WO2023134000A1 - 一种多维空间中的直线、平面和超平面的快速检测方法 - Google Patents
一种多维空间中的直线、平面和超平面的快速检测方法 Download PDFInfo
- Publication number
- WO2023134000A1 WO2023134000A1 PCT/CN2022/080650 CN2022080650W WO2023134000A1 WO 2023134000 A1 WO2023134000 A1 WO 2023134000A1 CN 2022080650 W CN2022080650 W CN 2022080650W WO 2023134000 A1 WO2023134000 A1 WO 2023134000A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- node
- data
- votes
- hyperplane
- space
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000001514 detection method Methods 0.000 claims description 39
- 238000012360 testing method Methods 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 abstract description 20
- 238000004364 calculation method Methods 0.000 abstract description 10
- 230000008569 process Effects 0.000 abstract description 10
- 238000013178 mathematical model Methods 0.000 abstract description 8
- 230000011218 segmentation Effects 0.000 abstract description 5
- 230000008901 benefit Effects 0.000 abstract description 4
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 238000010191 image analysis Methods 0.000 abstract description 2
- 230000000007 visual effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 9
- 238000013507 mapping Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 229910052704 radon Inorganic materials 0.000 description 1
- SYUHGPGVQRZVTB-UHFFFAOYSA-N radon atom Chemical compound [Rn] SYUHGPGVQRZVTB-UHFFFAOYSA-N 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
- G06V10/422—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation for representing the structure of the pattern or shape of an object therefor
Definitions
- the invention belongs to the technical field of image processing, and mainly relates to an integrated fast Hough transform, specifically a fast detection method for straight lines, planes and hyperplanes in a multi-dimensional space, which can be used for data mining, classification and pattern recognition of various data , including image data, financial data, array data, etc.; it can also be applied to image analysis, computer vision, autonomous driving, artificial intelligence, and data classification.
- Hough transform an efficient line detection algorithm in images, which is called Hough transform. So far, more than 2500 academic articles have proposed various improved algorithms of Hough transform.
- the 2D Hough transform has also been extended to plane and hyperplane detection in multi-dimensional space, and is used in many fields, including computer vision, machine learning, artificial intelligence, automatic driving and data classification.
- the goal of the Hough transform and most of its variants is the detection of straight lines in two-dimensional data, and it can also be used for the detection of specific curves and graphics through appropriate improvements.
- the use of Hough transform for straight line detection mainly includes two steps: setting a parameter space with limited length, designing a mapping function to convert the input data into a straight line in the parameter space; dividing the parameter space into multiple small counting units, using the input The data votes for each counting unit. The calculation units whose votes exceed the specified threshold correspond to the targets detected by the Hough transform.
- the algorithm flow is shown in Figure 1.
- F(j) [F 1 (j), F 2 (j),...,F n (j) ].
- a i (j) is a function of F(j), satisfying
- ⁇ 1 and y mx+c,
- the adaptive Hough transform Illingworth and Kittler, 1987).
- ⁇ 1 or x m′y+c′,
- the space consists of two subspaces (m,c) and (m',c').
- mapping data points (x, y) to a line a 0 +a 1 m+a 2 c 0(
- ⁇ 1) in the first subspace (m,c) of the line is and
- mapping the data point (x, y) to the straight line a 0 + a 1 m'+a 2 c' 0(
- ⁇ 1) in the second subspace (m',c') of the line is and
- the fast Hough transform uses an iterative algorithm to divide the parameter space into nested hypercubes from low to high resolution, and uses n k-byte trees to express and store.
- Each hypercube corresponds to a specific precision hyperplane in the data space, which we call an accumulator.
- each subspace is responsible for its information storage by an independent k-byte tree.
- Each subspace can be divided into k hypercubes, corresponding to k nodes of the k-byte tree, and each hypercube can be iteratively decomposed into k smaller hypercubes to form a deeper node of the k-byte tree.
- a hypercube does a deeper decomposition only when its corresponding hyperplane is voted by enough data points, and the subdivision stops when the size of the cube reaches the specified accuracy. This layered model greatly saves computation and storage space.
- the purpose of the present invention is to provide a fast detection method for straight lines, planes and hyperplanes in a multi-dimensional space, so as to overcome the deficiencies of the current prior art.
- the present invention provides a fast detection method for straight lines, planes and hyperplanes in a multi-dimensional space.
- the above method uses a straight line, plane or hyperplane in the data space as To represent, among them: ⁇ 1 ⁇ 0, d is a parameter specified by the user, ⁇ 1 , ⁇ 2 ,..., ⁇ n+1 ⁇ is a parameter, n corresponds to the dimension of the data, m is n-1 or n, ⁇ i is a data regularization
- the conversion factor is selected by the user.
- the method pre-sets system parameters through the following steps:
- the third step is to create a k-byte tree, and set the center position parameter of the root node of the k-byte tree and the half-length of each dimension;
- the fourth step is to convert each data point to the parameter space
- the fifth step is to calculate the distance from the root node to any data point j,
- test formula Whether it is satisfied, if it is satisfied, increase the number of votes accumulated by the root node;
- the seventh step is to use all N data points to vote for new nodes
- the eighth step is to detect the number of votes of the new node, if it is less than the threshold value T, stop processing the child node; if the level of the node has reached the q layer, then output the parameter information of the node, which corresponds to the detected target, and stop analyzing the node;
- the tenth step iteratively repeat the seventh to ninth steps until no new nodes are generated in the system and all nodes are processed.
- d is set to 0.
- the voting rule is to calculate the distance from any data point j to the child node:
- the present invention provides a computer device, including a memory, a processor, and computer-readable instructions stored in the memory and operable on the processor, and the processor executes the computer-readable instructions. Instructions implement the fast detection method described above.
- the present invention provides one or more readable storage media storing computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute Rapid detection method as described above.
- the invention establishes new mathematical models of straight lines, planes and hyperplanes, and develops an integrated fast Hough transform based on this.
- the integrated fast Hough transform allows us to use a single k-byte tree to be responsible for all the information in the search space, greatly reducing the amount of computation and storage requirements in the system through de-redundancy.
- This new approach also has two important advantages. First, the fitting mode of the fast Hough transform corresponds to the least squares method, while the integrated fast Hough transform corresponds to the total least squares fitting algorithm.
- the former assumes that the noise of the data only exists in one dimension, while the latter model assumes that the data noise is spread across all dimensions, so it has a better tolerance to data noise and solves the problem of fast Hough transform in practical applications.
- the detection accuracy of the target located on the parameter space segmentation line is too low.
- the targets that are close to each other in the data space are gathered together in the parameter space, which allows us to use intuitive graphics to display the data analysis process in the parameter space, so as to quickly judge The number of objects, visually guides us to set the system parameters and distinguish between repeatedly identified objects.
- Figure 1 is a standard flow chart of Hough transform used for straight line or plane detection.
- Figure 2 is a schematic diagram of testing the intersection of a straight line and a square. You can directly test whether the straight line intersects with a square, or you can test whether the straight line intersects with the circumcircle of the square. The latter will save a lot of time in calculation.
- Figure 3 is a schematic diagram of the parameter space of the multi-scale layered and refined integrated Hough transform; the counting unit has the same length in each dimension, and IFHT only needs to analyze the counting unit that intersects the unit circle in the n-1 dimension .
- Figure 4 is a schematic diagram of the entire parameter space represented by a k-byte tree; we use a vector b composed of 0 and 1 to represent different nodes, and each node generates 2 k child nodes except the root node, and the root node contains 2 k-1 child nodes node.
- Figure 7 is a schematic diagram of different objective functions used in FHT and IFHT line detection; where (a) is FHT corresponding to the least squares method fitting, minimizing the total square value of the deviation along the y-axis, (b) is IFHT corresponding to the total least squares A multiplicative fit that minimizes the sum of all Euclidean distance squared values.
- Figure 8 is a schematic diagram of the comparison of FHT and IFHT line detection performance. Two levels of noise are respectively added to the second dimension of the data; where (a) is the general line detection, and (b) is the parameters of the line corresponding to IFHT and FHT When the boundary value of the counting unit on the plane, the performance of FHT drops sharply, while the performance of IFHT remains basically unchanged.
- Figure 9 is a schematic diagram of the algorithm performance comparison of IFHT and FHT when the noise is uniformly distributed in all dimensions of the data; the standard deviations of the simulated noise are 0.02 and 0.067, respectively.
- FIG. 10 is a schematic diagram of a computer system that can be used to carry the present invention.
- the integrated fast Hough transform adopts a more intuitive mathematical model to represent the plane and hyperplane in straight line and multi-dimensional space.
- the integrated fast Hough transform utilizes the following mathematical model to characterize the hyperplane:
- ⁇ i is a data regularization factor, and the specific selection will be discussed later.
- ⁇ 1 ⁇ 0 we can simplify the above formula and ignore the condition that ⁇ 1 ⁇ 0 for now.
- ⁇ 1 , ⁇ 2 ,..., ⁇ n+1 ⁇ we know the value range of ⁇ 1 , ⁇ 2 ,..., ⁇ n+1 ⁇ , and assume that the half-length of their value range is ⁇ L 1 ,L 2 ,...,L n+1 ⁇ , the above formula can be Converted to:
- the detection target in the data space ⁇ F 1 ,F 2 ,...,F n ⁇ and ⁇ 1 ⁇ 0, transformed into the parameter space ⁇ X 1 ,X 2 ,...,X k ⁇ as
- the range of each dimension Xi is [-1,1], so the entire parameter space can be subdivided into multiple small hypercubes (counting units) for vote.
- the condition ⁇ 1 ⁇ 0 can also be realized simply by ignoring the counting units that do not meet the requirement.
- ⁇ 1 for i 1, 2, 3, the parameter space ( ⁇ 1 , ⁇ 2 , ⁇ 3 ) has a length of 2 in each dimension and can be subdivided into equal-length cubes as vote counting units .
- Each data point in the data space corresponds to a line or plane in the IFHT parameter space.
- measurement errors and noise in the data correspond to offsets of the corresponding objects in the parameter space.
- the IFHT parameter space by detecting a small interval instead of only detecting a specific value, noise tolerance can be achieved and detection accuracy can be improved.
- the length of the detected small interval corresponds to the accuracy of our final detection target.
- IFHT will first divide the parameter space into many small counting units, so that each data point can count voting units. Since IFHT has the same length in all dimensions of the parameter space, the divided sub-counting units are equal in all dimensions.
- the ticket counting units are composed of squares, cubes and hypercubes (depending on the data dimension).
- the IFHT parameter space is linear. If a large cube is supported by T data points, if the cube is subdivided into multiple small cubes and voted on them, it is impossible for any small cube to have more votes than the original cube. votes, and the data points supporting any small cube must be a subset of the data points supporting the large cube. Therefore, the segmentation of the IFHT parameter space can be carried out in a multi-scale and hierarchical manner.
- the IFHT parameter space is first quantified into a relatively large number of counting units, and then the counting units with enough votes are obtained and then refined again until the required accuracy is achieved. This approach gives IFHT a computationally huge advantage over other methods that use polar coordinates to implement the Hough transform.
- [C 1 , C 2 ..., C k ] is the center position of the counting unit, and r is the radius of the circumscribed circle of the counting unit.
- the IFHT parameter space can be regarded as a hollow hypercylindrical surface (as shown in Figure 3).
- the root node of the k-byte tree is a vector C 0 whose half length is S 0 , and each node in the tree can generate 2 k child nodes.
- We can use the vector b [b 1 ,...,b k ], where bi is a binary variable with value -1 or 1, to index any child node.
- the center value of its child nodes whose index values are [b 1 ,...,b k ] can be calculated by the following formula:
- IFHT draws on the method of multi-scale layered refinement and iterative calculation in the FHT algorithm.
- a vote counting unit whose center position of layer l is at [C l1 , C l2 ..., C lk ], its distance D l to the hyperplane defined by any formula 3 in the parameter space is equal to Regularizing the half-side length of this distance, the following formula holds:
- Equation 2 we use The constraints to ensure the uniqueness of the line or plane. If we have no additional information about the target to be detected, the initial range of ⁇ 1 ,..., ⁇ n ) is
- IFHT uses a line, plane or hyperplane in the data space with to represent, where ⁇ 1 ⁇ 0, d is a parameter specified by the user, which is less than or equal to the maximum distance d max of all data points from the origin. If there is no additional requirement on the target parameters, the initial range of the parameter space can be set as
- ⁇ 1, i 1,...,n.
- the integrated fast Hough transform IFHT needs to pre-set the system parameters:
- a voting threshold T is set to determine the minimum data support required to identify the target and the expected detection accuracy q.
- the IFHT calculation process is as follows:
- IFHT IFHT
- ⁇ 1 n+1 variables
- ⁇ 2 ⁇ 2 ,..., ⁇ n+1 ⁇
- ⁇ 1 can be uniquely determined, so it is essentially a dummy variable. Therefore, in the specific implementation, we can also directly calculate ⁇ 1 without introducing it into the parameter space.
- the quantization precision q of the parameter space is an important parameter. If the value is too low, the number of detected targets will be too much and the accuracy will not be enough; when the value of q is too high, the algorithm will be too sensitive to noise, resulting in missed detection of individual targets.
- FHT FHT
- the IFHT algorithm has some very interesting properties.
- the noise is not assumed to be uniformly distributed in all dimensions.
- y mx+c,
- ⁇ 1 and x m'y+c',
- ⁇ 1 are used together to describe all straight lines.
- the least squares method will minimize the total squared value of the deviation along the y-axis, as shown in Figure 7a.
- IFHT fits two dimensions at the same time, essentially converting the least squares fitting into a total least squares fitting.
- the size and boundary of the counting unit will be determined. Algorithm analysis shows that when a straight line corresponds to the boundary value of the counting unit, the accuracy of FHT detection will be greatly reduced (Illingworth and Kittler, 1987). Because of the different models, few lines will suffer from the same problem in both FHT and IFHT.
- Figure 7b shows the performance of FHT and IFHT when the noise standard deviation is 0.02 and 0.067. It can be seen that in this case, the performance of FHT drops sharply, but the performance of IFHT is hardly affected.
- the above invention can be realized by using software, or by dedicated hardware, or by combining software and hardware, and the hardware can even be a general-purpose computer system.
- the present invention can be integrated into a module, and the module and its functions can be realized by software or hardware.
- this module can be a process, a computer program, or a part thereof for implementing a specific function or related functions.
- this module can be a functional hardware unit for cooperating with other components.
- a module may be a digital electronic component, or a part of a digital circuit such as an application specific integrated circuit (ASIC).
- ASIC application specific integrated circuit
- FIG. 10 shows a schematic diagram of a computer system 400 that can be used to implement the above invention.
- the computer 402 is equipped with the memory needed to realize the software of the above invention and the execution of the software.
- the software which executes on the operating system of the computer system, is responsible for implementing and executing the techniques described herein and obtaining program results by computer system 400 .
- Computer software is a series of logical instructions that a processor (such as a computer CPU) can translate.
- Computer software can be composed of any language or expression, including a series of instructions to allow the processor to perform specific functions, and may also need to be translated into other languages, codes or symbols during the process.
- Computer software usually uses a specific computer language to write source programs, and the source programs are converted into machine language codes that can be executed by an operating system or a machine through a language compiler.
- the writing of computer software may involve other program components, class libraries, etc., which are used to provide support for the specified functions of the present invention.
- the computer system 400 includes: a computer 402, an input device such as a keyboard 404, a mouse 406 or an external memory 408 (such as a CD, DVD, USB flash drive, etc.), an output device 410, and a network connection device (a wide area network connection device 412, a local area network connection device) , or other data equipment (such as video equipment, image acquisition equipment, lidar, data acquisition equipment, data preprocessing equipment, etc.).
- Computer 402 comprises: processor 422, ROM 424, RAM 426, network interface 428 is used for being connected with external network connection equipment, and input-output interface 430 is used for being connected with input equipment, output equipment and other data equipment, data storage equipment. All parts of the computer 402 are connected through the system bus 436, so that the information and data exchange of each part can be kept smooth.
- the computer system 400 we describe here is schematic, and other configurations can also fulfill the functions of this patent.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
本发明公开了一种多维空间中的直线、平面和超平面的快速检测方法,该方法建立了新的直线、平面及超平面的数学模型,以此为基础开发了一体化快速霍夫变换,使用单一k字节树负责所有信息存储。这一新方法具有两个重要的优势:首先,模型对应总最小二乘拟合算法,对数据噪声容忍度更好,解决了快速霍夫变换对位于参数空间分割线上目标检测精确度过低的问题;其次,一体化快速霍夫变换中,数据空间中距离很近的目标在参数空间中是聚集在一起的,参数空间计算过程可以采用直观的图形显示,从而快速判断目标的数量,指导系统参数设定并区分重复识别的目标。本发明可应用于图像分析、计算机视觉、自动驾驶、人工智能和数据分类等。
Description
本发明属于图像处理技术领域,主要涉及一体化快速霍夫变换,具体是一种多维空间中的直线、平面和超平面的快速检测方法,其可用于多种数据的数据挖掘、分类和模式识别,包括图像数据、金融数据、阵列数据等;还可应用于图像分析、计算机视觉、自动驾驶、人工智能和数据分类等。
直线检测是物体识别的基础,在图像处理及识别中有着举足轻重的地位。1962年霍夫首次开发了图像中高效的直线检测算法并获得了专利,这一算法被称为霍夫变换。至今已经有超过2500份学术文章提出各种霍夫变换的改进算法。2D霍夫变换也被拓展到多维空间中的平面及超平面检测,应用于众多领域,包括计算机视觉、机器学习、人工智能、自动驾驶及数据分类等。
霍夫变换和其绝大多数的变种的目标是二维数据中的直线检测,通过适当的改进也可用于特定曲线和图形的检测。霍夫变换用于直线检测主要包括两个步骤:设定长度有限的参数空间,设计映射函数将输入数据转换成该参数空间的直线;将参数空间分割成多个小的计票单元,利用输入数据对各计票单元投票。票数超过规定阈值的计算单元对应霍夫变换检测到的目标。算法流程如图1所示。
近年来随着激光雷达,尤其是激光雷达LiDAR在自动驾驶及空间测量上的广泛应用,霍夫变换在3D数据点云中的应用也越来越广泛。当拓展到3D甚至更高维度的数据空间时,霍夫变换对应的最简单的检测目标是平面和超平面(hyperplane)。但是由于“数据维度的诅咒”,在多维空间中相应算法的计算量和复杂度往往会指数级的增加。目前只有为数不多的霍夫变换算法可以实际应用于多维空间数据。其中,由IBM公司的Li等人于1986年提出的快速霍夫变换(Li,et al.,1986)是公认的在计算和存储复杂度上性能表现突出的佼佼者,算法也被一些通用的计算机图像处理类库如GANDALF实现并获得广泛应用。有趣的是,我们发现一些文献常常把快速霍夫变换用来指代Brady等人提出的类似于Radon变换的一个算法(Brady,1998),为了避免误解我们在此声明,本文中提到的快速霍夫变换(FHT)都是专指Li等人于1986年开发的算法,除非特别说明。Li等人提出的快速霍夫变换在计算和存储复杂度上不论是在二维还是多维空间表现都很优异,这主要归功于它采用的由粗到细渐进式处理的策略。在n维度空间中,快速霍夫变换把数据点转换为参数空间中的超平面(在二维空间中是直线,三位空间中则是平面)并搜索他们的交集。
在一个n维的数据空间{F
1,F
2,…,F
n},各个数据点表示为F(j)=[F
1(j),F
2(j),…,F
n(j)].快速霍夫变换的计算流程将如下所示:
1、设定一个k维的参数空间{X
1,X
2,…,X
k},找到对应的映射函数将数据点F(j)映射成参数空间的直线或超平面。参数空间的直线或超平面数学表达式为
2、设定一个投票阈值T用于决定认定目标需要的最小数据支持,以及预期的检测的精度q。
3、利用迭代算法将参数空间按从低到高的解析度分割成超立方体,并使用n个k字节树来表达和存储。进一步的分割和分析只需要对获得足够投票支持的超立方体实行。
在快速霍夫变换中,参数空间需要被划分为n个子空间,这是由使用的数学模型决定的。举个例子,在二维空间中,一条直线通常利用y=mx+c,m→∞来描述。对空间[m,c]直接搜索是不现实的,因为m的值可以是无穷大。快速霍夫变换因此使用两个方程来描述这个空间y=mx+c,|m|≤1 and y=mx+c,|m|≤1。这两个方程实质上把整个搜索空间转换成两个子参数空间(m,c)and(m',c'),其中|m|≤1,|m′|<1来搜索。同样的策略也被广泛应用于其他霍夫变换变种中,比如自适应霍夫变换(Illingworth and Kittler,1987)。
以二维快速霍夫变换为例,数据点(x,y)可以位于直线y=mx+c,|m|≤1或x=m′y+c′,|m′|<1上.参数空间由两个子空间(m,c)和(m',c')构成。将数据点(x,y)映射成直线第一个子空间(m,c)中的直线a
0+a
1m+a
2c=0(|m|≤1)的方程是
and
将数据点(x,y)映射成直线第二个子空间(m',c')中的直线a
0+a
1m′+a
2c′=0(|m′|<1)的方程是
and
快速霍夫变换利用迭代算法将参数空间按从低到高的解析度分割成互相嵌套的超立方体,并使用n个k字节树来表达和存储。每一个超立方体对应于数据空间中一条特定精度的超平面,我们称为计票单元(Accumulator)。对一个参数空间的计票单元m∈[m
1,m
2],c∈[c
1,c
2],如果我们想检测数据点(x,y)是否满足y=mx+c,m∈[m
1,m
2],c∈[c
1,c
2]的条件,我们测试a
0+a
1m+a
2c=0在
m∈[m
1,m
2],c∈[c
1,c
2]条件下是否满足。后者可以重构成一个矩形m∈[m
1,m
2],c∈[c
1,c
2]和直线a
0+a
1m+a
2c=0是否相交的问题,其中a0,a1,a2均已知。如果我们利用数据正则化变化使[m
1,m
2]和[c
1,c
2]具有同样长度,上述测试可以进一步 放宽至测试直线和正方形的外接圆是否相交,从而大大缩减运算量(如图2所示)。以二维空间为例,我们只需检测|a
0+a
1m
*+a
2c
*|<r是否满足即可,其中m
*和c
*是正方形中心的坐标,r是正方形外接圆的半径。
快速霍夫变换在搜索过程中,参数空间被划分为n个子空间,每个子空间由一个独立的k字节树负责其信息存储。各子空间可以分成k个超立方体,对应k字节树的k个节点,每个超立方体可以迭代分解成k个更小的超立方体,形成k字节树的更深一层节点。超立方体只有当它对应的超平面获得足够多的数据点的投票支持时,才会做更深层次的分解,并直到立方体的大小到达指定的精度时,细分停止。这种分层模型极大地节省了计算量和存储空间。
对多个独立的子空间分别进行搜索带来了很多额外的负担。虽然理论上来说,任何一条直线都不可能同时满足y=mx+c,|m|≤1 and x=m′y+c′,|m′|<1,但是实际上,斜率接近1的直线在两个子空间都会被检测到,在数据受噪声影响时尤其如此。实际算法实现上,多个独立的子空间都会在边界上有少量重叠,避免在搜索空间中出现漏洞。随着数据维度的增加,计算量和存储空间的浪费也越来越严重。
更糟糕的是,快速霍夫变换的多个子空间使用完全不同的映射函数。测试在多个不同的子空间中检测到的目标(二维空间直线,三位空间的平面,高维空间的超平面)是不是来自于同一个对象变得极具挑战性。事实上,在最初的快速霍夫变换的文献中,也提到了这一难题,并把它作为未来的科研工作重点。但这么多年以来,这一难题从未被解决。
发明内容
本发明的目的在于提供一种多维空间中的直线、平面和超平面的快速检测方法,以克服当前现有技术的不足。
本发明采取的技术方案如下:
第一方面,本发明提供了一种多维空间中的直线、平面和超平面的快速检测方法,在数据点云模式,在n维空间{F
1,F
2,…,F
n}中,数据点云由N个数据点组成,每个数据点表示为:F(j)=[F
1(j),F
2(j),…,F
n(j)],j=1…N;所述方法将数据空间中的直线、平面或超平面采用
来表征,其中:
β
1≥0,d为用户指定的参数,{β
1,β
2,…,β
n+1}是参数,n对应数据的维度,m值为n-1或者n,τ
i为一个数据正则化因子,由用户自己选择。
进一步的,所述方法中,只检测某些特定的子集,包括:检测符合β
n+1=0的目标,或者任意一个β
i≥0的目标,或者β
i在特定取值区间内的目标。
进一步的,所述方法通过以下步骤预先设定系统参数:
第一步,如果使用m=n模型,设定直线、平面或超平面的截距的取值范围d,d小于等于输入的点云数据中各数据点离坐标原点的最大距离d
max,若使用m=n+1的模型,则可以默认设定d=1;
第二步,设定一个投票阈值T用于决定认定目标需要的最小数据点个数,设定预期的检测的精度q,q的值为整数;
第三步,创建k字节树,设定k字节树根节点中心位置参数及各维度半长;
第四步,转换各数据点到参数空间;
第五步,计算根节点到任意数据点j的距离,
第六步,若根节点票数小于T,则系统中无满足条件的目标,系统退出,如票数大于等于门限值T,则生成根节点的满足b
1=1条件的2
k-1子节点,为每个新生成的节点配备一个矢量b=[b
1,…,b
k];
第七步,利用全部N个数据点对新节点投票;
第八步,检测新节点的票数,若小于门限值T,则停止对该子节点处理;若该节点的层级已到达q层,则输出该节点的参数信息,其对应系统中检测到的目标,并停止对该节点的分析;
第九步,若一个新节点的票数大于等于门限值T,则生成该节点下一层的2
k个子节点,为每个新生成的节点配备一个矢量b=[b
1,…,b
k];对每个新生成的节点测试否满足公式
如果不满足,则放弃该子节点;
第十步,迭代重复第七到九步骤,直到系统中不再有新节点生成,且所有的节点都被处理完毕。
更进一步的,在所述第一步中,如果检测经过坐标原点的直线、平面或者超平面目标,d设定为0。
更进一步的,在所述第三步中,如果需要检测输入数据中的所有目标,根节点中心位置在C
0=(0,…,0),其在各维度半长为S
0=(1,…,1);如果只检测输入数据中的某些特定区域的目标,根节点中心位置则相应调整。
更进一步的,在所述第四步中,如果需要检测输入数据中的所有目标,使用以下方程:
如果只检测输入数据中的某些特定区域的目标,则使用以下公式推导:
更进一步的,在所述第七步中,投票规则是对任意数据点j,计算其到该子节点到的距离:
更进一步的,在所述第十步之后,统计第八步中输出的所有节点,如果输出节点有多个,对可能重复的目标合并,若无节点被输出,则系统没有检测到任何目标。
第二方面,本发明提供了一种计算机设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令,所述处理器执行所述计算机可读指令时实现如上所述的快速检测方法。
第三方面,本发明提供了一个或多个存储有计算机可读指令的可读存储介质,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行如上所述的快速检测方法。
本发明的有益效果是:本发明建立了新的直线、平面及超平面的数学模型,以此为基础开发了一体化快速霍夫变换。一体化快速霍夫变换使得我们可以使用一个单一k字节树负责所有搜索空间的信息,通过去冗余大大消减系统中的计算量和存储需求。这一新方法还具有两个重要的优势。首先,快速霍夫变换的拟合模式对应的是最小二乘法,而一体化快速霍夫变换则对应总最小二乘拟合算法。前者是假设数据的噪声只存在一个维度上,而后者的模型则假设数据噪声是遍布所有维度的,因此对数据噪声容忍度更好,并在实际应用中很好地解决了快速霍夫变换对位于参数空间分割线上目标的检测精确度过低的问题。其次,一体化快速霍夫变换中,在数据空间中彼此距离很近的目标在参数空间中是聚集在一起的,这使我们可以采用直观的图形显示参数空间中的数据分析过程,从而快速判断目标的数量,形象化的指导我们设定系统参数并区分被重复识别的目标。
图1是霍夫变换用于直线或者平面检测的标准流程图。
图2是测试一条直线和一个正方形相交示意图,可以直接测试直线和一个正方形是否相交,也可测试直线和正方形的外接圆是否相交,后者在计算上会节省大量时间。
图3是多尺度分层细化一体化霍夫变换的参数空间示意图;计票单元在各维度上都具 有同样的长度,IFHT只需要分析和n-1维度上的单位圆相交的计票单元。
图4是利用k字节树表示整个参数空间示意图;我们使用0和1构成的矢量b来代表不同的节点,除根节点外每个结点产生2
k个子节点,根节点包含2
k-1个子节点。
图5是IFHT参数空间的分割及分析的直观展示图;其中(a)为2D输入数据,其中有包含两条直线;(b)为IFHT在q=5层级所有计票单元的投票分布,票数超过门限值的计票单元聚集成两个独立的群,对应两条直线。
图6是一个3D数据集和对应的IFHT参数空间示意图;其中(a)为3D数据集和其中的平面,(b)为IFHT在q=5所有计票单元的分析结果,超过门限值的计票单元聚集成一个独立的群,对应数据集中的平面。
图7是FHT和IFHT直线检测时使用不同的目标函数示意图;其中(a)为FHT对应最小二乘法拟合,最小化沿y轴的偏差的总平方值,(b)为IFHT对应总最小二乘法拟合,最小化所有欧式距离平方值的总和。
图8是FHT和IFHT直线检测性能比较示意图,两个级别的噪声被分别添加到数据的第二个维度上;其中(a)为一般性直线检测,(b)为直线对应IFHT和FHT的参数平面上的计票单元的边界值时,FHT性能急剧下降,而IFHT性能基本不变。
图9是当噪声均匀分布在数据的各个维度上时,IFHT和FHT的算法性能比较示意图;模拟噪声的标准差分别为0.02和0.067。
图10是一个可以用于承载本发明的计算机系统的示意图。
下面结合附图和具体实施例对本发明进行详细说明。
1.一体化快速霍夫变换数学模型
一体化快速霍夫变换IFHT采用了一种更直观的数学模型来表征直线及多维空间中的平面和超平面。假设输入数据是n维空间{F
1,F
2,…,F
n}中的数据点F(j)=[F
1(j),F
2(j),…,F
n(j)],一体化快速霍夫变换利用以下数学模型来表征超平面:
上述公式中τ
i是一个数据正则化因子,具体选择我们后述讨论。简单起见,我们对上述公式可以简化一下暂不考虑β
1≥0的条件。假设我们知道了{β
1,β
2,…,β
n+1}的取值范围,并假设它们取值范围半长为{L
1,L
2,…,L
n+1},上述公式可以转化为:
其中:F(j)=[F
1(j),F
2(j),…,F
n(j)]到参数空间{X
1,X
2,…,X
k}的映射函数为
W(j)的值由
确定。在参数空间{X
1,X
2,…,X
k}中,各个维度X
i范围是[-1,1],整个参数空间因此可以细分成多个小超立方体(计票单元)用于投票。条件β
1≥0也可以简单的通过忽略不满足要求的计票单元来实现。
以二维空间为例,任何穿过点(x,y)的直线可以被表示为β′
1x+β′
2y+β′
3=0,β′
1
2+β′
2
2=1,β′
1≥0,|β′
3|≤d,其中d是一个指定参数,其小于等于所有数据点离原点的最大距离d
max。假设β′
3=d·β
3,β′
1=β
1,β′
2=β
2并暂不考虑β′
1≥0的限制,我们可以把以上公式转换成β
1x+β
2y+d·β
3=0,β
1
2+β
2
2=1,|β
i|≤1,i=1,2,3。(β
1,β
2,β
3)构成我们的参数空间,把数据点(x,y)转换到参数空间a
1β
1+a
2β
2+a
3β
3=0,β
1
2+β
2
2=1,|β
3|≤1的映射函数是
and
因为|β
i|≤1 for i=1,2,3,所以参数空间(β
1,β
2,β
3)在各个维度的长度都是2,可以细分成等长的立方体作为计票单元。
2.多尺度分层细化的计算模式
数据空间中的每个数据点对应IFHT参数空间的一条直线或平面。在现实中,数据中的测量误差及噪声对应会造成参数空间内相应对象的偏移。在IFHT参数空间中,通过对一个微小区间的检测而不是单单检测某个特定的值,可以实现对噪声容忍并提高检测精度。而所检测的小区间的长度则对应了我们最后的检测目标的精度。对一个数据空间中的数据点云,检测具有T个以上数据点支持的直线、平面和超平面等目标时,IFHT会首先把参数空间分割成众多小计票单元,让每个数据点对计票单元进行投票。由于IFHT在参数空间各维度的长度都相等,分割成的小计票单元在各维度都相等,在参数空间中计票单元由正方形、立方体及超立方体(取决于数据维度)构成。
IFHT参数空间是线性的,如果一个大立方体获得了T个数据点支持,如果对这个立方体再次细分获得的多个小立方体,并对它们投票,任何小立方体的票数都不可能超过原立方体的票数,而且支持任一小立方体的数据点一定是支持大立方体的数据点的子集。因此对IFHT参数空间的分割可以实行多尺度分层细化的方式。IFHT参数空间先量化成比较大的数个计票单元,获得足够票数的计票单元然后再次细化,直至达到要求的精度为止。 这一方式使得IFHT相对于利用极坐标来实现霍夫变换的其他方法具有计算上的巨大优势。
由于IFHT参数空间内的计票单元在各维度是等长的,我们可以利用如图2所示的近似算法来加速。判断一个数据点是否对一个计票单元投票的公式可写成
[C
1,C
2…,C
k]是计票单元的中心位置,r是计票单元外接圆的半径。
3.利用k字节树实现存储及计算
IFHT参数空间可以看作一个中空的超圆柱面(如图3所示)。我们使用多尺度的互相嵌套的超立方体来遍历整个参数空间,对应一个k字节树数据结构来存储信息并计算。k字节树的根节点是向量C
0其边的半长为S
0,树中每个节点可产生2
k个子节点。我们可以使用矢量b=[b
1,…,b
k],其中b
i是值为-1或1的二值变量,来索引任何一个子节点。一个l层,中心为C
l的节点,它的个索引值为[b
1,…,b
k]的子节点的中心值可以由下面公式来计算:
其中S
l+1是所有l+1层节点的半长,且S
l+1=S
l/2。
IFHT借鉴了FHT算法中的多尺度分层细化并迭代计算的方法。一个l层中心位置在[C
l1,C
l2…,C
lk]的计票单元,其到参数空间中任意公式3定义的超平面的距离D
l等于
对这一距离作半边长正则化则下述公式成立:
我们可以利用迭代方法计算任意计票单元的正则化距离。假设参数空间{X
1,X
2,…,X
k}的初使值的中心点在C
0=[C
01,C
02,…,C
0k],我们有:
公式4中的任意数据点是否对一个计票单元投票的公式则可以简化成:
在参数空间,这等同于测试一个超立方体是否与(n-1)维度的单位圆/球是否相交。利用公式4相同的简化办法,对一个在l层,中心位置于[C
1,C
2,…,C
k]的计票单元,如果立方体 的外接圆半径为r,测试公式可以简化为
4.一体化快速霍夫变换算法流程
n维空间{F
1,F
2,…,F
n}中,给定系列数据点云F(j)=[F
1(j),F
2(j),…,F
n(j)]。IFHT将数据空间中的直线、平面或超平面用
来表征,其中
β
1≥0,d是用户指定的参数,其小于等于所有数据点离原点的最大距离d
max。如果无对目标参数的额外要求,参数空间的初始范围可设定为|β
i|≤1,i=1,…,n。一体化快速霍夫变换IFHT需要预先设定一下系统参数:
设定一个k维的参数空间{X
1,X
2,…,X
k},找到对应的映射函数将数据点F(j)映射成参数空间的直线或超平面。参数空间的直线或超平面数学表达式为
其中a
i(j)是F(j)的函数,
当j=1,…,n,
其中W(j)是一个正则化标量用于满足
设定一个投票阈值T用于决定认定目标需要的最小数据支持,以及预期的检测的精度q。
IFHT计算过程如下:
设定k字节树根节点。根节点位置在(0,…,0),其在各维度半长为(1,…,1),计算根节点到所有数据点的距离并计算票数,如票数大于门限值T,则生成根节点的满足b
1=1条件的2
k-1子节点。
判断各子节点是否满足公式9,如果不满足,则放弃该子节点。对满足的子节点利用公式6计算各数据点正则化距离,并通过测试公式7是否满足完成对子节点的投票,如票数小于门限值T,则放弃该子节点,相反则对该子节点再次细化生成2
k个子节点。
对上述步骤中生成的子节点迭代处理。最后记录下所有达到指定的精度q且获得票数大于门限值T的节点,如果没有节点能达到指定的精度q且获得票数大于门限值T,可以汇报精度最高的节点(如有多个则记录票数最高的)。
5.一体化快速霍夫变换其他可能的变种
上述一体化霍夫变换,直线、平面及超平面使用以下数学模型来表征
在某些情况下,我们可以对上述公式作一些修改。其中一个可能的改变是将β
n+1也放入正则化等式中,这样就可以免去在一开始计算所有数据点距离原点最大距离的步骤。对应的方程将会转化成
另外,IFHT的具体实现作一定调整。在IFHT数学模型中,共有n+1个变量(β
1,…,β
n+1)。但是当{β
2,…,β
n+1}被确定时,β
1就能够被唯一的确定出来,所以它本质上是一个虚拟变量。因此,在具体实现上,我们也可以直接计算β
1,而不需要将它引入到参数空间中去。
6.一体化快速霍夫变换结果
在几乎所有的霍夫变换的变种中,参数空间的量化精度q是一个重要参数。如果值取得过低,检测到的目标数量会过多且精度不够;当q值选择的过高,算法会对噪声过于敏感,造成个别目标的漏检。在FHT实际应用中,我们经常能看同一个目标在精度q时,有多个计票单元同时报告发现检测目标且它们对应的是同一个目标,但是在精度q+1级别时,没有任何目标能被检测到。因此自动的识别同一个目标在霍夫或者快速霍夫变换中,依赖大量的后处理。
在IFHT中,参数空间的一体化使得直观地展示参数空间内的分割及计算进展变得可能,从而使我们可以非常直观地确定系统参数,并对结果出现中的可能的重复目标作出直观判断。图5和图6中,我们分别展示了一个2D数据集和一个3D数据集在IFHT运行中的参数空间的数据分析结果。我们可以看到,当量化精度q的值取得比较低时,有多个计票单元会发现同一个目标。但是,检测到相同目标的计票单元在参数空间总是聚集在一起的,因此用户可以非常直观地确定自己的数据中到底包含多少待检测目标,同时针对性的确定系统参数及合并重复检测到的目标。
IFHT算法具有一些非常有趣的特性。在快速霍夫变换FHT的模型中,噪声并不是被假设均匀地分布在所有维度上。以2D数据为例,y=mx+c,|m|≤1和x=m′y+c′,|m′|<1一起被用于描述所有的直线。当我们利用第一个公式来拟合输入的数据点云时,最小二乘法将最小化沿y轴的偏差的总平方值,如图7a所示。但是IFHT对个维度同时拟合,本质上将最小二乘拟合转换成总最小二乘法拟合。在2D空间中,我们寻找满足β
1x+β
2y+β
3=0和β
1
2+β
2
2=1的(β
1,β
2)值是通过最小化所有欧式距离平方值的总和,拟合过程如图7b所示。
我们利用模特卡洛的办法在合成数据上比较了IFHT和FHT的性能。在第一个实验中,我们产生二维空间直线y=0.8x+0.24+g上的N个数据点,其中x满足(-2,2)的均匀分布,g是高斯白噪声,其标准差为0.02和0.067,分别对应信噪比0.03%和0.1%。x,y~N(0,1)的N个数据点被添加到数据集中作为背景噪声。同样的参数我们产生100次,用IFHT和FHT以(T=N/2 and q=5)分别检测。图7a显示了在两种设置中,IFHT都比 FHT具有更高的检测率。
当参数初始值设定后,计票单元的尺寸和边界将会确定下来。算法分析表明,当一条直线对应于计票单元的边界值时,FHT检测的精度会大幅度降低(Illingworth and Kittler,1987)。由于模型不同,很少有直线在会在FHT和IFHT中同时遇到同样的问题。直线y=0.5在参数(q=5and d=2)是少数满足这一条件的,图7b展示了当噪声标准差为0.02和0.067时,FHT和IFHT的性能。可以看到,在这种情形下,FHT的性能急剧下降,但IFHT的性能几乎不受影响。
在上述模拟实验中,噪声只添加到数据的第二维度,在实际数据中,这种情形很少会发生。我们重复了第一个实验,但是在两个维度都添加了噪声。我们使用的模型是y=0.8t+0.24+g2,x=t+g1,t~U(-2,2)。这里g1和g2都是相同标准差的高斯噪声,IFHT和FHT参数(q=5 and d=2)用于目标检测,结果如图8所示。显然,在不同的噪声水平,IFHT都比FHT有更好的性能。
上述发明可以使用软件实现,或者通过专用硬件来实现,也可以通过软件和硬件结合的方式来实现,硬件甚至可以是通用的计算机系统。本发明可以集成为一个模块,模块及其功能可以通过软件或硬件来实现。在软件实现中,这个模块可以是一个进程,一个计算机程序,或者它们的一部分用于实现一个特定的功能或相关功能。在硬件实现中,这个模块可以是一个功能性硬件单元,用于和其他元件协同工作。比如模块可以是一个数字电子元件,或者是数字电路如特殊应用数字集成电路(ASIC)的一部分。
图10展示了可用于实现上述发明的计算机系统400的示意图。其中计算机402装配有实现上述发明软件及软件执行时需要的内存。软件在计算机系统的操作系统上执行,负责实现并执行本发明描述的技术,并由计算机系统400获取程序结果。
计算机软件是一系列处理器(如计算机CPU)能翻译的逻辑指令。计算机软件可以由任何语言或者表达式构成,包含一系列指令让处理器完成特定的功能,这一过程中也可能需要翻译成其他语言、代码或标示。
计算机软件通常利用特定计算机语言编写源程序,经过语言编译程序将源程序转换成操作系统或者机器可以执行的机器语言代码。计算机软件的编写可能涉及到别的程序组件、类库等,用于为本发明完成指定的功能提供支持。
计算机系统400包括:计算机402,输入设备如键盘404,鼠标406或者外置内存408(如CD、DVD、USB闪存盘等),输出设备410,网络连接设备(广域网连接设备412,局域网连接设备),或者其他数据设备(如视频设备、图像采集设备、激光雷达、数据采集设备、数据预处理设备等)。计算机402包括:处理器422,ROM 424,RAM 426,网络 接口428用于和外部网络连接设备相连,输入输出接口430用于和输入设备、输出设备及其他数据设备、数据存储设备相连。计算机402的各部分都通过系统总线436相连,使各部分的信息和数据交换维持畅通。这里我们描述的计算机系统400是示意性的,别的配置也同样可以完成的本专利的功能。
以上显示和描述了本发明的基本原理、主要特征和优点。本领域的普通技术人员应该了解,上述实施例不以任何形式限制本发明的保护范围,凡采用等同替换等方式所获得的技术方案,均落于本发明的保护范围内。本发明未涉及部分均与现有技术相同或可采用现有技术加以实现。
Claims (10)
- 根据权利要求1所述的一种多维空间中的直线、平面和超平面的快速检测方法,其特征在于,所述方法中,只检测某些特定的子集,包括:检测符合β n+1=0的目标,或者任意一个β i≥0的目标,或者β i在特定取值区间内的目标。
- 根据权利要求1或2所述的一种多维空间中的直线、平面和超平面的快速检测方法,其特征在于,所述方法通过以下步骤预先设定系统参数:第一步,如果使用m=n模型,设定直线、平面或超平面的截距的取值范围d,d小于等于输入的点云数据中各数据点离坐标原点的最大距离d max,若使用m=n+1的模型,则可以默认设定d=1;第二步,设定一个投票阈值T用于决定认定目标需要的最小数据点个数,设定预期的检测的精度q,q的值为整数;第三步,创建k字节树,设定k字节树根节点中心位置参数及各维度半长;第四步,转换各数据点到参数空间;第五步,计算根节点到任意数据点j的距离,第六步,若根节点票数小于T,则系统中无满足条件的目标,系统退出,如票数大于等于门限值T,则生成根节点的满足b 1=1条件的2 k-1子节点,为每个新生成的节点配备一个矢量b=[b 1,…,b k];第七步,利用全部N个数据点对新节点投票;第八步,检测新节点的票数,若小于门限值T,则停止对该子节点处理;若该节点的层级已到达q层,则输出该节点的参数信息,其对应系统中检测到的目标,并停止对该节点的分析;第九步,若一个新节点的票数大于等于门限值T,则生成该节点下一层的2 k个子节点,为每个新生成的节点配备一个矢量b=[b 1,…,b k];对每个新生成的节点测试否满足公式 如果不满足,则放弃该子节点;第十步,迭代重复第七到九步骤,直到系统中不再有新节点生成,且所有的节点都被处理完毕。
- 根据权利要求3所述的一种多维空间中的直线、平面和超平面的快速检测方法,其特征在于,在所述第一步中,如果检测经过坐标原点的直线、平面或者超平面目标,d设定为0。
- 根据权利要求3所述的一种多维空间中的直线、平面和超平面的快速检测方法,其特征在于,在所述第三步中,如果需要检测输入数据中的所有目标,根节点中心位置在C 0=(0,…,0),其在各维度半长为S 0=(1,…,1);如果只检测输入数据中的某些特定区域的目标,根节点中心位置则相应调整。
- 根据权利要求3所述的一种多维空间中的直线、平面和超平面的快速检测方法,其特征在于,在所述第十步之后,统计第八步中输出的所有节点,如果输出节点有多个,对可能重复的目标合并,若无节点被输出,则系统没有检测到任何目标。
- 一种计算机设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令,其特征在于,所述处理器执行所述计算机可读指令时实现如权利要求1至8中任一项所述的快速检测方法。
- 一个或多个存储有计算机可读指令的可读存储介质,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行如权利要求1至8中任一项所述的 快速检测方法。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210033412.X | 2022-01-12 | ||
CN202210033412.XA CN114387442A (zh) | 2022-01-12 | 2022-01-12 | 一种多维空间中的直线、平面和超平面的快速检测方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023134000A1 true WO2023134000A1 (zh) | 2023-07-20 |
Family
ID=81201144
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2022/080650 WO2023134000A1 (zh) | 2022-01-12 | 2022-03-14 | 一种多维空间中的直线、平面和超平面的快速检测方法 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN114387442A (zh) |
WO (1) | WO2023134000A1 (zh) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130222696A1 (en) * | 2012-02-28 | 2013-08-29 | Sony Corporation | Selecting between clustering techniques for displaying images |
CN106104573A (zh) * | 2014-02-04 | 2016-11-09 | 弗劳恩霍夫应用研究促进协会 | 霍夫处理器 |
CN109557532A (zh) * | 2018-10-18 | 2019-04-02 | 西安电子科技大学 | 基于三维霍夫变换的检测前跟踪方法、雷达目标检测系统 |
US20190251330A1 (en) * | 2016-06-13 | 2019-08-15 | Nanolive Sa | Method of characterizing and imaging microscopic objects |
-
2022
- 2022-01-12 CN CN202210033412.XA patent/CN114387442A/zh active Pending
- 2022-03-14 WO PCT/CN2022/080650 patent/WO2023134000A1/zh unknown
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130222696A1 (en) * | 2012-02-28 | 2013-08-29 | Sony Corporation | Selecting between clustering techniques for displaying images |
CN106104573A (zh) * | 2014-02-04 | 2016-11-09 | 弗劳恩霍夫应用研究促进协会 | 霍夫处理器 |
US20190251330A1 (en) * | 2016-06-13 | 2019-08-15 | Nanolive Sa | Method of characterizing and imaging microscopic objects |
CN109557532A (zh) * | 2018-10-18 | 2019-04-02 | 西安电子科技大学 | 基于三维霍夫变换的检测前跟踪方法、雷达目标检测系统 |
Also Published As
Publication number | Publication date |
---|---|
CN114387442A (zh) | 2022-04-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Liu et al. | Artificial intelligence in the 21st century | |
Chen et al. | Linear spectral clustering superpixel | |
Du et al. | Study on density peaks clustering based on k-nearest neighbors and principal component analysis | |
Hernández-Orallo | ROC curves for regression | |
Gao et al. | Manifold Siamese network: A novel visual tracking convnet for autonomous vehicles | |
CN111242208A (zh) | 一种点云分类方法、分割方法及相关设备 | |
CN104392253B (zh) | 一种草图数据集的交互式类别标注方法 | |
Ji et al. | A piecewise linear representation method based on importance data points for time series data | |
Wang et al. | Dynamic position-aware network for fine-grained image recognition | |
Kumar et al. | Survey on techniques for plant leaf classification | |
Zhang et al. | Sparse learning-based correlation filter for robust tracking | |
Liu et al. | Target recognition of sport athletes based on deep learning and convolutional neural network | |
WO2015001416A1 (en) | Multi-dimensional data clustering | |
He et al. | A Method of Identifying Thunderstorm Clouds in Satellite Cloud Image Based on Clustering. | |
Deng et al. | Selective clustering for representative paintings selection | |
He et al. | Practical tracking method based on best buddies similarity | |
Zhou et al. | PEA-YOLO: a lightweight network for static gesture recognition combining multiscale and attention mechanisms | |
Sun et al. | Learning isometry-invariant representations for point cloud analysis | |
Murtagh et al. | Random projection towards the Baire metric for high dimensional clustering | |
WO2023134000A1 (zh) | 一种多维空间中的直线、平面和超平面的快速检测方法 | |
Liu et al. | Deep learning-based image target detection and recognition of fractal feature fusion for BIOmetric authentication and monitoring | |
WO2023050461A1 (zh) | 一种数据的聚类方法、系统及存储介质 | |
Kharinov | Image segmentation method by merging and correction of sets of pixels | |
Martı́nez Carrillo et al. | A compact and recursive Riemannian motion descriptor for untrimmed activity recognition | |
Yuan et al. | Anisotropic neighborhood searching for point cloud with sharp feature |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22919643 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |