A Two-Stage Feature Point Detection and Marking Approach Based on the Labeled Multi-Bernoulli Filter
Round 1
Reviewer 1 Report
This paper claims to propose a two-stage RFS technique for SLAM. A labeled random finite set is followed by an LMB technique to detect feature points and marking. Authors propose to path plan in two stages, (a) global path planning and (b) local path planning. Authors claim the method provides a better estimation accuracy and a better back-end optimization.
here are a few points:
- This paper must be proofread for example lines 48, 114, 115, etc..
- Contribution is moderate at best. Highlighting the main contribution can help. It seems like the paper is just a combination of techniques which makes this paper highly incremental.
- literature review suffers from the lack of comparison between RFS techniques and other Bayesian techniques such as Bayesian nonparamterics. Also, there are many related papers that are swept under the rug.
- Using LMB is costly. What is the computational cost of this method? In addition due to many issues with LMB, I was wondering if the authors could justify the choice of LMB compared to other state-of-the-art RFS techniques.
- There is no evidence that this method works better compared to other methods. Not only does this method seem to suffer from high computational cost but this algorithm does also not seem to be able to outperform other existing techniques.
- Lack of enough experimental results: It is suggested authors provide experimental results regarding (a) computational costs (b) comparison to other existing RFS-based SLAM techniques (c) comparison to other Bayesian techniques. The simulation results do not confirm the authors' claim regarding better efficiency and better back-end optimization.
Author Response
We thank two reviewers for their suggestions. These comments are valuable and helpful to our paper. We have read all these suggestions carefully and made corrections and improvements one by one. Based on the instructions, we upload the revised manuscript file. We hope the new version is smoother and clearer. For the specific reviewer's responses, we present in following.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors consider a complex tracking problem, with random finite sets (RFS) to solve the issue of simultaneous localization and mapping (SLAM). Moreover, they also consider that the sensors are moving during the tracking problem. I have some suggestions for improving the manuscript. See my comments below.
- Please, improve the caption of Figure 1 adding more explanations.
-The paper is quite difficult to follow. Can you provide a brief summary/ sketch of the overall paper after the introduction? in order to help the reader in following each step.
- The state-of-the-art discussion must be improved. Several adjacent particle filtering schemes for tracking and model selection can be used here for addressing your problem. For instance, consider
C. C. Drovandi, J. McGree, and A. N. Pettitt. A sequential Monte Carlo algorithm to incorporate model uncertainty
in Bayesian sequential design." Journal of Computational and Graphical Statistics, 23(1):324, 2014.
L. Martino, J. Read, V. Elvira, F. Louzada, Cooperative Parallel Particle Filters for on-Line Model Selection and Applications to Urban Mobility, Digital Signal Processing Vol. 60, pp. 172-185, 2017.
I. Urteaga, M. F. Bugallo, and P. M. Djuric. Sequential Monte Carlo methods under model uncertainty, IEEE Statistical Signal Processing Workshop (SSP), pages 15, 2016.
C. M. Carvalho, M. S. Johannes, H. F. Lopes, and N. G. Polson. Particle Learning and Smoothing. Statist. Sci., Volume 25, Number 1 (2010), 88-106.
This discussion can increase also the impact of your work.
Author Response
We thank two reviewers for their suggestions. These comments are valuable and helpful to our paper. We have read all these suggestions carefully and made corrections and improvements one by one. Based on the instructions, we upload the revised manuscript file. We hope the new version is smoother and clearer. For the specific reviewer's responses, we present in following.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
I appreciate the authors' response and their effort to proofread the paper however the following points still stand:
- this paper can be hard to follow at times but this is not a significant issue.
- The contribution is in fact moderate. It is basically a combination of models. But I appreciate the authors highlighting their contribution.
- Literature review lacks a discussion on the Bayesian methods, and several references on this topic are missing. Lines 82-83 are indeed false or at least need more justifications. It has been shown that Bayesian nonparametric techniques can and do outperform RFS techniques in both computational cost and accuracy, check for instance:
+ Random infinite tree and dependent Poisson diffusion process for nonparametric Bayesian modeling in MOT.
+ Inference for MOT: A Bayesian nonparametric approach.
It is highly recommended authors provide a literature review on the Bayesian nonparametric techniques and highlight why would one would want to choose their technique over their counterparts. The lines added to the paper are insufficient and have no reference to back up their claims.
- As mentioned above, there has been extensive work on this topic from the Bayesian nonparametric point of view and an exhaustive literate review is needed to show the advantage of their technique.
- I still strongly believe there is no evidence this method has an advantage over other techniques, especially nonparametric methods. I would also like to point out that the provided graphs on the rebuttal are obtained by running only 10 MC which is not many iterations to provide accurate results. the more iteration the closer you get to the true underlying parameters.
- It is necessary to show the OSPA comparison between their method, RFS-based techniques, and Bayesian nonparametric techniques. As mentioned earlier BNP techniques are proven to outperform RFS methods, e.g. see Bayesian nonparametric modeling for predicting dynamic dependencies in MOT.
Overall, this paper lacks justification as to why this approach would be of interest and would be chosen over its counterparts. Simulation results are not sufficient to support the authors' claim. There should not be any unjustified claims in the paper. I highly recommend authors correct those unjustified claims and be precise regarding the significance of their method. A careful design of the experiment can show if their method outperforms other techniques and if not provide justification.
Author Response
We thank reviewer for their suggestions. These comments are valuable and helpful to our paper. We have read all these suggestions carefully and made corrections and improvements. Based on the instructions, we upload the revised manuscript file. We hope the new version is smoother and clearer. For the specific reviewer's responses, we present in following.
Author Response File: Author Response.pdf
