Driving Speed Estimation and Trapped Drivers’ Detection inside Tunnels Using Distributed MIMO Bluetooth Devices
, Anestis Kastellos 2, George Kokkonis 3, Theodosios Gkamas 2 and Christos Pikridas 1Round 1
Reviewer 1 Report
The authors presented Driving Speed Estimation and Trapped Drivers’ Detection Inside Tunnels Using Distributed MIMO Bluetooth Devices. I have the following comments for this work:
- Please make the abstract a single paragraph.
- The conclusion section is too lengthy. It should be focused on your contributions and results, repeating the work done as part of the implementation is not a good approach.
- Figure 3 shows very basic components in the fabrication of the prototype. Is the prototype up to the industrial requirements to be used in real-time?
- How frequently does the RSSI data need to be re-collected? RSSI data is time, location, and environment-sensitive. Once you build a database, it needs to be frequently updated, a drawback of such approaches. Also is there any pre-or post-processing of the database involved in this work? Please refer to https://doi.org/10.3390/electronics8020195 for more details on it.
- Please avoid the use of long sentences, check the grammar of the manuscript, and remove typos.
Author Response
Comment 1: Please make the abstract a single paragraph.
Response: Abstract occupies now one paragraph, after the reviewer's suggestion.
Comment 2: The conclusion section is too lengthy. It should be focused on your contributions and results, repeating the work done as part of the implementation is not a good approach.
Response: The conclusion is shortened and focused on our contributions and results after the reviewer's suggestion.
Comment 3: Figure 3 shows very basic components in the fabrication of the prototype. Is the prototype up to the industrial requirements to be used in real-time?
Response: The previously presented end-node BL-detector prototype is a Technology Readiness Level 9 (TRL 9) end node device that has been successfully tested and validated in its appliance environment. The BL-detector equipment ensures conformity with the EU-wide requirements for operational types of equipment inside tunnels and the local government laws and regulations for wireless transmissions, making it an Industry ready real-time detection implementation. Therefore, an appropriate paragraph has been added at the end of Figure 3.
Comment 4: How frequently does the RSSI data need to be re-collected? RSSI data is time, location, and environment-sensitive. Once you build a database, it needs to be frequently updated, a drawback of such approaches. Also is there any pre-or post-processing of the database involved in this work? Please refer to https://doi.org/10.3390/electronics8020195 for more details on it.
Response: Appropriate response and citation has been added into new paragraphs (last paragraphs at the end of section 3.1, lines: 280 and 293)
RSSI data are collected in real-time when Bluetooth transponders are detected. The real-time data acquired passes through a pre-processing filtering step and a post-processing training step to achieve speed predictions, similar to [citation added]. Each Bluetooth detector can acquire more than one RSSI value of each detected Bluetooth MAC. The number of RSSI values/Bluetooth/MAC depends on the users’/vehicles' moving speed. The BLE devices, which detect only close to the tunnel exits, require no post-processing or estimation, apart from the appliance of the Friis equation and a pre-processing data cleansing step including impairments compensation. This pre-processing step interval is 500ms up to 1min. It is close to the interval for the Application service to reload and renew its dashboards. The same pre-processing step applies for the Bluetooth class 1 device, followed by a classifier speed prediction. The post-processing training interval for the estimators training is periodically instantiated in each node every month. During this time, each detector node retrieves RSSI data as well as TMS-IL output speed data from the MongoDB service. Then, it passes the RSSI through a pre-processing filtering step before feeding them as input data to the classifier. During the testing phase, the APC values per class are calculated. The newly trained classifier is used if a max(APC) >0.7 is achieved. If not, it is discarded, and a new training process is scheduled that shall use the aggregated data of the two intervals for its next post-processing step.
Comment 5: Please avoid the use of long sentences, check the grammar of the manuscript, and remove typos.
Response: Comment has been taken into account.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
An engineering-level system is presented for detection of people and vehicles inside tunnels based on BLE and Bluetooth sensing. Both the device design and the processing steps are explained. The system is an interesting. complement/replacement to current aid detection systems in tunnels, and it is shown too have good performance in comparison with the state-of-the-art technology.
The paper contains a few scattered typos and English grammar errors that should be corrected, including a repeated paragraph in Section 4.2. From a technical point of view, the description of the classification process should be complemented by indicating the size of the training size (number of samples). Also, some discussion on the variability of APC among the tested classifier should be added, in order to explain better the choices and the reasons for such disparity.
Author Response
An engineering-level system is presented for detection of people and vehicles inside tunnels based on BLE and Bluetooth sensing. Both the device design and the processing steps are explained. The system is an interesting. complement/replacement to current aid detection systems in tunnels, and it is shown too have good performance in comparison with the state-of-the-art technology.
Comment 1: The paper contains a few scattered typos and English grammar errors that should be corrected
Response: The typos have been corrected
Comment 2: including a repeated paragraph in Section 4.2.
Response: The repeated paragraph has been deleted
Comment 3: From a technical point of view, the description of the classification process should be complemented by indicating the size of the training size (number of samples).
Response: The size of the dataset and test size are presented in the paragraph at line: 220
Comment 4: Also, some discussion on the variability of APC among the tested classifiers should be added, in order to explain better the choices and the reasons for such disparity.
Response: Variability of the APC is because the amount of data for the medium-speed classes is larger compared to low or high-speed classes (To support this claim also EU regulations forbid low and high-speed vehicle movements inside tunnels). As a result, the classifiers cannot acquire sufficient data to be trained correctly due to the lack of RSSI and TMS-IL measurements. The authors set the better evaluation and training of their proposed classifiers for low and high-speed classes as future work. – Paragraph added at line 472, end of results summary section.
Additionally, an extra paragraph and Table have been added in section 3.1 (line 253 and Table 2), describing the post-processing optimization step of the classifiers' parameters.
Reviewer 3 Report
A very interesting paper with real practical application and experimental verification in the field environment. Before publish the authors should read one more time their paper can correct very minor English issues like in row 313 "and the can be no other scans from" perhaps it should be "and they can be no other scans from" also they may want to consider publishing it in a higher impact factor journal.
Author Response
Comment: A very interesting paper with real practical application and experimental verification in the field environment. Before publish the authors should read one more time their paper can correct very minor English issues like in row 313 "and the can be no other scans from" perhaps it should be "and they can be no other scans from" also they may want to consider publishing it in a higher impact factor journal.
Response: The reviewer's comments have been taken into consideration. Minor English issues have been carefully corrected.
Round 2
Reviewer 1 Report
The authors have well-addressed all my queries. I have no hesitation in accepting the manuscript to be published in its current form.
