A Novel 4D Track-before-Detect Approach for Weak Targets Detection in Clutter Regions

Round 1

Reviewer 1 Report

1. There are several typos in the paper. A careful revision of the text is recommended.
   
2. The acronyms are not all defined, which leads to a difficult understanding.
3. The discussion of the problem in question seems difficult to understand.
   
4. It would be worth mentioning some fundamental papers on TBD as:
   • Track-before-detect algorithms for bistatic sonars
     D Orlando, F Ehlers, G Ricci
     2010 2nd International Workshop on Cognitive Information Processing, 180-185
   • Batch tracking algorithm for multistatic sonars
     F Ehlers, D Orlando, G Ricci
     IET Radar, Sonar & Navigation 6 (8), 746-752
   • Track-before-detect strategies for STAP radars
     D Orlando, L Venturino, M Lops, G Ricci
     IEEE Transactions on Signal Processing 58 (2), 933-938

 

Author Response

Answer to the Editor and the Reviewers’ comments on the paper
A Novel 4D track-before-detect approach for weak targets detection in clutter regions
Yan Bo, Zhang hua, Xu Luping, Chen Yu, Lu Hongmin
Remotesensing-1429182
SUMMARY OF REVISIONS

 

The authors would like to thank the Editor and the Reviewers for the comments and suggestions expressed during the review process of our paper. We revised the manuscript to improve the quality of the presentation and satisfy the requests of the Reviewers. Based on the comments of the Reviewers, the main changes are succinctly summarized in the following:

• The most recent literature on TBD methods has been added in the introduction

to make the literature review more comprehensive.

• 3.4 theoretical model and Sec.3.5 implementation of whole algorithm have been added. In Sec.3.4, the superiority of method is explained and proved in theory. In Sec.3.5, the whole algorithm is explained by using Pseudocode and diagram.
• We compared our solution with three new state-of-the-art methods (CPHD, Bernoulli filter and GLMB filter) present in the literature using both Monte Carlo simulation and experimental results.
• We summarized all the parameters using in experiments in Tab.3 to make the method more understandable.
• The parameters and corresponding description using in this manuscript have been well listed in appendix.
• The Sec. 4.1. Synthetic data experiment has been reorganized. Subtitles has been added to make it easier to understand.
• The measurement of two real experiments and corresponding ground truth have been patched with the paper to make the research more reproducible.

 

All the comments of the Reviewers have been carefully addressed, and the paper has been revised accordingly. The answers to all the comments are reported in detail below and the changes with respect to the previous manuscript are explained. In the revised manuscript, the most relevant changes are reported in yellow.

 

 

 

 

 

Q1: There are several typos in the paper. A careful revision of the text is recommended.
R1: Thanks for pointing out the typos. The manuscript has been polished several times already.

 

 

Q2:The acronyms are not all defined, which leads to a difficult understanding.

R2: In the revised version, the acronyms have been checked, meanwhile all the symbols and corresponding description using in this manuscript have been well listed at the end of paper, which make this work much easier to understand.

 

 

Q3:The discussion of the problem in question seems difficult to understand.
R3: Two improvements have been made to make it easier to understand.

(1)  The Sec.3.4 theoretical model and Sec.3.5 implementation of whole algorithm have been added. In Sec.3.4, the superiority of method is explained and proved in theory.

In Sec.3.5, the whole algorithm is explained by using Pseudocode and diagram.

(2)  The Sec. 4.1. Synthetic data experiment has been reorganized. It has been split into several smaller sections:

4.1. Synthetic data experiment

  4.1.1. Optimal threshold estimation

  4.1.2. Scenario and parameters of synthetic experiment

  4.1.3. Result of the proposed method

4.1.4. Result of non TBD methods

4.1.5. Result of Multi-dimensional HT-TBD

4.2. Real data experiment

  4.2.1. Experiment 1

  4.2.2. Experiment 2

 

Q4:It would be worth mentioning some fundamental papers on TBD as:

R4: The references [6-8] have been included in the revised manuscript to make the literature review more comprehensive.

 

[6]. Orlando, Danilo and Ehlers, Frank and Ricci, Giuseppe. Track-before-detect algorithms for bistatic sonars 2010 2nd International Workshop on Cognitive Information Processing 2010.180–185.

[7]. Ehlers, F and Orlando, D and Ricci, G, Batch tracking algorithm for multistatic sonars IET Radar Sonar Navigation 2012.6,8.746-752.

[8]. D. Orlando and L. Venturino and M. Lops and G. Ricci, Track-Before-Detect Strategies for STAP Radars IEEE Transactions on Signal Processing 2010.58,2.933-938.

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

A four-dimensional (4D) track-before–detect (TBD) approach for closely weak extended target tracking and overcoming heterogeneous clutter background is proposed in this manuscript. The manuscript isn’t well written. Authors are given theoretical analysis and experimental results related to clutter region suppression, track fusion, determining thresholds of the proposed method, but there is no complete theoretical basis for a proposed algorithm. The manuscript needs to be corrected to make it more understandable to the reader. The theoretical model of the proposed algorithm is missing, as well as the complete mathematical model of the same, so that the reader can reconstruct and test the proposed method. Instead, a flowchart representation of the proposed algorithm is given (Figure 1).

Based on the given equations and figures, the authors did not present evidence for the claims that were made. Authors are recommended to apply the methodology used by other authors from the references they cited in their work (e.g. [13-15], [24-28], etc…) in order to make it clear to the reader on which (nonlinear) known methods the proposed algorithm is based on.

Especially, more details should be given on the application of the PHD algorithm used by the authors to compare with the proposed algorithm.

 

Suggestions for improvement the manuscript:

 

• Page 2, Section 2, Equations 1: It is common to use a state vector instead of individual axes, for dynamical model.
• Page 4, Section 3: Give a complete (probably nonlinear) theoretical model of the proposed algorithm
• Page 19 Figure 7: Provide more details for the approach of the PHD algorithm you used when compiling the proposed algorithm

Author Response

Answer to the Editor and the Reviewers’ comments on the paper
A Novel 4D track-before-detect approach for weak targets detection in clutter regions
Yan Bo, Zhang hua, Xu Luping, Chen Yu, Lu Hongmin
Remotesensing-1429182
SUMMARY OF REVISIONS

 

The authors would like to thank the Editor and the Reviewers for the comments and suggestions expressed during the review process of our paper. We revised the manuscript to improve the quality of the presentation and satisfy the requests of the Reviewers. Based on the comments of the Reviewers, the main changes are succinctly summarized in the following:

• The most recent literature on TBD methods has been added in the introduction

to make the literature review more comprehensive.

• 3.4 theoretical model and Sec.3.5 implementation of whole algorithm have been added. In Sec.3.4, the superiority of method is explained and proved in theory. In Sec.3.5, the whole algorithm is explained by using Pseudocode and diagram.
• We compared our solution with three new state-of-the-art methods (CPHD, Bernoulli filter and GLMB filter) present in the literature using both Monte Carlo simulation and experimental results.
• We summarized all the parameters using in experiments in Tab.3 to make the method more understandable.
• The parameters and corresponding description using in this manuscript have been well listed in appendix.
• The Sec. 4.1. Synthetic data experiment has been reorganized. Subtitles has been added to make it easier to understand.
• The measurement of two real experiments and corresponding ground truth has been patched with the paper to make the research more reproducible.

 

All the comments of the Reviewers have been carefully addressed, and the paper has been revised accordingly. The answers to all the comments are reported in detail below and the changes with respect to the previous manuscript are explained. In the revised manuscript, the most relevant changes are reported in yellow

 

 

 

 

 

Q1: A four-dimensional (4D) track-before–detect (TBD) approach for closely weak extended target tracking and overcoming heterogeneous clutter background is proposed in this manuscript. The manuscript isn’t well written. Authors are given theoretical analysis and experimental results related to clutter region suppression, track fusion, determining thresholds of the proposed method, but there is no complete theoretical basis for a proposed algorithm. The manuscript needs to be corrected to make it more understandable to the reader.

R1: Thank you very much for the time and efforts spent in reading our manuscript and providing constructive comments that significantly improved our work. We are glad the Reviewer appreciated the relevance of our contribution. We have addressed all the observations of this Reviewer, trying to clarify all the unclear aspects. To make it more understandable, following revisions have been addressed.

(1) Sec.3.4 theoretical model and Sec.3.5 implementation of whole algorithm have been added.

(2) Detailed explanation about Pseudocode and example in diagram have been added.

(3) All the description of parameters and symbols are listed at the end of paper.

 

Q2:The theoretical model of the proposed algorithm is missing, as well as the complete mathematical model of the same, so that the reader can reconstruct and test the proposed method. Instead, a flowchart representation of the proposed algorithm is given (Figure 1).

R2: This is a very good point to improve this paper. Therefore, Sec.3.5 implementation of whole algorithm have been added. Each step of proposed method is detailly explained. Meanwhile, the description of example given in Fig.2 has been added. The Fig.2 greatly improves the visualization of our methods. We think the Sec.3.5 is helpful to readers to reconstruct and test the proposed method.    

 

Q3:Based on the given equations and figures, the authors did not present evidence for the claims that were made. Authors are recommended to apply the methodology used by other authors from the references they cited in their work (e.g. [13-15], [24-28], etc…) in order to make it clear to the reader on which (nonlinear) known methods the proposed algorithm is based on.

R3: You are right, the proposed method is based on a nonlinear model. Therefore, in the revised version, Sec.3.4 theoretical model has been added here, the merit of our method is explained and proved in theory. Meanwhile, more state-of-art works have been compared using Monte Carlo simulation and experimental results. We think it is helpful to present the superiority of our method.

Meanwhile, the Pseudocode is explained line by line, the description of example in Fig.2 has also been added. All of this are helpful to readers to understand the proposed method.

 

Suggestions for improvement the manuscript:

• Q4: Page 2, Section 2, Equations 1: It is common to use a state vector instead of individual axes, for dynamical model.

R4: The Eq.1 has been revised in using state vector.

• Q5: Page 4, Section 3: Give a complete (probably nonlinear) theoretical model of the proposed algorithm

R5: Sec.3.4 and Sec.3.5 have been added in the revised version. In Sec.3.4 Theoretical model has been added to explain the proposed method and its superiority in theory. In Sec.3.5, whole algorithm is detailed introduced by Pseudocode and example in diagram.

 

 

• Q6:Page 19 Figure 7: Provide more details for the approach of the PHD algorithm you used when compiling the proposed algorithm

R6: All the parameters using in experiments are summarized in Tab.3 to make the methods (ET-PHD, CPHD, Bernoulli filter and GLMB filter) more understandable.

The parameters of Multi-dimensional HT-TBD is also mentioned in Sec. 4.1.5. Such as “The width of gradient bin and intercept bin is 25 and 100 respectively”.

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper is interesting and proposes novel tracking approach and demonstrate it with synthetic and real dataset.
Some review comments to improve the paper further -

1. Comparison with other approaches than 'ET-PHD [32]' is recommended.
2. Normalized innovation square (NIS) over frames along with the confidence score plots are suggested
3. Furthermore, probability of detection vs probability of false alarm would be a good metric to also benchmark the proposed solution
4. The authors can publicly upload the dataset to make the research more reproducible.

Author Response

Answer to the Editor and the Reviewers’ comments on the paper
A Novel 4D track-before-detect approach for weak targets detection in clutter regions
Yan Bo, Zhang hua, Xu Luping, Chen Yu, Lu Hongmin
Remotesensing-1429182
SUMMARY OF REVISIONS

 

The authors would like to thank the Editor and the Reviewers for the comments and suggestions expressed during the review process of our paper. We revised the manuscript to improve the quality of the presentation and satisfy the requests of the Reviewers. Based on the comments of the Reviewers, the main changes are succinctly summarized in the following:

• The most recent literature on TBD methods has been added in the introduction

to make the literature review more comprehensive.

• 3.4 theoretical model and Sec.3.5 implementation of whole algorithm have been added. In Sec.3.4, the superiority of method is explained and proved in theory. In Sec.3.5, the whole algorithm is explained by using Pseudocode and diagram.
• We compared our solution with three new state-of-the-art methods (CPHD, Bernoulli filter and GLMB filter) present in the literature using both Monte Carlo simulation and experimental results.
• We summarized all the parameters using in experiments in Tab.3 to make the method more understandable.
• The parameters and corresponding description using in this manuscript have been well listed in appendix.
• The Sec. 4.1. Synthetic data experiment has been reorganized. Subtitles has been added to make it easier to understand.
• The measurement of two real experiments and corresponding ground truth has been patched with the paper to make the research more reproducible.

 

All the comments of the Reviewers have been carefully addressed, and the paper has been revised accordingly. The answers to all the comments are reported in detail below and the changes with respect to the previous manuscript are explained. In the revised manuscript, the most relevant changes are reported in yellow

 

 

 

 

 

Q1. The paper is interesting and proposes novel tracking approach and demonstrate it with synthetic and real dataset. Some review comments to improve the paper further

R1: Thank you very much for the time and efforts spent in reading our manuscript and providing constructive comments that significantly improved our work. We are glad the Reviewer appreciated the relevance of our contribution. We have addressed all the observations of this Reviewer, trying to clarify all the unclear aspects.

 

Q2. Comparison with other approaches than 'ET-PHD [32]' is recommended.

R2: We compared our solution with three new state-of-the-art methods (CPHD[37], Bernoulli filter [38] and GLMB filter[40]) present in the literature using both Monte Carlo simulation and experimental results.

 

The result of methods using Synthetic data is summarized in Tab.4 and corresponding analysis is patched in Sec.4.1.4. Result of non TBD methods.  

The result of methods using real data is presented in Fig.8 and Fig.11, the overall detection rate, false alarm rate, OSPA distance are summarized in Tab.5.

 

The result further presents the superiority of our method in decreasing false alarm track rate and in increasing detection rate.

[37]. B.-T. Vo and B.-N. Vo and A. Cantoni, Analytic Implementations of the Cardinalized Probability Hypothesis Density Filter IEEE Transactions on Signal Processing 2007.55,7.3553-3567.

[38]. B.-T. Vo and C.M. See and N. Ma and W.T. Ng, Multi-Sensor Joint Detection and Tracking with the Bernoulli Filter IEEE Transactions on Aerospace and Electronic Systems 2012.48,2.1385-1402.

[40]. B.-N. Vo and B.-T. Vo and H. Hung, An Efficient Implementation of the Generalized Labeled Multi-Bernoulli Filter IEEE Transactions on Signal Processing 2017.65,8.1975-1987.

 

Q3. Normalized innovation square (NIS) over frames along with the confidence score plots are suggested

R3: The innovation square over frames is presented in Fig.8c and Fig.11c. The innovation square is applied here mainly because the covariance in calculating NIS is not used in the proposed method. Therefore, only the difference between the prediction and the selected measurement point is estimated.

In Fig.8c and Fig.11c, innovation square which evaluates the difference between measurement and corresponding prediction is presented. In most of scans, even the scans in which target is maneuvering, innovation of method is stable and similar with the others. It means that our tracking method is stable to access correct measurement point and to get the suitable prediction even if the target is maneuvering.

Q4. Furthermore, probability of detection vs probability of false alarm would be a good metric to also benchmark the proposed solution

R4: In Tab.4 (Synthetic data) and Tab.5 (real cases), the probability of detection vs probability of false alarm is calculated. The result infers that the detection vs false alarm rate of the proposed method is much higher than that of the other four comparisons.

Q5. The authors can publicly upload the dataset to make the research more reproducible.

R5: The measurement of two real experiments and corresponding ground truth has been patched with the paper to make the research more reproducible. Meanwhile, a Matlab demo is also patched to present the data.

 

Author Response File: Author Response.pdf