Real-Time Detection and Spatial Localization of Insulators for UAV Inspection Based on Binocular Stereo Vision

Round 1

Reviewer 1 Report

The article prepared very good quality. It will be intereste to the scientist who has interest field of UAV and aerial imagery. 

The manuscript could be accept in present form.

Author Response

We'd like to thank you for your careful readings and positive feedback.

Reviewer 2 Report

The authors describe a UAV-based (DJI M600-pro) power transmission line inspection system capable of detecting and positioning insulators, by using depth information acquired from two onboard cameras, real-time processing of the collected data and GPS-assisted photogrammetric techniques for the insulator's location. The results are well presented, together with a thorough description of the theoretical background (with many useful references for the interested reader. I would just suggest to add a reference for the Hough transform techniques cited in Line 331). English style is good, with very few typos (Line 208, "Pablo et al." should read "Arbelàez et al."; Line 389, "hVAU" should read "hUAV"). The discussion and the conclusions are technically sound, pointing out interesting developments of the technique, such as adding thermal cameras for insulator temperature detection and fault analysis.

In conclusion, I would recommend publication of the paper in its present form, with the minor corrections suggested.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This paper presents a real application for image segmentation with the aim of visual recognition of specific items on an image. Authors trust an image saliency technique supported by RGB-D and the final approach completed with GPS data since it operates outdoors, looking for insulators of power towers.

The outlook of this work is sound, well written and generally well detailed and explain.

I suggest authors work on the following comments before going further steps.

sec 2.1 Dataset with capital letter

sec 2.1 What is the dimension of the dataset in terms of physical extension, length of wire, or similar? Can be this dataset be publicly available? At least visual examples are required.

sec 3.1: please revise titles of subsections to be initiated with capital letters.

sec 3.1: please be here more specific about the real challenges mentioned for depth estimation, ie with stereo pairs.

line 208: Pablo is the main name of the author, not his surname.

sec 3.1.2: success of SURF is acknowledged, but please confirm why selected here. Any other results with other feature detector?

Results:

How algorithms [14][38][16] have been tested? Are they available for benchmarking?

Explain better what the average processing time is.

All in all, I miss a better-argued reason for this approach to: 1) operate real time. Is not possible to post-process data and identify insulators? 2) be a real advantage against deep learning techniques for object detection/classifiers.

Author Response

Please see the attachment.

Author Response File: Author Response.docx