RBCN-Net: A Data-Driven Inertial Navigation Algorithm for Pedestrians

Round 1

Reviewer 1 Report

In this manuscript, the authors investigate a data-driven inertial navigation algorithm for pedestrians. Specifically, the residual network ResNet18 is combined with NAM and CBAM attention modules to enhance the learning ability of the network for channel and spatial features and with the BiLSTM module to enhance the network’s capability of long-distance learning. Experimental results are shown to demonstrate the developed network design through comparison with other existing networks.

 

First of all, the importance and necessity of the considered problem are not well explained. Why is the accurate navigation for pedestrians are of particular interest and practical significance? Some background is needed before technical discussions.

 

More importantly, a large body of existing works on improvements of the IMU technology is missed in this manuscript, without which the approach proposed in this work can be hardly justified. In fact, the issues such as drift of IMU identified in this manuscript have been well addressed in the past decades for wide implementation including automobile, aerospace, etc. Combining the existing techniques such as NAM, CBAM, and BiLSTM with the existing residual network ResNet18 to enhance the ability of the network to learn channel and spatial features may be a selling point but not a major contribution to this area. As such, the novelty and contribution of this work is deemed to be minor.

 

Also, some important issues such as data security and privacy are ignored in this work. A big assumption of this work is the availability of access to the data from pedestrians. How about if pedestrians are not willing to share her or his data? How potential data security and privacy issues be addressed?

 

Last but not least, majority of the manuscript is rough to read due to missing words, lack of conjunctions and articles, grammar errors, and stray sentences.

 

Author Response

Response to Reviewer 1 Comments

 

Point 1: First of all, the importance and necessity of the considered problem are not well explained. Why is the accurate navigation for pedestrians are of particular interest and practical significance? Some background is needed before technical discussions.

Response 1: Thank you for your suggestion. As per your recommendation, we have added a section on the significance of high-precision pedestrian inertial navigation to the second paragraph of Chapter 1 “Introduction”.

Here is the section that we have added based on your suggestion：

The demand for indoor positioning is increasing as personalized networks become more popular. Indoor positioning technologies rely on various technologies, such as Bluetooth [1,2], WiFi [3], machine vision [4], ultra-wide band (UWB) [5], and inertial navigation [6]. The first four technologies are highly precise and relatively mature. However, each of these technologies has its own challenges to consider. Bluetooth and WiFi can be affected by electromagnetic interference and indoor obstacles. Machine vision needs to take into account user privacy concerns. UBW requires advanced facility deployment. Finally, inertial navigation can be challenged by drift during extended operation. Despite these limitations, no single positioning technology can provide high-accuracy positioning service stably. Multi-sensor fusion technology is currently under development to address this challenge. In the complex indoor environment, high-precision inertial navigation can provide reliable positioning service for pedestrians for a short period of time when other positioning technologies are temporarily disabled by electromagnetic interference. Additionally, it can provide correction information for other positioning technologies because inertial navigation is not disturbed by external factors and does not rely on any external signals. Therefore, high-precision pedestrian inertial navigation has high research value.

 

Point 2: More importantly, a large body of existing works on improvements of the IMU technology is missed in this manuscript, without which the approach proposed in this work can be hardly justified. In fact, the issues such as drift of IMU identified in this manuscript have been well addressed in the past decades for wide implementation including automobile, aerospace, etc. Combining the existing techniques such as NAM, CBAM, and BiLSTM with the existing residual network ResNet18 to enhance the ability of the network to learn channel and spatial features may be a selling point but not a major contribution to this area. As such, the novelty and contribution of this work is deemed to be minor.

Response 2: Thank you for your suggestion. We apologize for our negligence in not including this background information in our paper. There are two types of IMU sensors: platform inertial guidance systems and jetlink inertial guidance systems. Platform inertial guidance systems have high accuracy but are more expensive and have more complex hardware than jetlink systems. They are typically used in automotive, military, aerospace, and other industries. Platform inertial guidance systems rely on a mechanical stabilization platform structure, with changes in carrier heading detected by the gyroscope. The rotation of the stabilized platform is then controlled so that it tracks the east and north direction, allowing the accelerometer sensitive axis to always point in that direction.

Jetlink inertial guidance systems do not rely on a physical platform; instead, the accelerometer and gyroscope are fixed directly to the carrier, and the attitude matrix is updated based on the gyroscope output. The ratio vector is then projected and transformed into the navigation coordinate system. The IMU sensor built into a smartphone can be seen as a micro-electro-mechanical technology-based Jetlink inertial guidance system.

In 2010, Apple's iPhone 4 was the first smartphone to adopt ST's three-axis accelerometer LIS331DLH and three-axis gyroscope L3G4200D as built-in inertial sensors. Other mobile phone manufacturers followed suit, and accelerometers and gyroscopes gradually became standard sensors for smartphones. However, as shown in Table 1, the random wander noise of most smartphones with built-in IMUs is much higher than that of the typical platform-based KY-INS300A inertial guide.

Despite this limitation, we believe that using IMU sensors in smartphones for pedestrian navigation is a feasible solution. Since smartphones are ubiquitous and most people carry them while walking, using the built-in IMU to collect data does not require users to incur additional expenses or carry extra equipment.

Table 1 Performance Comparison of Classic Mobile IMU and KY-INS300A

Device	IMU	Acce Random Walk	GyroRandom Walk
KY-INS300A	KY-INS300A	0.05
iPhone7 Plus	InvenSense ICM-20600	0.1	4
iPhone 4	L3G4200D and LIS331DLH	0.218	8. 75
MPU6050	MPU6050	0.4	0.15
Huawei Mate 20	LSM330D and FXAS21002C	0.15	25

Based on your suggestion, we add the following sentence to the third paragraph of Chapter 1, “Introduction”.

The development of micro-electro-mechanical technology has led to the inclusion of inertial measurement units (IMUs) based on these systems in smart devices such as mobile phones, watches, and wristbands. The use of IMU sensors in these devices for inertial navigation is very convenient and does not require the purchase of additional expensive equipment. However, compared to platform-based inertial guidance, IMU sensors in smart devices have several limitations due to their lower cost and other factors. Specifically, they generally have lower accuracy and higher noise levels.

 

Point 3: Also, some important issues such as data security and privacy are ignored in this work. A big assumption of this work is the availability of access to the data from pedestrians. How about if pedestrians are not willing to share her or his data? How potential data security and privacy issues be addressed?

Response 3: Thank you for your suggestion. Currently, our work is in the algorithm research stage. Similar to the RIDI [9], RONIN[10], and IONet[11] algorithms, our aim is to enhance positioning accuracy and reduce trajectory deviation at turns. The IMU data required for our experiments is collected by our team members and teachers, and there has been no data leakage or privacy infringement. We will make our code and dataset publicly available on Github. In the future, we plan to streamline our code and port it to ARM hardware, with a focus on improving real-time performance, data security, and privacy concerns.

 

Point 4: Last but not least, majority of the manuscript is rough to read due to missing words, lack of conjunctions and articles, grammar errors, and stray sentences

Response 4: Thank you for your suggestions. We have made corrections to the incorrect sentences and have addressed grammatical errors.

In “Abstract”, we have revised the sentence "And by adding 8 the BiLSTM module, it enhances the network’s ability of long distance learning" to "Adding the BiLSTM module can enhance the network's ability to learn over long distances.".

In Chapter 1 “Introduction”, we have revised the sentence ”Inspired by RoNIN, we design a new deep neural network model, RBCN-Net (see Fig. 1), to improve the network’s ability to regress pedestrian motion features by combining the residual network ResNet18 [10] with NAM [11], CBAM [12] attention mechanisms, and BiLSTM [13]. This network improves the problem that the network cannot distinguish the importance of features by adding an attention mechanism and conducts comparison experiments in constructing multiple network models with RBCN-Net on the dataset VOIMU to verify the ability of RBCN-Net to improve navigation accuracy.” to ” Taking inspiration from RoNIN, we have designed a new deep neural network model, called RBCN-Net (see Fig. 1), to improve the ability of the network to regress pedestrian motion features. We achieve this by combining ResNet18 [16], bidirectional long short term memory (BiLSTM) [17], a convolutional block attention module (CBAM) [18], and a normalization-based attention module (NAM) [19]. By adding the attention mechanism, we address the problem of the network's inability to distinguish the importance of features. To demonstrate the effectiveness of RBCN-Net, we conduct comparative experiments by constructing multiple network models using the VOIMU dataset to verify its ability to improve navigation accuracy.”.

In section 4.5 "Comparison of Base Network Models", we have revised "the" to "The" at the beginning of the paragraph.

In section 3.4 "RBCN-Net", we have revised the sentence ”After passing 8 Basicblock-NAM, the  obtains 512×7 feature vectors, which are then input to the BiLSTM layer, which has 2 hidden layers with input and output dimensions of 7. ” to ” After passing 8 Basicblock-NAM, the  obtains 512×7 feature vectors, and then input to the BiLSTM layer, which has 2 hidden layers with input and output dimensions of 7.”.

Author Response File: Author Response.docx

Reviewer 2 Report

In this paper, the authors proposed a positioning scheme. The topic is interesting, but some improvements are required.

1. In Section 1, the contents "The rest of the paper is organized as follows:..." are not the contributions of this paper, so it is better to correct as a separate paragraph.

2. All acronyms need to be well explained when they appear in your paper for the first time. "NAM" appears many times in the abstract and introduction, but its explanation appears in Section 2.

3. It would be better to add another chapter "The background and related work" to describe the basic research background and the related work in detail.

4. I think the following papers are related to your work.

[1]. "A vision-based indoor positioning systems utilizing computer aided design drawing"

[2]. "A 3-Dimensional Triangulation Scheme To Improve The Accuracy of Indoor Localization for IoT Services"

[3]. "A Slotted Random Request Scheme for Connectionless Data Transmission in Bluetooth Low Energy 5.0"

5. Is the proposed scheme compared with others?

 

Author Response

Response to Reviewer 2 Comments

 

Point 1: In Section 1, the contents "The rest of the paper is organized as follows:..." are not the contributions of this paper, so it is better to correct as a separate paragraph.

Response 1: Thanks to your suggestions, we have subdivided the 9th paragraph of Chapter 1 “Introduction”.

 

Point 2: All acronyms need to be well explained when they appear in your paper for the first time. "NAM" appears many times in the abstract and introduction, but its explanation appears in Section 2.

Response 2: Thanks to your suggestion, we revised paragraph 5 of Chapter 1 “Introduction”, by adding the full spelling of NAM, CBAM, and BiLSTM. Here is the section that we have modified based on your suggestion：

Taking inspiration from RoNIN, we have designed a new deep neural network model, called RBCN-Net (see Fig. 1), to improve the ability of the network to regress pedestrian motion features. We achieve this by combining ResNet18 [16], bidirectional long short term memory (BiLSTM) [17], a convolutional block attention module (CBAM) [18], and a normalization-based attention module (NAM) [19]. By adding the attention mechanism, we address the problem of the network's inability to distinguish the importance of features. To demonstrate the effectiveness of RBCN-Net, we conduct comparative experiments by constructing multiple network models using the VOIMU dataset to verify its ability to improve navigation accuracy.

 

Point 3: It would be better to add another chapter "The background and related work" to describe the basic research background and the related work in detail.

Response 3: Thanks to your suggestion, we have added new paragraphs 1-3 to Chapter 1 “Introduction”, in the text to provide background information and explain the significance of the study. Here is the section that we have modified based on your suggestion：

A successful outdoor positioning system based on the global positioning system (GPS) has been developed. However, satellite signals are weakened indoors, such as underground and inside large buildings, making GPS useless for positioning. Indoor navigation and positioning services can be made available to users through indoor location-based services, which can boost pedestrian travel efficiency and make it easier for managers to prioritize pedestrian traffic.

The demand for indoor positioning is increasing as personalized networks become more popular. Indoor positioning technologies rely on various technologies, such as Bluetooth [1, 2], WiFi [3], machine vision [4], ultra-wide band (UWB) [5], and inertial navigation [6]. The first four technologies are highly precise and relatively mature. However, each of these technologies has its own challenges to consider. Bluetooth and WiFi can be affected by electromagnetic interference and indoor obstacles. Machine vision needs to take into account user privacy concerns. UBW requires advanced facility deployment. Finally, inertial navigation can be challenged by drift during extended operation. Despite these limitations, no single positioning technology can provide high-accuracy positioning service stably. Multi-sensor fusion technology is currently under development to address this challenge.In the complex indoor environment, high-precision inertial navigation can provide reliable positioning service for pedestrians for a short period of time when other positioning technologies are temporarily disabled by electromagnetic interference. Additionally, it can provide correction information for other positioning technologies because inertial navigation is not disturbed by external factors and does not rely on any external signals. Therefore, high-precision pedestrian inertial navigation has high research value.

With the development of micro electro mechanical technology, Inertial measurement units (IMU) based on micro-electro-mechanical systems are typically included in smart devices like mobile phones, watches, and wristbands. Using IMU sensors in these smart devices for inertial navigation is very convenient and does not require the purchase of additional expensive equipment. However, IMU sensors in smart devices are much less capable than platform-based inertial guidance due to price and other factors, and generally have the disadvantages of low accuracy and high noise.

 

Point 4: I think the following papers are related to your work.

[1]. "A vision-based indoor positioning systems utilizing computer aided design drawing"

[2]. "A 3-Dimensional Triangulation Scheme To Improve The Accuracy of Indoor Localization for IoT Services"

[3]. "A Slotted Random Request Scheme for Connectionless Data Transmission in Bluetooth Low Energy 5.0"

Response 4: Thank you for bringing this to our attention, as it was an oversight on our part. We have added a new paragraph to Chapter 1 of the text to complement the classical indoor positioning techniques. In addition to citing the papers you suggested, we have also included references to the following papers: [3], [5], and [6]. These papers provide further insights into the state-of-the-art indoor positioning techniques and will help readers to better understand the advancements and challenges in this field. Thank you again for your valuable input.

 

Point 5: Is the proposed scheme compared with others?

Response 5: In section 4.5 "Comparison of Base Network Models", the navigation accuracy of the algorithm proposed in this paper is compared to that of the RoNIN [10] algorithm (using ResNet18) from the University of Washington. The comparison is conducted on the dataset VOIMU used in this paper, and the detailed results can be found in Table 3 of the paper.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thanks for addressing my comments.

Reviewer 2 Report

Thanks for the author's reply. I have no more questions