CN107038726B

CN107038726B - Gray code coded 3D feature descriptor simplifying method

Info

Publication number: CN107038726B
Application number: CN201710224529.5A
Authority: CN
Inventors: 闵华松; 林云汉; 杜梁杰; 周昊天
Original assignee: Wuhan University of Science and Engineering WUSE
Current assignee: Wuhan University of Science and Engineering WUSE
Priority date: 2017-04-07
Filing date: 2017-04-07
Publication date: 2020-07-31
Anticipated expiration: 2037-04-07
Also published as: CN107038726A

Abstract

The invention provides a 3D feature descriptor simplifying method for Gray code coding, which has the following characteristics that: step A, analyzing a descriptor, and determining a simplified unit and a simplified coding bit number; step B, solving the value range of each simplified unit and dividing the value range into 2^NDividing equally; step C, traversing each value in the descriptor vector, and coding the descriptor by adopting a corresponding Gray code according to a value domain interval where the value falls; and step D, combining the values of each simplified unit to form a complete Gray code descriptor.

Description

Gray code coded 3D feature descriptor simplifying method

Technical Field

The invention relates to the field of robots and computer vision, in particular to a Gray code coded 3D feature descriptor simplifying method.

Background

With the advent of Microsoft Kinect, Asus x position live, intel RealSense cameras and Structure sensors, the acquisition cost of RGB-D data becomes inexpensive. The application of these cheap depth sensors in the fields of robots and computer vision, and the ability of robots to perceive three-dimensional environments is a research hotspot and difficulty in recent years. The purpose of robot three-dimensional environment perception is to find matched object clusters in a model library and a real scene so as to realize the identification of three-dimensional objects in the real scene. In the application of mobile robots and handheld 3D data acquisition equipment, in order to realize the convenience and the real-time performance of the system, the system needs to have lower memory occupancy rate and lower computation complexity, and efficient matching is realized between source points and target point clouds in two arbitrary directions. In general, three-dimensional object recognition can be divided into object recognition based on a global descriptor and object recognition based on a local descriptor. Because the object identification based on the local descriptor can realize the robustness to the problems of occlusion, disorder, view angle change and the like, the method is widely researched.

The expression of descriptors is currently of two types: floating point type descriptions and binary descriptions. Although the floating point type descriptor has good performance in recognition accuracy, the performance in time complexity is not good due to the disadvantages of relatively slow matching speed, large memory consumption and the like. In view of this, methods of expressing the feature descriptors using binary have appeared in recent years. There are two generally feasible methods of binary descriptor acquisition: training a binary descriptor through machine learning and directly binarizing a floating point descriptor. The binary descriptor can be trained through machine learning to obtain a good recognition effect, but a large amount of learning and calculation are needed, and the algorithm is complex.

Prakhya et al in 2015 applied the direct binarization idea to simplification of a 3D point cloud feature descriptor SHOT descriptor for the first time, and proposed B-SHOT. In 2016, the author conducts a plurality of extension experiments on the original algorithm, applies the binarization idea to the other two latest descriptors, namely RoPS and FPFH, establishes B-RoPS and B-FPFH descriptors, and compares B-SHOT, B-RoPS and B-FPFH through a series of experiments, and the experimental results show that the comprehensive performance of the B-SHOT is the best.

However, B-SHOT has two problems:

the first problem is that: B-SHOT takes each 4-dimension of the SHOT descriptor as a simplified unit, which is unreasonable. Because the SHOT descriptor is composed of 32 spatial grids, each spatial grid is divided into 11 equal parts according to the included angle of the normal vector between two point clouds, and the equal parts form a descriptor of a 352(32x11) dimensional vector in total. However, 11 is not evenly divisible by 4, and there are cases where some packets cross the spatial grid, resulting in discontinuity of information, i.e., there is a problem that the simplified cells are not reasonable.

The second problem is that: when the SHOT feature descriptor is converted into the B-SHOT feature descriptor, there may be a serious information loss. Such as: assume that there is such a set of groups S₀,S₁,S₂,S₃Where {0.8,0.24,0,0}, as can be seen from the rule of B-SHOT, the data S of the group is obtained₀And S₁Is over S₀，S₁，S₂And S₃90% of the sum, andto connect S₀And S₁The code is 1 and the remaining codes are 0. Carrying out binarization on the obtained binary image to obtain: { B₀,B₁,B₂,

B

₃1,1,0, 0. As can be seen, B after binarization₀And B₁The values of (a) are all 1, and in fact, the original values are very different, i.e. there may be a problem of serious information loss.

Disclosure of Invention

The present invention is made to solve the above problems, and an object of the present invention is to provide a simple and rational 3D feature descriptor simplifying method for gray code encoding.

The invention provides a 3D feature descriptor simplifying method for Gray code coding, which has the following characteristics that:

step A, analyzing a descriptor, and determining a simplified unit and a simplified coding bit number;

step B, solving the value range of each simplified unit and dividing the value range into 2^NDividing equally;

step C, traversing each value in the descriptor vector, and coding the descriptor by adopting a corresponding Gray code according to a value domain interval where the value falls; and

and D, combining the values of each simplified unit to form a complete Gray code descriptor.

The invention provides a method for simplifying a 3D feature descriptor of Gray code coding, which also has the following characteristics: the descriptor vector is 352-dimensional, the descriptor vector is composed of 32 grids of space and 11 equal included angles in each grid, and the descriptor is simplified by taking the space grid as a unit and serving as a simplifying unit.

The invention provides a method for simplifying a 3D feature descriptor of Gray code coding, which also has the following characteristics: the descriptor vector is 352 dimensions, and the descriptor vector simplifies the descriptor by taking the 352 dimensions as a simplification unit.

The invention provides a method for simplifying a 3D feature descriptor of Gray code coding, which also has the following characteristics: in which a four-valued simplification of the descriptor vector is performed with 2-bit encoding and an octalization is performed with 3-bit binary code.

The invention provides a method for simplifying a 3D feature descriptor of Gray code coding, which also has the following characteristics: in step B, the binary Gray code is used to divide the value range of each simplified unit into four equal intervals, and each interval is assigned with 00,01,11 and 10 according to the size.

Action and Effect of the invention

According to the 3D feature descriptor simplifying method for Gray code coding, the reasonable analysis simplifying unit is adopted to adjust the number of coding bits according to the actual effect (for example, four-value simplification of a vector is carried out by 2-bit coding, and 8-valued of the vector is carried out by 3-bit coding) so as to simplify the descriptor. The problems that a simplified unit is unreasonable and the existing coding mode possibly has serious information loss in the prior art are solved.

Drawings

Fig. 1 is a flowchart of a 3D profile descriptor reduction method for gray code encoding according to an embodiment of the present invention.

Detailed Description

The 3D characterizer simplification method of gray code coding according to the present invention is described in detail below with reference to the accompanying drawings and embodiments.

Examples

And simplifying the feature descriptors by adopting a Gray code coding method, and providing a general method for simplifying the 3D feature descriptors by adopting Gray codes. The simplified method can be defined as equation (1).

B_di＝G(R_di,μ,N) (1)

Wherein R is_diRepresenting the actual value of the vector in the original descriptor, mu being the simplified unit, N being the number of encoding bits, B_diThe binary values representing the simplified descriptors obtained by the method of gray coding herein. According to the above definitions, when applying this equation to the simplification of the SHOT descriptor, and taking μ ═ 1 and N ═ 4, the obtained simplified descriptor is B-SHOT. The simplification method proposed herein will be an extension of the binary method proposed by Prakhya et al, which is a general binary simplification method. For different original descriptors, different simplified descriptors can be generated by changing the two variable parameters mu and N by adopting the method provided by the invention.

As shown in fig. 1, a 3D feature descriptor reduction method for gray code encoding has the following steps:

step A: and analyzing the descriptor to determine the simplified unit and the simplified coding bit number, and entering the step B.

Analysis descriptor D { D₀,D₁,D₂,…,D_m-1Determining a simplified unit mu and a simplified coding bit number N; for example, after analyzing the SHOT descriptor, it is determined that the 352D whole is to be adopted as a simplification unit, and the descriptor is simplified by performing quaternization with two-bit Gray codes.

And the descriptor is simplified by properly increasing the number of the simplified coding bits, and the vector in the simplified unit is subdivided to reduce the information loss degree. Such as: four-valued simplification of the vector is done with 2-bit encoding and 8-valued with 3-bit binary code. I.e. a floating point vector of one dimension is encoded with 2 or more bits.

And B: the value Range of each simplified unit is determined and divided into 2^NAnd (5) dividing equally, and entering the step C. The binary Gray code is used to divide the value range of each simplified unit into four equal intervals, and each interval is assigned with 00,01,11 and 10 according to the size.

For binary coding with more than or equal to 2 digits, if a common binary coding rule is adopted, jump of more than one digit exists between partial codes, and the problem of discontinuous head and tail exists. For example, two-bit binary codes are used to represent four values, which are arranged in the order from small to large as 00,01,10, and 11, respectively. Wherein, there are two jumps when the second code 01 is transformed into the third code 10; there are two transitions in the transition from the fourth code to the first code. Whereas for gray coding there is only one bit transition between each bit (including the head and tail). Such as: 2-bit gray codes are sequentially ordered from small to large as 00,01,11 and 10; gray codes with 3 bits are sorted from small to large as follows: 000,001,011,010,110,111,101,100.

And C: and traversing each value in the descriptor vector, and coding the descriptor by adopting a corresponding Gray code according to a value domain interval to which the value falls, and entering the step D. D_j<range/2^NThe value of the dimension is encoded as 00.

The descriptor vector is 352-dimensional, the descriptor vector is composed of 32 grids of space and 11 equal included angles in each grid, and the descriptor is simplified as a simplifying unit by taking the space grid as a unit. Or the descriptor vector simplifies the descriptor by taking 352 dimensions of the whole as a simplified unit.

The geometric information (statistical histogram information, statistical signature histogram information) of the descriptor is simplified by a binary gray code encoding method. Since the simplification made here is to convert the actual information of the original descriptor into binary numbers (codes containing only 0 and 1), and in addition, considering that only one bit jump exists between every two adjacent codes from small to large of gray codes, the value range of the simplified unit vector is divided equally and the coding is simplified by the gray codes.

Effects and effects of the embodiments

According to the 3D feature descriptor simplifying method for Gray code coding, the reasonable analysis simplifying unit adjusts the number of coding bits according to the actual effect (such as four-value simplification of vectors by 2-bit coding and 8-valued simplification of vectors by 3-bit coding) to simplify the descriptor. The problems that a simplified unit is unreasonable and the existing coding mode possibly has serious information loss in the prior art are solved.

The above embodiments are preferred examples of the present invention, and are not intended to limit the scope of the present invention.

Claims

1. A Gray code coded 3D feature descriptor simplifying method is characterized by comprising the following steps:

step A, analyzing a descriptor, and determining a simplified unit mu and a simplified coding bit number N; encoding a floating point vector of one dimension with 2 or more bits;

performing four-value simplification of the descriptor vector with a 2-bit binary gray code when the encoding bit number N is 2; namely: dividing the value range of each simplified unit into four equal intervals by using a binary Gray code, and respectively assigning values of each interval from small to large as: 00,01,11, 10;

when the encoding bit number N is 3, performing eight-valued simplification of the descriptor vector with a 3-bit binary gray code; namely: dividing the value range of each simplified unit into eight equal intervals by using Gray codes with 3 bits, and respectively assigning values of each interval from small to large as: 000,001,011,010,110,111,101,100, respectively;

step C, traversing each value in the descriptor vector, and coding the descriptor by adopting a corresponding Gray code according to the value domain interval where the value falls; only one bit of jump exists between every two adjacent codes from small to large of the Gray code;

step D, combining the values of each simplified unit to form a complete Gray code descriptor;

the binary value B of the simplified descriptor obtained by the simplification according to the method from the step A to the step D_diIs composed of

B_di＝G(R_di,μ,N)

Wherein R is_diRepresenting the actual value of the vector in the original descriptor, μ is the simplified unit and N is the number of encoding bits.

2. A method for 3D feature descriptor reduction for gray code coding according to claim 1, characterized in that:

wherein the descriptor vector has 352 dimensions,

the descriptor vector is composed of 32 grids of space and 11 equal included angles in each grid, and the descriptor is simplified by taking the space grids as a unit as a simplifying unit.

3. A method for 3D feature descriptor reduction for gray code coding according to claim 1, characterized in that:

wherein the descriptor vector has 352 dimensions,

the descriptor vector simplifies the descriptor in 352 dimensions as a whole as a simplification unit.