Optimization of the 3D Point Cloud Registration Algorithm Based on FPFH Features
, Enzhong Zhang 1,*, Deqiang Mu 1, Shijun Ji 2, Ziqiang Zhang 1, Hongwei Liu 1 and Zheng Fu 1
Round 1
Reviewer 1 Report
The author proposed a Optimization of 3D Point Cloud Registration Algorithm Based 2 on FPFH Features: before proceeding furthers, i need to clarify few comments:
1. why author used FPFH features as taken.
2. Kindly check table 1, i dont feel its nice way to represent.
3. The author has to compare more existing results with existing architectures.
4. The literature should be in resent paper and compare the same.
5. kindly add resent advancement references.
6. Kindly improve the explanation of your results.
Author Response
Point 1: Why author used FPFH features as taken.
Response 1: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Fast Point Feature Histograms (FPFH) are invariant to the attitude of point cloud data, and are still robust when noisy data points exist. In addition, it is also common to use normal vectors and included angles between points when extracting feature points from point cloud data, and the FPFH feature is a feature descriptor based on the included angles of normal vectors of point clouds, and the overall complexity is low, so the FPFH feature is used in this paper.
Point 2: Kindly check table 1, I don’t feel its nice way to represent.
Response 2: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Table 1 Number of point cloud filters
|
Filtering method |
Total number of point clouds |
Number of filtered point clouds |
Effective filtering number |
Filter rate |
|||
|
Single statistical filtering |
454751 |
41858 |
3614 |
90.35% |
|||
|
Optimization algorithm in this paper |
41660 |
3812 |
95.3% |
||||
|
Single statistical filtering |
442562 |
40346 |
3910 |
97.7% |
|||
|
Optimization algorithm in this paper |
40314 |
3942 |
98.6% |
||||
|
Single statistical filtering |
253163 |
21612 |
3704 |
92.6% |
|||
|
Optimization algorithm in this paper |
21577 |
3739 |
93.5% |
||||
1 Teapot point cloud dataset
2 Bun000 point cloud dataset
3 Free-form surface point cloud dataset
Point 3: The author has to compare more existing results with existing architectures.
Response 3: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 4: The literature should be in resent paper and compare the same.
Response 4: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 5: Kindly add resent advancement references.
Response 5: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 6: Kindly improve the explanation of your results.
Response 6: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
An improved point cloud registration method proposed in this paper. In the rough registration stage, ISS feature point extraction is combined with octree to extract feature points, and then noise points are set to verify the feasibility of the method. Finally, the mismatching point pairs are eliminated by combining edge feature point removal algorithm with RANSAC algorithm. In the precise registration stage, a new nearest neighbor selection method is proposed, and the ICP(Iterative Closest Point) precise registration algorithm is improved by using curvature and FPFH features, and is verified by experiments. Compared with the traditional algorithm, the speed and accuracy of registration in this paper have been improved, and the research conclusions are as follows:
1 In the rough registration stage, this paper improves the point cloud feature point extraction method and point cloud filtering method, and then verifies it by experiments. From the data in Table 1, it can be seen that the removal rates of noise points on the three point cloud data sets have reached 95.3%, 98.6% and 93.5% respectively, and the accuracy and robustness of rough registration have been improved.
2 In the precise registration stage, the feasibility of this method is verified by experiments. Taking Stanford rabbit and free-form surface as examples, it can be seen from the data in Figure 12 that the error of the improved algorithm is reduced by 40.16% and 36.27% respectively, and the number of iterations is reduced by 42.9% and 37.14% respectively. Compared with the traditional ICP algorithm, the improved algorithm has higher registration accuracy, faster running speed and stronger versatility.
The point cloud registration method proposed in this paper improves the registration accuracy and iteration rate of complex surface grinding and polishing precision machining, and has good registration effect. The matching analysis between complex surface parts (machined blanks or semi-finished parts) and theoretical models can be completed in the precise machining process of complex surface grinding and polishing, and the machining allowance of each position of complex surface grinding and polishing parts can be obtained, which optimizes the subsequent machining process parameters and process planning, and further improves the machining accuracy and efficiency of complex surface grinding and polishing. This study has good theoretical significance and important engineering application value for the future development of grinding and polishing precision machining of complex surfaces.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Although authors presented a level of novel in the paper however there are many corrections to be attended to before the paper can be recommended for publication
1. Many acronyms were not defined. For examples ICP; GO-ICP; SHOT etc
2. What do you mean by datas
3. The reference format within the paper are not appropriate
4. Figure 1 should be referenced
5. many of the parameters in eqn. (1) - (5) are not defined
6. Authors should reference eqn. (12)
7. What is RANRAC?
8. Figure 12 not too clear
Author Response
Point 1: Many acronyms were not defined. For examples ICP; GO-ICP; SHOT etc
Response 1: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
ICP (Iterative Closest Point);
Go-ICP (Globally optimal Iterative Closest Point );
SHOT (Signature of Histogram of Orientation);
G-ICP (Generalized Iterative Closest Point);
N-ICP (Normal Iterative Closest Point);
4PCS (4-Points Congruent Sets);
ISS (Internal Shape Signatures) ;
SPFH (Simple Point Feature Histograms);
RANSAC (Random Sample Consensus);
RMSE (Root Mean Square Error)
Point 2: What do you mean by datas?
Response 2: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Regarding the word "datas", I just want to express the effect achieved on three point cloud data sets. Thank you very much for your valuable comments.
Point 3: The reference format within the paper are not appropriate
Response 3: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 4: Figure 1 should be referenced
Response 4: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 5: many of the parameters in eqn. (1) - (5) are not defined
Response 5: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 6: Authors should reference eqn. (12)
Response 6: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 7: What is RANRAC?
Response 7: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
RANSAC (Random Sample Consensus) algorithm is an algorithm to calculate the mathematical model parameters of data according to a group of sample data sets containing abnormal data, and obtain effective sample data. It is assumed that the sample contains correct data (data that can be described by the model) and abnormal data (data that deviates far from the normal range and cannot adapt to the mathematical model), that is, the data set contains noise. These abnormal data may be caused by wrong measurement, wrong assumption, wrong calculation, etc. At the same time, RANSAC also assumes that given a set of correct data, there is a way to calculate the model parameters that conform to these data.
Point 8: Figure 12 not too clear（图12不太清晰）
Response 8: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Author Response File: Author Response.docx
Reviewer 3 Report
The term ICP appears first in the abstract but is ambiguous for general readers because it is not spelled out.
Please use a high-resolution figure to avoid blurring of edges, e.g., Fig. 1. Also, the font appears too small.
Proof-read the paper again. Please use technical language and avoid conjugation of words, e.g., p. 3 line 95 "can't." Check the use of punctuations and fragments/phrases in your sentences, e.g., p. 5 line 130.
Please ensure that proper references are used to support claims and statements, e.g., p.6 line 164 - where references are missing to attribute the data set used. Also, the equations used should cite where they are extracted from.
It is not very clear how validation/testing of the algorithms was carried out in the experiment. Please clarify.
How much of the error and iteration improvement in the new algorithm is attributed to the noisy data points? This is very important because this will tell how discriminatory the model is against noise. Please clarify.
The value of the results from the findings of this study is not clear. Please elaborate in the conclusion section on the significance of the findings and how they contribute to the growing body of knowledge in this field.
Author Response
Point 1:The term ICP appears first in the abstract but is ambiguous for general readers because it is not spelled out.
Response 1: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 2:Please use a high-resolution figure to avoid blurring of edges, e.g., Fig. 1. Also, the font appears too small.
Response 2: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 3:Proof-read the paper again. Please use technical language and avoid conjugation of words, e.g., p. 3 line 95 "can't." Check the use of punctuations and fragments/phrases in your sentences, e.g., p. 5 line 130.
Response 3: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
Point 4:Please ensure that proper references are used to support claims and statements, e.g., p.6 line 164 - where references are missing to attribute the data set used. Also, the equations used should cite where they are extracted from.
Response 4: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
The three point cloud data sets used in this paper are the public data sets bunny and teapot provided by Stanford University and the free-form surface point cloud data set measured by self-built machine tools, and the public data sets bunny and teapot can be downloaded and consulted in the Stanford University database.
Point 5:It is not very clear how validation/testing of the algorithms was carried out in the experiment. Please clarify.
Response 5: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
In the experiment, we first calculate the Root Mean Square Error (RMSE). Comparing the calculated error value between our improved method and the comparison method at the same iteration number, the relative accuracy of the smaller error value is higher; On the contrary, when the two methods are at the same error value, the improved method can be achieved by fewer iterations, which proves that the registration speed has been improved.
Point 6:How much of the error and iteration improvement in the new algorithm is attributed to the noisy data points? This is very important because this will tell how discriminatory the model is against noise. Please clarify.
Response 6: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
For noise data points, the step of removing noise points in the new algorithm mainly appears in the rough registration stage, and we use three point cloud data sets to verify it. Because the measurement environment of Stanford University's public data set is good, we manually added 4000 noise points. It can be seen that the new algorithm improves the noise points of bunny, teapot and freeform surface by 200, 32 and 35 respectively compared with the original denoising method, so the improved algorithm can improve the noise point removal rate by about 2% for point cloud data in complex environment, it accounts for about 5% of the total noise points.
Point 7:The value of the results from the findings of this study is not clear. Please elaborate in the conclusion section on the significance of the findings and how they contribute to the growing body of knowledge in this field.
Response 7: The authors sincerely thank the reviewer for raising such a comment. We have made correction according to the reviewer’s comments.
The point cloud registration method proposed in this paper improves the registration accuracy and iteration rate of complex surface grinding and polishing precision machining, and has good registration effect. The matching analysis between complex surface parts (machined blanks or semi-finished parts) and theoretical models can be completed in the precise machining process of complex surface grinding and polishing, and the machining allowance of each position of complex surface grinding and polishing parts can be obtained, which optimizes the subsequent machining process parameters and process planning, and further improves the machining accuracy and efficiency of complex surface grinding and polishing. This study has good theoretical significance and important engineering application value for the future development of grinding and polishing precision machining of complex surfaces.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I have no comments
Author Response
I am very grateful to the reviewers for their guidance and recognition of my work, and I would like to express my heartfelt thanks to the reviewers.
