FFT-Based Scan-Matching for SLAM Applications with Low-Cost Laser Range Finders

Round 1

Reviewer 1 Report

nice piece of work in general. you definitely need to update your reference list because in a total of 37 refs only the 10 (!) come out of the last 5 years. In a technological paper like this, the relevant bibliography needs to be more that 50% from the last 5 years. In order to enrich your list, you could include relevant works, like:

‘Design of an Autonomous Robotic Vehicle for Area Mapping and Remote Monitoring’, International Journal of Computer Applications, (ISSN: 0975 – 8887), Vol. 167, No 167, June 2017

Author Response

Thanks very much for the reminder of the Perfect-match algorithm. We have downloaded the papers and studied carefully. The reference of these worked is also added to this paper( Line 512, reference [38] and [39] ). However, we were unable to found their source code and evaluate their work objectively in such a short time. Where as the main contribution of our work is to improve the robustness of scan-to-scan matching, as a result, we compared our method with well-known algorithms of ICP and NDT. The experiment part of this paper is revised.

Author Response File: Author Response.docx

Reviewer 2 Report

This manuscript addresses an approach of localization with low cost lidar with FFT methodology. It is well written and organized. The proposed method is well presented and justified.

It should be proper to compare the results with the Perfect-match well-known algorithm.

There are several literature papers that present the perfect match, such as:

https://link.springer.com/chapter/10.1007%2F11780519_13

https://link.springer.com/article/10.1007/s10846-017-0765-5

Author Response

Thanks very much for the reminder of the Perfect-match algorithm. We have downloaded the papers and studied carefully. The reference of these worked is also added to this paper( Line 512, reference [38] and [39] ). However, we were unable to found their source code and evaluate their work objectively in such a short time. Whereas the main contribution of our work is to improve the robustness of scan-to-scan matching, as a result, we compared our method with well-known algorithms of ICP and NDT. The experiment part of this paper is revised.

Author Response File: Author Response.docx

Reviewer 3 Report

Authors propose a mapping approach sustained by a scan-mathing method, enhanced with the use of FFT. Though the soundness is valid, I have some concersn authors should focus on:

l16. robot IS running.

It is necessary to specify in the Abstract whether the FFT is applied to image data or to laser data.

l41. Sustain all these methods with references. Check the endless list of up-to-date works in the MDPI data base (either of this Jornals: Sensors, Remote Sensing, Electronics, ...)

l41. Montecarlo, rather than Carlo?

l47. Consider citing some of these references:

- Visual Information Fusion through Bayesian Inference for Adaptive Probability-Oriented Feature Matching

- Robust Visual Localization with Dynamic Uncertainty Management in Omnidirectional SLAM

I have a main concern with the Introducction: no specific discussion on appearance-based methods (such as FFT or Fourier Signature...or either HOG, Gist, etc) is presented. I would expect another paragraph with cites are introduced to articles where localization or mapping is achieved by using such sort of appearance-based visual descriptors. Citation to authors such as E. Menegatti and L. Payá may be required. If contrarily, the FFT approach works only over laser data, establish an analog discussion on the treatment of such data with these sort of appearance methods.

l153. Take care of the inline equations and nomenclature terms to appear aligned with the text. Aply to the rest of the manuscript. You can just use $variable$ in Latex.

l252. "About" is not an accepted term for a scientific paper. State the accurate characteristic of the entire equipment and their brandname.

l271. Table I. Specify how the ground truth is established.

Other major concers:

-If digital images are used, please show real results, before and after registration.

- Why not the PCL library (C++) is used? Justify.

- What is the real difference between this proposal and the former works of the authors? This should be clearly explained.

- Comparison results with other methods should be stated. Either with the authors' formers approaches, if applicable. Revision marked as MAJOR due to this aspect. Authors should take time to produce more real, and comparative results with other methods.

Author Response

Thank you very much for the valuable comments. We have seriously considered these comments and suggestions, and carefully revised the manuscript accordingly. The responses to comments are as follows.

Comment 1:

“L16. robot IS running.”

Response 1:

Corresponding parts have been modified.

Line 18: “Traditional scan matching algorithms may often fail while robot is running too fast in complex environments.” 

 

Comment 2:

“It is necessary to specify in the Abstract whether the FFT is applied to image data or to laser data.”

Response 2:

The FFT is applied to the image data.

Line 22 : “FFT is applied to the images to get the rotation angle and translation parameters. ”

 

Comment 3:

“L41. Montecarlo, rather than Carlo?”

Response 3:

Corresponding parts have been modified.

Line 46: ”Kalman Filter, Markov, Montecarlo and Extended Kalman Filter (EKF).“

 

Comment 4:

”L41. Sustain all these methods with references. Check the endless list of up-to-date works in the MDPI data base (either of this Jornals: Sensors, Remote Sensing, Electronics, ...)

Consider citing some of these references:

- Visual Information Fusion through Bayesian Inference for Adaptive Probability-Oriented Feature Matching

- Robust Visual Localization with Dynamic Uncertainty Management in Omnidirectional SLAM

I have a main concern with the Introduction: no specific discussion on appearance-based methods (such as FFT or Fourier Signature...or either HOG, Gist, etc) is presented. I would expect another paragraph with cites are introduced to articles where localization or mapping is achieved by using such sort of appearance-based visual descriptors. Citation to authors such as E. Menegatti and L. Payá may be required. If contrarily, the FFT approach works only over laser data, establish an analog discussion on the treatment of such data with these sort of appearance methods.“

Response 4:

Added 10 more recent related works, according to the reviewer’s suggestions.

The two listed papers is added in reference [11] (Line 456) and [12] (458) .

modified the introductions：

Line 52：Vision-based approaches are wildly studied in recent years [4]. They analysis and track natural landmarks [5], significant features [6,7], or image pixel value directly [8,9,10] positioning and mapping. Special cameras are used for getting better performance under different situations, like omnidirectional camera approaches [11,12] , RGB-D camera approaches [10,13], etc.

modified the introduction in Line 114:

“If the scan data is considered as image, which indicates the outline of obstacles near the robot. The scan matching problem can be transformed to image registration problem. There are many image registration approaches that can match two positions in long distance without any good initial guess for searching. However, scan image is quite different from normal images, since it is sparse with points collected by lidar. As a result, normal feature-based image registration methods, such as SIFT [40], SURF [41], ORB [42], etc., may not be suitable. We need to find the shift of the image by considering all the data in the image. By changing the image to the frequency domain, the FFT transform is suitable for this aspect. The idea in this paper is mainly derived from FFT based image registration [43-46], because the scan image is simple and lies in one plane.”

 

Comment 5:

“L153. Take care of the inline equations and nomenclature terms to appear aligned with the text. Aply to the rest of the manuscript. You can just use $variable$ in Latex.”

Response 5:

Corresponding parts have been modified.

Comment 6:

”L252. "About" is not an accepted term for a scientific paper. State the accurate characteristic of the entire equipment and their brand name.“

Response 6:

Added Table I to list the parameters of RPLidar A1. (Line 318).

 

Comment 7:

“L271. Table I. Specify how the ground truth is established.”

Response 7:

In Section 5(A) (Line 321), we added :“The ground truth of real-world experiments were collected by manual measurement of key positions.“

 

Other major concerns:

Comment 8:

“-If digital images are used, please show real results, before and after registration. ”

Response 8:

The example of scan image is shown in Figure 2, Line 170. An overview framework with middle results is added in Line 311 ,Figure 5. 

As the main job of this work is to estimate the pose between scans. We also added the step output to the scans pose in the experiment part of Figure 7 on Line 345.

 

Comment 9:

“- Why not the PCL library (C++) is used? Justify.”

Response 9:

The Point Cloud Library (PCL) is a large scale, open project for point cloud processing. It contains numbers of numerous state-of-the art algorithms including filtering, feature estimation, surface reconstruction, 3D registration, model fitting and segmentation. However, for the scan matching problem needed for SLAM, especially when code should be ready for running on a low-cost platform, most of the parts of PCL are not needed. As a result, we realized our work in C++ part of LIBICP, another popular low-weight library (http://www.cvlibs.net/software/libicp/).

 

Comment 10:

“- What is the real difference between this proposal and the former works of the authors? This should be clearly explained.”

Response 10:

As expressed in Line 123:

In our previous works [47,48], we purposed the FFT based scan-matching algorithm and its improvement FFT-ICP. In this paper, we detailed previous work in the data pretreatment part. According to the low detection range problem of low-cost LRF sensor, we added a pre-alignment module based on missing data features, to get the rough rotation angle before FFT processing under certain conditions. We also propose a new solution for FFT-ICP scan-matching with low-cost LRF sensors.

Comment 11:

“- Comparison results with other methods should be stated. Either with the authors' formers approaches, if applicable. Revision marked as MAJOR due to this aspect. Authors should take time to produce more real, and comparative results with other methods.”

Response 11:

We did a major revise of the experiment part. We compared purposed work with ICP、NDT, and our previous work FFT, FFT-ICP1, in processing speed, successful rate, and accuracy. We implemented dynamic experiments under different speed, compared with ICP and NDT.

 

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Authors have tacked my comments. Well done.

Please go on a deep revision of the writing and punctuation.

Ex: line 54 --> they ANALYSE.