of the transformation matrix, the full-resolution source point cloud and the full-resolution target point cloud.
2. The method of claim 1, wherein, The preprocessing comprises: noise reduction processing and down-sampling; the pre-processing of the current point cloud to obtain the full-resolution current point cloud and the down-sampled current point cloud comprises: performing noise reduction processing on the current point cloud to obtain the full-resolution current point cloud; down-sampling the full-resolution current point cloud to obtain the down-sampled current point cloud.
3. The method of claim 1, wherein, extracting features from the down-sampled current point cloud to obtain geometric features, color features and fusion features of each point in the down-sampled current point cloud, comprising: respectively calculate the FPFH of each point in the down-sampled current point cloud to obtain the geometric features of each point in the down-sampled current point cloud; respectively convert the RGB values of each point in the down-sampled current point cloud into Lab values to obtain the color features of each point in the down-sampled current point cloud; based on the first preset color weight, the geometric features of each point in the down-sampled current point cloud and the color features of each point in the down-sampled current point cloud, determine the fusion features of each point in the down-sampled current point cloud.
4. The method of claim 1, wherein, based on the fusion features of each point in the down-sampled source point cloud and the fusion features of each point in the down-sampled target point cloud, pair each point in the down-sampled source point cloud with each point in the down-sampled target point cloud to obtain a first set of point pairs, including: pair any point in the down-sampled source point cloud with any point in the down-sampled target point cloud to obtain a plurality of first initial point pairs; respectively determine the norm of the difference between the fusion features of the two points in each first initial point pair, and determine all first initial point pairs with a norm less than a first similarity threshold as the point pairs in the first set of point pairs.
5. The method of claim 1, wherein, The first energy function includes: wherein E1(R, t) is a function value of the first energy function; C law is a first point pair set; w ij is a color similarity weight between p i and p j ; p i is a point in the first point pair set, which belongs to the source point cloud after down-sampling; p j is a point in the first point pair set, which belongs to the target point cloud after down-sampling; p μ is a robust kernel function with a control non-convex degree of μ; R is a rotation matrix; FPFH(p i ) is a geometric feature of p i ; t is a translation vector; FPFH(p j ) is a geometric feature of p j ; ||·|| is a norm.
6. The method of claim 1, wherein, perform feature extraction on the rough-transformed source point cloud to obtain the geometric features, color features and fusion features of each point in the rough-transformed source point cloud, including: respectively calculate the FPFH of each point in the rough-transformed source point cloud to obtain the geometric features of each point in the rough-transformed source point cloud; respectively convert the RGB values of each point in the rough-transformed source point cloud into Lab values to obtain the color features of each point in the rough-transformed source point cloud; based on the second preset color weight, the geometric features of each point in the rough-transformed source point cloud and the color features of each point in the rough-transformed source point cloud, determine the fusion features of each point in the rough-transformed source point cloud.
7. The method of claim 1, wherein, based on the fusion features of each point in the rough-transformed source point cloud and the fusion features of each point in the down-sampled target point cloud, pair each point in the rough-transformed source point cloud with each point in the down-sampled target point cloud to obtain a second set of point pairs, including: pair any point in the rough-transformed source point cloud with any point in the down-sampled target point cloud to obtain a plurality of second initial point pairs; respectively determine the norm of the difference between the fusion features of the two points in each second initial point pair, and determine all second initial point pairs with a norm less than a second similarity threshold as the point pairs in the second set of point pairs.
8. A computer apparatus comprising: A memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the plant three-dimensional point cloud registration method of any one of claims 1-7.
9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the plant three-dimensional point cloud registration method of any one of claims 1-7.
10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the plant three-dimensional point cloud registration method of any one of claims 1-7.