CN114494368A - Low-overlapping-rate point cloud registration method combining dimensionality reduction projection and feature matching - Google Patents
Low-overlapping-rate point cloud registration method combining dimensionality reduction projection and feature matchingDownload PDFInfo
- Publication number
- CN114494368A CN114494368ACN202111558350.6ACN202111558350ACN114494368ACN 114494368 ACN114494368 ACN 114494368ACN 202111558350 ACN202111558350 ACN 202111558350ACN 114494368 ACN114494368 ACN 114494368A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- dimensionality reduction
- point
- matching
- matches
- 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.)
- Granted
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
Abstract
Description
Translated fromChinese技术领域technical field
本发明涉及三维视觉中的点云配准应用技术领域，尤其涉及一种结合降维投影和特征匹配的低重叠率点云配准方法。The invention relates to the technical field of point cloud registration application in three-dimensional vision, in particular to a low overlap rate point cloud registration method combining dimension reduction projection and feature matching.
背景技术Background technique
点云配准的目的是求解出同一坐标下不同姿态点云的变换矩阵，利用该变换矩阵将多视角扫描点云统一在同一坐标系下，最终获取完整的三维模式、场景。近年来，点云配准得到广泛的研究，但传统的点云配准算法性能受限于较好的初始位置和较大的重叠率。其中，重叠率定义为重叠区域的点数在完整点云所占比例，当源点云与目标点云的重叠率低于60％时，两片点云的重叠程度低。因此，针对低重叠率点云的快速而准确的配准算法仍是点云配准的难点。The purpose of point cloud registration is to solve the transformation matrix of point clouds with different attitudes under the same coordinate, and use this transformation matrix to unify the multi-view scanning point cloud in the same coordinate system, and finally obtain the complete three-dimensional mode and scene. In recent years, point cloud registration has been widely studied, but the performance of traditional point cloud registration algorithms is limited by good initial positions and large overlap ratios. Among them, the overlapping rate is defined as the proportion of the points in the overlapping area in the complete point cloud. When the overlapping rate of the source point cloud and the target point cloud is lower than 60%, the overlapping degree of the two point clouds is low. Therefore, a fast and accurate registration algorithm for point clouds with low overlap rate is still a difficult point in point cloud registration.
目前点云配准算法有很多，大体可分为两类，基于全局搜索和基于局部特征的配准算法。At present, there are many point cloud registration algorithms, which can be roughly divided into two categories: global search-based and local feature-based registration algorithms.
基于全局搜索的配准算法，是按照几何位置关系的约束，从待匹配点云中寻找对应点，计算变换矩阵，多次循环迭代得到最优变换矩阵。基于全局搜索的配准算法经典的是迭代最近点(Iterative Closest Point，ICP)算法，但该算法对点云的初始位置要求较高。根据ICP算法又衍生出多种改进形式，如PLICP、GICP等，此类改进算法在点云配准的精度、速度以及抗噪性等方面有了一定程度的提高，但仍无法较好完成低重叠率点云配准；基于概率统计学的配准方法也是以全局为单位，利用概率密度函数来估计点云分布，具有代表性的如Biber等人提出的正态分布变换(Normal distributions transform，NDT)算法；另有其他基于全局搜索的配准算法，如基于随机采样一致性(Random sample consensus，RANSAC)的算法，Mellado等人提出的基于四点一致性(4-points congruent sets，4PCS)的算法及其改进形式D4PCS算法、Super4PCS算法等。但基于全局搜索的配准算法普遍具有计算复杂度高的特点，耗时较长；同时针对低重叠率点云配准，非重叠区域的搜索对配准不起作用，反而增加计算量，降低整体配准效率。The registration algorithm based on global search searches for corresponding points from the point cloud to be matched according to the constraints of the geometric position relationship, calculates the transformation matrix, and obtains the optimal transformation matrix through multiple loop iterations. The classic registration algorithm based on global search is the Iterative Closest Point (ICP) algorithm, but this algorithm has higher requirements on the initial position of the point cloud. According to the ICP algorithm, a variety of improved forms have been derived, such as PLICP, GICP, etc. These improved algorithms have improved the accuracy, speed, and noise resistance of point cloud registration to a certain extent, but they are still unable to perform well. Overlap rate point cloud registration; the registration method based on probability statistics is also based on the global unit, using the probability density function to estimate the point cloud distribution, such as the normal distribution transform proposed by Biber et al. NDT) algorithm; there are other registration algorithms based on global search, such as the algorithm based on random sample consensus (RANSAC), the four-point consensus (4-points congruent sets, 4PCS) proposed by Mellado et al. ) algorithm and its improved form D4PCS algorithm, Super4PCS algorithm and so on. However, registration algorithms based on global search generally have the characteristics of high computational complexity and take a long time; at the same time, for point cloud registration with low overlap rate, the search in non-overlapping areas does not work for registration, but increases the amount of calculation and reduces the Overall registration efficiency.
基于局部特征的配准算法，主要工作在于定义特征描述子，并建立点云间特征的对应关系。Rusu等提出对相邻两个点的法向量夹角进行统计的点特征直方图(Pointfeature histogram，PFH)作为描述子，及其改进形式快速点特征直方图(FPFH)；DanielF.Huber等提出的将点云转化为网格形式的自旋图像(spin images,SI)特征量；Tombari等将点签名和点特征直方图的思想相结合，提出SHOT特征描述子；Andrea等提出三维形状上下文(3DSC)，在对数极坐标系下利用直方图描述形状轮廓特征，是二维形状上下文的扩展；Li等通过对目标三维点云进行聚类，得到小规模、特征明显的聚类点云，有效提高配准效率和精度；除此以外，还有尺度不变特征变换(SIFT)，旋转不变特征变换(RIFT)等常见的局部特征描述子。由于点云局部特征提取算法复杂度高，且易出现误匹配等问题，所以不能直接用于低重叠率点云的配准；同时局部特征提取时易受杂乱和遮挡的滋扰，即当三维点云中存在杂乱和遮挡时，配准的效果将大大降低。The registration algorithm based on local features mainly works on defining feature descriptors and establishing the correspondence between point clouds. Rusu et al. proposed a point feature histogram (PFH) that counts the angle between the normal vectors of two adjacent points as a descriptor, and its improved form, fast point feature histogram (FPFH); Daniel F. Huber et al. proposed Convert the point cloud into a spin image (SI) feature quantity in the form of a grid; Tombari et al. combined the ideas of point signature and point feature histogram to propose the SHOT feature descriptor; Andrea et al. proposed a three-dimensional shape context (3DSC). ), using histograms to describe the shape contour features in the log-polar coordinate system, which is an extension of the two-dimensional shape context; Li et al. Improve registration efficiency and accuracy; in addition, there are common local feature descriptors such as Scale Invariant Feature Transform (SIFT) and Rotation Invariant Feature Transform (RIFT). Due to the high complexity of the local feature extraction algorithm for point clouds, and problems such as mismatching, it cannot be directly used for the registration of point clouds with low overlap rates; at the same time, local feature extraction is easily disturbed by clutter and occlusion. When there is clutter and occlusion in the point cloud, the effect of registration will be greatly reduced.
无论是基于全局搜索还是基于局部特征的传统配准算法，都无法直接应用于低重叠率点云配准，为解决这一难题，很多学者提出了两步配准策略，即先提取目标点和重叠区域，而后在重叠区域范围内迭代搜索得到最优解。Wang等提出一种基于区域分割的重叠区域提取方法，在点云采集视角差异较大的情况下仍能完成对重叠区域的提取，提高了配准效率；Li等提出一种基于贡献因子的改进TrICP算法，算法自动计算重叠度，能对含大量噪声、部分重叠、非同源的激光与影像重建点云进行可靠高效地自动配准；Lu等将点云划分成多个区域，先对每个区域进行D4PCS配准，再放大重叠区域的作用完成全局配准，但算法耗时较长，不适用于大规模点云数据；张元等将区域分块和凸优化结合，分别提取重叠区域、优化对应关系，最后利用ICP算法完成配准；虽然上述算法在一定程度上提高了低重叠率点云配准的精度，但都是在三维空间内进行特征点计算和提取，算法存在计算复杂度高、耗时长等问题。Whether it is based on global search or traditional registration algorithm based on local features, it cannot be directly applied to low overlap rate point cloud registration. The overlapping area is then iteratively searched within the overlapping area to obtain the optimal solution. Wang et al. proposed an overlapping region extraction method based on region segmentation, which can still complete the extraction of overlapping regions in the case of large differences in point cloud collection perspectives, improving the registration efficiency; Li et al. proposed an improvement based on contribution factors TrICP algorithm, the algorithm automatically calculates the degree of overlap, and can perform reliable and efficient automatic registration of laser and image reconstruction point clouds with a lot of noise, partial overlap, and non-homogeneity; Lu et al. Perform D4PCS registration on each area, and then enlarge the overlapping area to complete the global registration, but the algorithm takes a long time and is not suitable for large-scale point cloud data; Zhang Yuan et al. combined regional block and convex optimization to extract overlapping areas separately. , optimize the corresponding relationship, and finally use the ICP algorithm to complete the registration; although the above algorithm improves the accuracy of the low overlap rate point cloud registration to a certain extent, but the calculation and extraction of feature points are performed in three-dimensional space, and the algorithm is computationally complex. problems such as high speed and time-consuming.
发明内容SUMMARY OF THE INVENTION
为了解决现有技术存在的不足，本发明的目的在于提供一种结合降维投影和特征匹配的低重叠率点云配准方法，面向低重叠率点云配准问题，通过将降维投影与图像特征匹配相结合的方式，在降维后的图像空间完成特征点的提取、描述及匹配，解决低重叠率点云特征点提取困难少、易出现误匹配、迭代耗时长等问题。In order to solve the shortcomings of the prior art, the purpose of the present invention is to provide a low overlap rate point cloud registration method combining dimension reduction projection and feature matching. Facing the low overlap rate point cloud registration problem, by combining dimension reduction projection with The combination of image feature matching can complete the extraction, description and matching of feature points in the image space after dimensionality reduction, and solve the problems of low overlap rate point cloud feature points with less difficulty in extraction, prone to false matching, and long iteration time.
为实现上述目的，本发明提供的结合降维投影和特征匹配的低重叠率点云配准方法，包括以下步骤：In order to achieve the above object, the low overlap rate point cloud registration method combining dimension reduction projection and feature matching provided by the present invention includes the following steps:
1)将点云的均值法矢方向作为初始降维方向；1) Take the mean normal vector direction of the point cloud as the initial dimension reduction direction;
2)对所述初始降维方向进行方向矫正，得到最终降维方向；2) Perform direction correction on the initial dimension reduction direction to obtain the final dimension reduction direction;
3)沿所述最终降维方向进行投影生成降维后的深度图像；3) Projecting along the final dimension reduction direction to generate a dimension-reduced depth image;
4)根据栅格大小生成多分辨率图像；4) Generate multi-resolution images according to the grid size;
5)根据投票机制和刚体变换的距离不变性进行两步筛选，得到可靠的匹配点对；5) Perform two-step screening according to the voting mechanism and the distance invariance of rigid body transformation to obtain reliable matching point pairs;
6)利用投票机制和刚体变换的距离不变性进行两步筛选；6) Two-step screening using voting mechanism and distance invariance of rigid body transformation;
7)对最终目标点对求解旋转矩阵，利用ICP算法迭代优化，得到最终的精确转换矩阵。7) Solve the rotation matrix for the final target point pair, and use the ICP algorithm to iteratively optimize to obtain the final accurate transformation matrix.
进一步地，所述步骤1)还包括，Further, the step 1) also includes,
由主成分分析法求解离散点的法向量，pi法向量取矩阵的最小特征值对应的特征向量：The normal vector of the discrete points is solved by the principal component analysis method, and thepi normal vector takes the eigenvector corresponding to the minimum eigenvalue of the matrix:
Eξi＝λiξi，i＝{1，2，3}，Eξi =λi ξi , i={1, 2, 3},
其中R为邻域半径，N为领域内点的个数，pi表示领域内第i个点的三维坐标向量，p′表示领域质心的三维坐标向量，di表示领域内第i个点到质心的欧式距离，ξi为矩阵E的特征向量，i为大于等于1的整数；where R is the radius of the neighborhood, N is the number of points in the field, pi represents the three-dimensional coordinate vector of theith point in the field, p′ represents the three-dimensional coordinate vector of the center of mass of the field, and di represents the ith point in the field to Euclidean distance of centroid, ξi is the eigenvector of matrix E, i is an integer greater than or equal to 1;
pi点法向量取矩阵的最小特征值对应的特征向量，记为n＝(xi，yi，zi)，则确定降维的初始投影方向为点云P各点的均值法矢：The normal vector of point pi takes the eigenvector corresponding to the minimum eigenvalue of the matrix, and denoted as n=(xi , yi , zi ), then the initial projection direction of dimension reduction is determined as the mean normal vector of each point of point cloud P:
进一步地，所述步骤2)还包括，Further, the step 2) also includes,
沿初始投影方向均匀划分空间邻域，并沿多个邻域方向分别投影，得到不同的降维图像；The spatial neighborhood is evenly divided along the initial projection direction, and projected along multiple neighborhood directions to obtain different dimensionality reduction images;
利用哈希算法求出降维图像的hash指纹图，并比较其汉明距离，将距离值最小的图像视为相似度最高的图像，将其对应的方向作为矫正后的方向；Use the hash algorithm to obtain the hash fingerprint of the dimension-reduced image, and compare its Hamming distance. The image with the smallest distance value is regarded as the image with the highest similarity, and the corresponding direction is used as the corrected direction;
对原始点云进行旋转变换，使旋转后的点云空间坐标系的Z轴方向与投影方向对齐：Rotate the original point cloud to align the Z-axis direction of the rotated point cloud space coordinate system with the projection direction:
其中，in,
进一步地，所述步骤3)还包括，Further, the step 3) also includes,
计算每个栅格c内的最大深度值maxD(c)，然后定义点集P′(c)为栅格内最外层的点集：Calculate the maximum depth value maxD (c) in each grid c, and then define the point set P'(c) as the outermost point set in the grid:
P′(c)＝{p∈P||maxD(c)-pz|＜δ1}；P′(c)={p∈P||maxD (c)-pz |<δ1 };
对P′(c)中点的Z坐标进行高斯权重插值，计算出栅格对应的像素值f(c)：Perform Gaussian weight interpolation on the Z coordinate of the midpoint of P'(c) to calculate the pixel value f(c) corresponding to the grid:
其中，归一化参数W表示为：Among them, the normalization parameter W is expressed as:
高斯函数g表示为：The Gaussian function g is expressed as:
插值计算每个栅格对应的像素值，从而得到降维后的深度图像。Interpolate the pixel values corresponding to each grid to obtain a reduced-dimensional depth image.
进一步地，所述步骤4)还包括，Further, the step 4) also includes,
建立栅格大小d与点云离散点的均值间距D的关系：Establish the relationship between the grid size d and the mean distance D of the discrete points of the point cloud:
d＝l·D，d=l·D,
其中，l为系数；Among them, l is the coefficient;
分辨率表示为：Resolution is expressed as:
进一步地，所述步骤5)还包括，Further, the step 5) also includes,
采用ORB算子，对n个不同图层进行ORB匹配，每个图层对应图像特征点对表示为：The ORB operator is used to perform ORB matching on n different layers, and the image feature point pairs corresponding to each layer are expressed as:
matches[i]，i∈[0，n-1]，matches[i], i∈[0, n-1],
其中，matches[0]对应l0下分辨率最高的图层；Among them, matches[0] corresponds to the layer with the highest resolution under l0 ;
统计matches[0]中匹配关系在其他图层的重复次数，当下式成立时票数加一：Count the number of repetitions of the matching relationship in matches[0] in other layers, and add one to the number of votes when the following formula holds:
l0×(matches[0][j].X，matches[0][j].Y)l0 ×(matches[0][j].X，matches[0][j].Y)
＝li×(matches[i][k].X，matches[i][k].Y)=li ×(matches[i][k].X, matches[i][k].Y)
其中，matches[0][j]代表第1个图层中第j对匹配点，matches[i][k]代表第i+1图层的第k对匹配点。Among them, matches[0][j] represents the jth pair of matching points in the first layer, and matches[i][k] represents the kth pair of matching points in the i+1th layer.
进一步地，所述步骤6)还包括，Further, the step 6) also includes,
根据图层数确定票数阈值δ2，大于δ2的匹配点视为可靠匹配点，映射到原始点云得到Matches；Determine the vote number threshold δ2 according to the number of layers, the matching points greater than δ2 are regarded as reliable matching points, and are mapped to the original point cloud to obtain Matches;
Matches中任意两个点对(pi，qi)和(pj，qj)如果是正确的匹配点对，则根据刚体变换的距离不变性存在关系dist(pi，pj)＝dist(qi，qj)；任选两对点，选取δ3＞0，使得：If any two point pairs (pi , qi ) and (pj ,qj ) in Matches are correct matching point pairs, then there is a relationship dist(pi , pj )= dist according to the distance invariance of rigid body transformation (qi , qj ); choose two pairs of points, choose δ3 >0, so that:
其中，分子表示匹配点距离之差，分母是为了降低距离约束条件对点云规模的敏感性，满足上式则确定为满足距离约束的最终的匹配点对。Among them, the numerator represents the difference between the distances of matching points, and the denominator is to reduce the sensitivity of the distance constraint to the point cloud scale, and the final matching point pair that satisfies the distance constraint is determined to satisfy the above formula.
更进一步地，所述步骤7)还包括，Further, the step 7) also includes,
利用四元数法，求得最终的匹配点对的初始转换矩阵，将待配准点云转换到相近的位置；Using the quaternion method, the initial transformation matrix of the final matching point pair is obtained, and the point cloud to be registered is transformed to a similar position;
应用ICP算法，进行迭代计算得到精确的转换矩阵。Applying the ICP algorithm, the iterative calculation is performed to obtain an accurate transformation matrix.
为实现上述目的，本发明还提供一种电子设备，包括存储器和处理器，所述存储器上储存有在所述处理器上运行的计算机程序，所述处理器运行所述计算机程序时执行如上文所述的结合降维投影和特征匹配的低重叠率点云配准方法的步骤。In order to achieve the above object, the present invention also provides an electronic device, comprising a memory and a processor, the memory stores a computer program running on the processor, and the processor executes the above when running the computer program. The steps of the low overlap rate point cloud registration method combining dimension reduction projection and feature matching.
为实现上述目的，本发明还提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序运行时执行如上文所述的结合降维投影和特征匹配的低重叠率点云配准方法的步骤。In order to achieve the above object, the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program runs, the low-overlap rate point cloud matching as described above in combination with dimensionality reduction projection and feature matching is executed. steps of the quasi-method.
本发明的结合降维投影和特征匹配的低重叠率点云配准方法，与现有技术相比较，具有以下有益效果：针对低重叠率点云配准问题，通过将降维投影与图像特征匹配相结合的方式，在降维后的图像空间完成特征点的提取、描述及匹配，不但解决了低重叠率点云特征点提取困难少、易出现误匹配、迭代耗时长等问题，而且提高了配准的精度和效率以及对低重叠率的适应性；随着点云规模越大，速度优势越显著。Compared with the prior art, the low overlap rate point cloud registration method combining dimension reduction projection and feature matching of the present invention has the following beneficial effects: aiming at the low overlap rate point cloud registration problem, by combining dimension reduction projection with image features The combination of matching and the extraction, description and matching of feature points is completed in the image space after dimensionality reduction, which not only solves the problems of less difficulty in extracting feature points of low-overlap rate point clouds, prone to mis-matching, and time-consuming iterations, but also improves the performance. The accuracy and efficiency of registration and the adaptability to low overlap rates are improved; as the size of the point cloud becomes larger, the speed advantage becomes more significant.
本发明的其它特征和优点将在随后的说明书中阐述，并且，部分地从说明书中变得显而易见，或者通过实施本发明而了解。Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the invention.
附图说明Description of drawings
附图用来提供对本发明的进一步理解，并且构成说明书的一部分，并与本发明的实施例一起，用于解释本发明，并不构成对本发明的限制。在附图中：The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and together with the embodiments of the present invention, are used to explain the present invention, but not to limit the present invention. In the attached image:
图1为根据本发明的结合降维投影和特征匹配的低重叠率点云配准方法流程图；1 is a flow chart of a low-overlap rate point cloud registration method combining dimension reduction projection and feature matching according to the present invention;
图2为根据本发明的点云降维示意图；2 is a schematic diagram of point cloud dimensionality reduction according to the present invention;
图3为根据本发明的方向矫正示意图；3 is a schematic diagram of a direction correction according to the present invention;
图4为根据本发明的不同系数降维后的图像效果示意图；4 is a schematic diagram of an image effect after dimension reduction with different coefficients according to the present invention;
图5为根据本发明的特征匹配和位置解算流程图；5 is a flow chart of feature matching and position solution according to the present invention;
图6为根据本发明的方向矫正实验图；Fig. 6 is a direction correction experiment diagram according to the present invention;
图7为根据本发明的降维图像hash指纹图；7 is a dimensionality reduction image hash fingerprint according to the present invention;
图8为根据本发明的特征匹配及筛选结果图；8 is a feature matching and screening result diagram according to the present invention;
图9为根据本发明与其他配准方法所得到的配准结果对比示意图；9 is a schematic diagram showing the comparison of registration results obtained according to the present invention and other registration methods;
图10为根据本发明的误差曲线和速度曲线示意图。FIG. 10 is a schematic diagram of an error curve and a speed curve according to the present invention.
具体实施方式Detailed ways
以下结合附图对本发明的优选实施例进行说明，应当理解，此处所描述的优选实施例仅用于说明和解释本发明，并不用于限定本发明。The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, but not to limit the present invention.
实施例1Example 1
图1为根据本发明的结合降维投影和特征匹配的低重叠率点云配准方法流程图，下面将参考图1，对本发明的结合降维投影和特征匹配的低重叠率点云配准方法进行详细描述。1 is a flow chart of a method for registering a low-overlap rate point cloud according to a combination of dimensionality reduction projection and feature matching of the present invention, below will refer to FIG. method is described in detail.
首先，在步骤101，根据主成分分析法求取点云的法矢信息，并以均值法矢做为降维的初始投影方向。First, in
本发明实施例中，由主成分分析法求解离散点的法向量，pi法向量取矩阵的最小特征值对应的特征向量：In the embodiment of the present invention, the normal vector of the discrete points is solved by principal component analysis, and thepi normal vector is the eigenvector corresponding to the minimum eigenvalue of the matrix:
Eξi＝λiξi，i＝{1，2，3}，Eξi =λi ξi , i={1, 2, 3},
其中，R为邻域半径，N为领域内点的个数，pi表示领域内第i个点的三维坐标向量，p′表示领域质心的三维坐标向量，di表示领域内第i个点到质心的欧式距离，ξi为矩阵E的特征向量，i为大于等于1的整数；Among them, R is the radius of the neighborhood, N is the number of points in the field, pi represents the three-dimensional coordinate vector of theith point in the field, p′ represents the three-dimensional coordinate vector of the center of mass of the field, and di represents the ith point in the field. Euclidean distance to the centroid, ξi is the eigenvector of matrix E, i is an integer greater than or equal to 1;
pi点法向量取矩阵的最小特征值对应的特征向量，记为n＝(xi，yi，zi)，则确定降维的初始投影方向为点云P各点的均值法矢：The normal vector of point pi takes the eigenvector corresponding to the minimum eigenvalue of the matrix, and denoted as n=(xi , yi , zi ), then the initial projection direction of dimension reduction is determined as the mean normal vector of each point of point cloud P:
在步骤102，通过对初始投影方向的邻域划分以及哈希算法进行方向矫正。In
本发明实施例中，为了使点云重叠区沿近似方向降维，以增加图像特征匹配点，对步骤101中得到的初始投影方向进行矫正：In the embodiment of the present invention, in order to reduce the dimensionality of the point cloud overlapping area along the approximate direction, so as to increase the image feature matching points, the initial projection direction obtained in
1)确定初始投影方向后，沿该方向均匀划分空间邻域，并沿多个邻域方向分别投影，得到不同的降维图像；1) After determining the initial projection direction, evenly divide the spatial neighborhood along this direction, and project along multiple neighborhood directions to obtain different dimensionality reduction images;
2)利用哈希算法求出上述图像的hash指纹图，并比较其汉明距离，距离值最小的图像视为相似度最高的图像，以此图像对应的方向作为矫正后的方向；2) utilize the hash algorithm to obtain the hash fingerprint of the above image, and compare its Hamming distance, the image with the smallest distance value is regarded as the image with the highest similarity, and the direction corresponding to this image is used as the corrected direction;
3)对原始点云进行旋转变换，使旋转后的点云空间坐标系的Z轴方向与投影方向对齐：3) Rotate the original point cloud to align the Z-axis direction of the rotated point cloud space coordinate system with the projection direction:
其中，in,
在步骤103，通过高斯函数对投影栅格内的离散点进行插值，得到栅格对应的像素值。In
本发明实施例中，由于像素栅格内可能包含有多个离散点，取其中一点的深度作为像素值无法准确表示区域内的所有点的信息，因此考虑采用高斯权重插值的方法得到像素值。In the embodiment of the present invention, since the pixel grid may contain multiple discrete points, taking the depth of one of the points as the pixel value cannot accurately represent the information of all points in the region, therefore, the Gaussian weight interpolation method is considered to obtain the pixel value.
1)首先计算每个栅格c内的最大深度值maxD(c)，然后定义点集P′(c)为栅格内最外层的点集：1) First calculate the maximum depth value maxD (c) in each grid c, and then define the point set P'(c) as the outermost point set in the grid:
P′(c)＝{p∈P||maxD(c)-pz|＜δ1}P′(c)={p∈P||maxD (c)-pz |＜δ1 }
2)对P′(c)中点的Z坐标进行高斯权重插值，计算出栅格对应的像素值f(c)：2) Perform Gaussian weight interpolation on the Z coordinate of the midpoint of P'(c), and calculate the pixel value f(c) corresponding to the grid:
其中，归一化参数W表示为：Among them, the normalization parameter W is expressed as:
高斯函数g表示为：The Gaussian function g is expressed as:
插值计算每个栅格对应的像素值，从而得到降维后的深度图像。Interpolate the pixel values corresponding to each grid to obtain a reduced-dimensional depth image.
图2为根据本发明的点云降维示意图，如图2所示，首先由点云的法矢信息和方向矫正，确定投影方向，对原始点云数据进行旋转、平移变换，将均值法矢与z轴方向对齐；然后将点云数据向xoy平面投影，根据点云的均值间距确定分辨率并进行栅格化；最后对每一栅格内的Z坐标进行插值得到像素值，将三维点云转换为二维深度图。Figure 2 is a schematic diagram of point cloud dimensionality reduction according to the present invention. As shown in Figure 2, first, the normal vector information and direction correction of the point cloud are used to determine the projection direction, and the original point cloud data are rotated and translated, and the mean normal vector Align with the z-axis direction; then project the point cloud data to the xoy plane, determine the resolution and rasterize according to the mean spacing of the point cloud; finally, interpolate the Z coordinate in each grid to obtain the pixel value, and convert the three-dimensional point Clouds are converted to 2D depth maps.
图3为根据本发明的方向矫正示意图，如图3所示，对于部分重叠的点云，正确的匹配关系主要集中于重叠区内，而对于此类点云降维后的图像空间，获取这些正确匹配关系的前提是重叠区沿近似相同的方向投影，本发明在降维过程中通过方向矫正来获取这一方向。Fig. 3 is a schematic diagram of direction correction according to the present invention. As shown in Fig. 3, for partially overlapping point clouds, the correct matching relationship is mainly concentrated in the overlapping area. The premise of the correct matching relationship is that the overlapping regions are projected in approximately the same direction, and the present invention obtains this direction through direction correction in the process of dimensionality reduction.
将目标点云初始方向的邻域空间均匀划分为k部分，取与初始方向夹角为θ(可取30°或45°)的方向作为候选。The neighborhood space of the initial direction of the target point cloud is evenly divided into k parts, and the direction with an included angle of θ (preferably 30° or 45°) with the initial direction is taken as a candidate.
如图3所示，n1、n2、...、nk为均匀分布在
图4为根据本发明的不同系数l降维后的图像效果示意图，如图4所示，不同系数l降维得到的图像效果，图4中(a)：l＝1；图4中(b)：l＝1.5；图4中(c)：l＝2；图4中(d)l＝3；由图4可知，当l≦1.2时，在图像中不均匀分布一些像素值为0的点(异常点)，这是由于栅格较小，无投影点落在其中导致，这种情况下将在特征提取中增加较多的伪特征点；当l≧3时，图像在边缘处出现较为明显的锯齿状，同时还丢失了细微特征，这是由于栅格较大，栅格内包含较多的点，经过插值造成细节丢失，可提取的特征点将大幅度减少。当1.2<l<3时，图像分辨率较为合适，降维过程既能减少点信息丢失，又能准确表达点云信息。综上，在(1.2,3)区间内选取多组l，生成多分辨率降维图像，用于进一步的特征匹配。4 is a schematic diagram of the image effect after dimension reduction with different coefficients l according to the present invention. As shown in FIG. 4 , the image effect obtained by dimension reduction with different coefficients l, in FIG. ): l=1.5; (c) in Figure 4: l=2; (d) l=3 in Figure 4; it can be seen from Figure 4 that when l≦1.2, some pixels with a value of 0 are unevenly distributed in the image Points (abnormal points), which are caused by the small grid and no projected points falling in it. In this case, more pseudo feature points will be added to the feature extraction; when l≧3, the image appears at the edge The more obvious jagged, but also the loss of subtle features, this is because the grid is large, the grid contains more points, the details are lost after interpolation, and the feature points that can be extracted will be greatly reduced. When 1.2<l<3, the image resolution is more suitable, and the dimensionality reduction process can not only reduce the loss of point information, but also accurately express the point cloud information. In summary, multiple groups of l are selected in the (1.2, 3) interval to generate multi-resolution dimensionality-reduced images for further feature matching.
在步骤104，根据点云的均值间距选取不同大小的栅格，以生成不同分辨率的降维图像。In
本发明实施例中，不同设备获取的点云，点集规模不同，分辨率也不同，为了使配准方法对点云规模具有自适应性，在降维过程中建立栅格大小d与点云离散点的均值间距D的关系：In the embodiment of the present invention, the point clouds acquired by different devices have different point set scales and different resolutions. In order to make the registration method adaptive to the point cloud scale, the grid size d and the point cloud scale are established in the process of dimensionality reduction. The relationship between the mean distance D of discrete points:
d＝l·Dd=l·D
栅格大小d决定了栅格内离散点的个数，也即决定了降维后图像的分辨率，分辨率可表示为：The grid size d determines the number of discrete points in the grid, that is, the resolution of the image after dimensionality reduction. The resolution can be expressed as:
在步骤105，根据分辨率划分图层，不同图层间分别进行ORB特征匹配，统计匹配关系重复出现的票数。In
本发明实施例中，由步骤104生成多分辨率图像后，在图像特征匹配中，本发明采用ORB算子，对n个不同图层进行ORB匹配，每个图层对应图像特征点对表示为matches[i]，i∈[0，n-1]，其中matches[0]对应l0下分辨率最高的图层。统计matches[0]中匹配关系在其他图层的重复次数，当下式成立时票数加一：In the embodiment of the present invention, after the multi-resolution image is generated in
l0×(matches[0][j].X，matches[0][j].Y)l0 ×(matches[0][j].X，matches[0][j].Y)
＝li×(matches[i][k].X，matches[i][k].Y)=li ×(matches[i][k].X, matches[i][k].Y)
其中matches[0][j]代表第1个图层中第j对匹配点，matches[i][k]代表第i+1图层的第k对匹配点。where matches[0][j] represents the jth pair of matching points in the first layer, and matches[i][k] represents the kth pair of matching points in the i+1th layer.
在步骤106，根据投票机制和刚体变换的距离不变性进行两次筛选，得到可靠的匹配关系。In
本发明实施例中，根据图层数确定票数阈值δ2，大于δ2的匹配点视为可靠匹配点，映射到原始点云得到Matches。Matches中任意两个点对(pi，qi)和(pj，qj)如果是正确的匹配点对，那么根据刚体变换的距离不变性存在关系dist(pi，pj)＝dist(qi，qj)，故任选两对点，选取适当的δ3＞0，使得：In the embodiment of the present invention, the vote number threshold δ2 is determined according to the number of layers, and matching points greater than δ2 are regarded as reliable matching points, and are mapped to the original point cloud to obtain Matches. If any two point pairs (pi , qi ) and (pj ,qj ) in Matches are correct matching point pairs, then there is a relationship dist(pi , pj )= dist according to the distance invariance of rigid body transformation (qi , qj ), so choose two pairs of points, and select the appropriate δ3 > 0, so that:
其中，分子表示匹配点距离之差，分母是为了降低距离约束条件对点云规模的敏感性，满足上式则确定为满足距离约束的最终的匹配点对。Among them, the numerator represents the difference between the distances of matching points, and the denominator is to reduce the sensitivity of the distance constraint to the point cloud scale, and the final matching point pair that satisfies the distance constraint is determined to satisfy the above formula.
在步骤107，对最终匹配点应用四元数法，求取其位置转换矩阵。In
本发明实施例中，经过步骤106得到最终的匹配点对，经过四元数法求得初始转换矩阵R、T，将待配准点云在此基础下转换到相近的位置。In the embodiment of the present invention, the final matching point pair is obtained through
在步骤108，利用ICP算法迭代计算，求出配准的精确解。In
本发明实施例中，经过四元数法初配准，待配准点云转换到相近的位置后，对待配准点云应用ICP算法，只需较少次数的迭代计算即可得到精确的转换矩阵。In the embodiment of the present invention, after initial registration by the quaternion method, after the point cloud to be registered is converted to a similar position, the ICP algorithm is applied to the point cloud to be registered, and an accurate transformation matrix can be obtained with a small number of iterative calculations.
图5为根据本发明的特征匹配和位置解算流程图，下面将参考图5，对本发明的特征匹配和位置解算流程进行进一步描述。FIG. 5 is a flow chart of feature matching and position calculation according to the present invention. Referring to FIG. 5 , the flow of feature matching and position calculation of the present invention will be further described below.
在步骤501，将二维降维图像进行特征匹配：将待配准点云在同一系数下降维图像视作一个图层，在不同图层下分别进行图像特征匹配；In step 501, feature matching is performed on the two-dimensional dimensionality reduction image: the point cloud to be registered is regarded as a layer in the dimensionality reduction image with the same coefficient, and image feature matching is performed under different layers respectively;
在步骤502，进行特征点筛选，得到三维匹配点对：统计在不同图层下同一匹配点对重复出现的次数，票数小于阈值的匹配点对视为不可靠匹配，予以剔除；将保留的匹配点对映射回点云，得到三维匹配点对，根据刚体变换的距离不变性进行筛选得到最终目标点对；In step 502, screening of feature points is performed to obtain three-dimensional matching point pairs: the number of times the same matching point pair appears repeatedly under different layers is counted, and the matching point pairs whose votes are less than the threshold are regarded as unreliable matches and are eliminated; The point pair is mapped back to the point cloud to obtain a three-dimensional matching point pair, and the final target point pair is obtained by screening according to the distance invariance of rigid body transformation;
在步骤503，转换矩阵计算：利用四元数法解算目标点对间的位置关系，从而完成点云的初始配准；最后采用ICP细化配准，得到精确结果。In step 503, conversion matrix calculation: use the quaternion method to solve the positional relationship between the target point pairs, thereby completing the initial registration of the point cloud; finally, use ICP to refine the registration to obtain accurate results.
图6为根据本发明的方向矫正实验图，如图6所示，图6中(a)为候选方向；，图6中(b)为方向降维图像；图6中(c)为方向降维图像；图6中(d)为方向降维图像，待配准点云初始方向角差别较大，分别沿初始方向降维得到的图像，不适合进行特征匹配，易出现较多的错误匹配关系，因此需要进行方向矫正。Fig. 6 is an experimental diagram of direction correction according to the present invention, as shown in Fig. 6, Fig. 6 (a) is a candidate direction; Fig. 6 (b) is a direction dimension reduction image; Fig. 6 (c) is a direction reduction image Dimensional image; (d) in Figure 6 is an image of directional dimensionality reduction. The initial directional angle of the point cloud to be registered is quite different, and the images obtained by dimensionality reduction along the initial direction are not suitable for feature matching, and many wrong matching relationships are prone to occur. , so orientation correction is required.
图7为根据本发明的降维图像hash指纹图，如图7所示，利用pHash算法解得q0至qk与p的相似度。图7中的(a)-(d)对应p、q0、q1、q2的hash指纹图，其中计算汉明距离得到图7中的(a)与(d)相似度最高，故取(d)对应的q2作为Q方向矫正后的降维图像。FIG. 7 is a hash fingerprint of a dimension-reduced image according to the present invention. As shown in FIG. 7 , the similarity between q0 to qk and p is obtained by using the pHash algorithm. (a)-(d) in Figure 7 correspond to the hash fingerprints of p, q0 , q1 , and q2 , in which the Hamming distance is calculated to obtain the highest similarity between (a) and (d) in Figure 7, so take (d) The corresponding q2 is used as the dimension-reduced image corrected in the Q direction.
图8为根据本发明的特征匹配及筛选结果图，如图8所示，确定降维方向后改变取值生成多分辨率图像，在每个图层分别采用ORB算法进行特征点匹配，筛除票数较低的特征点对，得到图像匹配结果；然后将匹配特征点映射到点云，根据刚体变换的距离不变性进行第二次筛选，得到匹配结果如下图所示。Fig. 8 is a result diagram of feature matching and screening according to the present invention. As shown in Fig. 8, after determining the dimension reduction direction, the value is changed to generate a multi-resolution image. The feature point pair with the lower number of votes is used to obtain the image matching result; then the matching feature point is mapped to the point cloud, and the second screening is performed according to the distance invariance of rigid body transformation, and the matching result is obtained as shown in the following figure.
表1给出了配准方法各步骤耗时。该配准算法耗时比例较高的是特征匹配与方向矫正，这是由于特征匹配经过不同图层多次匹配，方向矫正包含点云的多次降维运算占单独运算。Table 1 shows the time-consuming of each step of the registration method. The most time-consuming part of the registration algorithm is feature matching and direction correction. This is because the feature matching is matched by different layers for many times, and the direction correction includes multiple dimensionality reduction operations of the point cloud, which are separate operations.
表1配准方法各步骤耗时Table 1 Time-consuming of each step of the registration method
图9为根据本发明与其他配准方法所得到的配准结果对比示意图，如图9所示，结合图9与表2，可以看出直接对低重叠率点云应用ICP算法，会导致配准失败；Super4PCS算法也不适用于重叠率低且规模大的点云配准；而本发明方法对于重叠率低至30％的点云能获得较好的配准效果。Fig. 9 is a schematic diagram showing the comparison of registration results obtained according to the present invention and other registration methods. As shown in Fig. 9, in combination with Fig. 9 and Table 2, it can be seen that directly applying the ICP algorithm to the low overlap rate point cloud will result in the The registration fails; the Super4PCS algorithm is also not suitable for point cloud registration with low overlap rate and large scale; and the method of the present invention can obtain a better registration effect for point clouds with an overlap rate as low as 30%.
表2配准结果对比Table 2 Comparison of registration results
图10为根据本发明的误差曲线和速度曲线示意图，如图10所示，本发明的结合降维投影和特征匹配的低重叠率点云配准方法，适用于低重叠率点云配准，相比于列举的其他算法，本发明在速度和精度上皆具优势，且随着点云规模越大，速度优势越显著。Fig. 10 is a schematic diagram of an error curve and a velocity curve according to the present invention. As shown in Fig. 10 , the low-overlap rate point cloud registration method of the present invention, which combines dimensionality reduction projection and feature matching, is suitable for low-overlap rate point cloud registration. Compared with other listed algorithms, the present invention has advantages in both speed and accuracy, and as the point cloud scale becomes larger, the speed advantage becomes more significant.
本发明的一个实施例中，还提供一种电子设备，包括存储器和处理器，所述存储器上储存有在所述处理器上运行的计算机程序，所述处理器运行所述计算机程序时执行如上文所述的适用于大规模集成电路后端设计的内存管理方法的步骤。In an embodiment of the present invention, an electronic device is also provided, including a memory and a processor, the memory stores a computer program running on the processor, and the processor executes the above when running the computer program The steps of the memory management method described in this paper are suitable for the back-end design of large-scale integrated circuits.
本发明的一个实施例中，还提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序运行时执行如上文所述的适用于大规模集成电路后端设计的内存管理方法的步骤。In one embodiment of the present invention, there is also provided a computer-readable storage medium on which a computer program is stored, and when the computer program runs, the memory management method applicable to the back-end design of large-scale integrated circuits as described above is executed. A step of.
本发明的适用于大规模集成电路后端设计的内存管理方法，采用虚拟内存技术，内存二进制存储技术以及脏页标记技术，统一实现了高效的适用于大规模集成电路后端设计的内存管理方法，能够快速的进行数据的存储以及加载，利用内存二进制直接存储结合脏页标记技术，能够自然的实现脏页保存技术以适应未来分布式云化的技术发展趋势；同时利用linux系统虚拟内存文件隐射的Swap机制，可以把冷数据放置在硬盘以突破实际物理内存容量的限制。The memory management method suitable for the back-end design of the large-scale integrated circuit of the present invention adopts the virtual memory technology, the memory binary storage technology and the dirty page marking technology to uniformly realize an efficient memory management method suitable for the back-end design of the large-scale integrated circuit. , can quickly store and load data, use memory binary direct storage combined with dirty page marking technology, can naturally realize dirty page storage technology to adapt to the future trend of distributed cloud technology development; at the same time use Linux system virtual memory file to hide The Swap mechanism of injection can place cold data on the hard disk to break through the limit of actual physical memory capacity.
本领域普通技术人员可以理解：以上所述仅为本发明的优选实施例而已，并不用于限制本发明，尽管参照前述实施例对本发明进行了详细的说明，对于本领域的技术人员来说，其依然可以对前述各实施例记载的技术方案进行修改，或者对其中部分技术特征进行等同替换。凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。Those of ordinary skill in the art can understand that the above are only preferred embodiments of the present invention, and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, It is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements to some of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111558350.6ACN114494368B (en) | 2021-12-20 | 2021-12-20 | A low overlap point cloud registration method combining dimensionality reduction projection and feature matching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111558350.6ACN114494368B (en) | 2021-12-20 | 2021-12-20 | A low overlap point cloud registration method combining dimensionality reduction projection and feature matching |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114494368Atrue CN114494368A (en) | 2022-05-13 |
| CN114494368B CN114494368B (en) | 2024-11-08 |
Family
ID=81493372
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111558350.6AActiveCN114494368B (en) | 2021-12-20 | 2021-12-20 | A low overlap point cloud registration method combining dimensionality reduction projection and feature matching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114494368B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114882086A (en)* | 2022-05-26 | 2022-08-09 | 南京工业大学 | Point cloud accurate registration method based on optimal slice transmission |
| CN114972448A (en)* | 2022-05-26 | 2022-08-30 | 合肥工业大学 | ICP algorithm-based dimensionality reduction acceleration point cloud registration method |
| CN115267811A (en)* | 2022-06-22 | 2022-11-01 | 上海擎朗智能科技有限公司 | Positioning method for robot and robot using the same |
| CN116245921A (en)* | 2022-12-20 | 2023-06-09 | 华中科技大学 | A fine registration method for 3D measurement point clouds with low overlap rate and weak features by introducing plane constraints |
| CN116740156A (en)* | 2023-08-10 | 2023-09-12 | 西南交通大学 | Registration method of arbitrary pose construction element based on Gaussian sphere and principal plane distribution |
| CN117132850A (en)* | 2023-09-07 | 2023-11-28 | 北京百度网讯科技有限公司 | Model pre-training and training methods, point cloud detection and segmentation methods and devices |
| CN117961197A (en)* | 2024-04-01 | 2024-05-03 | 贵州大学 | Self-adaptive deviation rectifying method of unmanned turbine blade micropore electric machining unit |
| CN119672076A (en)* | 2024-11-19 | 2025-03-21 | 中国科学技术大学 | Point cloud registration method, device and equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105469388A (en)* | 2015-11-16 | 2016-04-06 | 集美大学 | Building point cloud registration algorithm based on dimension reduction |
| CN112001955A (en)* | 2020-08-24 | 2020-11-27 | 深圳市建设综合勘察设计院有限公司 | Point cloud registration method and system based on two-dimensional projection plane matching constraint |
| US11037346B1 (en)* | 2020-04-29 | 2021-06-15 | Nanjing University Of Aeronautics And Astronautics | Multi-station scanning global point cloud registration method based on graph optimization |
- 2021
- 2021-12-20CNCN202111558350.6Apatent/CN114494368B/enactiveActive
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105469388A (en)* | 2015-11-16 | 2016-04-06 | 集美大学 | Building point cloud registration algorithm based on dimension reduction |
| US11037346B1 (en)* | 2020-04-29 | 2021-06-15 | Nanjing University Of Aeronautics And Astronautics | Multi-station scanning global point cloud registration method based on graph optimization |
| CN112001955A (en)* | 2020-08-24 | 2020-11-27 | 深圳市建设综合勘察设计院有限公司 | Point cloud registration method and system based on two-dimensional projection plane matching constraint |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114882086A (en)* | 2022-05-26 | 2022-08-09 | 南京工业大学 | Point cloud accurate registration method based on optimal slice transmission |
| CN114972448A (en)* | 2022-05-26 | 2022-08-30 | 合肥工业大学 | ICP algorithm-based dimensionality reduction acceleration point cloud registration method |
| CN115267811A (en)* | 2022-06-22 | 2022-11-01 | 上海擎朗智能科技有限公司 | Positioning method for robot and robot using the same |
| CN116245921A (en)* | 2022-12-20 | 2023-06-09 | 华中科技大学 | A fine registration method for 3D measurement point clouds with low overlap rate and weak features by introducing plane constraints |
| CN116740156A (en)* | 2023-08-10 | 2023-09-12 | 西南交通大学 | Registration method of arbitrary pose construction element based on Gaussian sphere and principal plane distribution |
| CN116740156B (en)* | 2023-08-10 | 2023-11-03 | 西南交通大学 | Registration method of arbitrary pose construction element based on Gaussian sphere and principal plane distribution |
| CN117132850A (en)* | 2023-09-07 | 2023-11-28 | 北京百度网讯科技有限公司 | Model pre-training and training methods, point cloud detection and segmentation methods and devices |
| CN117961197A (en)* | 2024-04-01 | 2024-05-03 | 贵州大学 | Self-adaptive deviation rectifying method of unmanned turbine blade micropore electric machining unit |
| CN119672076A (en)* | 2024-11-19 | 2025-03-21 | 中国科学技术大学 | Point cloud registration method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114494368B (en) | 2024-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114494368A (en) | Low-overlapping-rate point cloud registration method combining dimensionality reduction projection and feature matching | |
| CN109887015B (en) | A Point Cloud Automatic Registration Method Based on Local Surface Feature Histogram | |
| CN112446327B (en) | Remote sensing image target detection method based on non-anchor frame | |
| CN111553409B (en) | A Point Cloud Recognition Method Based on Voxel Shape Descriptors | |
| CN114677418B (en) | Registration method based on point cloud feature point extraction | |
| JP5705147B2 (en) | Representing 3D objects or objects using descriptors | |
| Salti et al. | SHOT: Unique signatures of histograms for surface and texture description | |
| JP6216508B2 (en) | Method for recognition and pose determination of 3D objects in 3D scenes | |
| CN117788539B (en) | Point cloud data registration method, system and electronic device | |
| CN110942515A (en) | Point cloud-based target object three-dimensional computer modeling method and target identification method | |
| CN114358166B (en) | Multi-target positioning method based on self-adaptive k-means clustering | |
| CN114648445B (en) | A multi-view high-resolution point cloud stitching method based on feature point extraction and fine registration optimization | |
| CN108830888B (en) | Coarse matching method based on improved multi-scale covariance matrix characteristic descriptor | |
| CN113160129B (en) | Combined type simplified point cloud data rapid registration method | |
| Wang et al. | Robust line feature matching based on pair-wise geometric constraints and matching redundancy | |
| Wang et al. | Two-view geometry estimation using RANSAC with locality preserving constraint | |
| CN112652003A (en) | Three-dimensional point cloud registration method based on RANSAC measure optimization | |
| CN117274339A (en) | Point cloud registration method based on improved ISS-3DSC characteristics combined with ICP | |
| CN115564811A (en) | Point cloud registration method for transformer wiring terminal | |
| WO2024103846A1 (en) | Three-dimensional registration reconstruction method based on multi-domain multi-dimensional feature map | |
| CN116110040A (en) | Substation equipment identification method and device based on combination of SHOT features and ESF features | |
| CN116630662A (en) | Feature point mismatching eliminating method applied to visual SLAM | |
| CN118967702A (en) | Position estimation method, medium and device combining two-dimensional image and three-dimensional point cloud | |
| CN111127667B (en) | Point cloud initial registration method based on region curvature binary descriptor | |
| CN116416287A (en) | Point cloud registration method based on feature description of key points and neighborhood points, electronic equipment and storage medium |
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 |