Moving Car Recognition and Removal for 3D Urban Modelling Using Oblique Images

Round 1

Reviewer 1 Report

Why is the presented method very important ? If 3D Urban Modelling is very import for smart cities why they don't stop the traffic for a few minutes, make the very important picture and resume the traffic after ? 

Why only the moving cars have to be removed ? For clarity and efficiency why not remove also the static cars ? 

The pictures that are presented as results clearly show that some of the cars are parked, how can you be sure that a car is stationary or moving ? How do you define (time related) short or long stays ? 

The related work section presents 2 subsections that are related to the topic but I would have expected to see something that also contains research information about : moving object removal, applicability of the solution in the presented context. 

The text formatting seems a bit strange, it looks more like a template problem. The images are looking okey but the text (in some portions) has a very big left indent. This needs to be fixed.

Since you mention the device and resolution use for taking the shots, why do you talk about sample like they are from a third party ? I suggest to make things clear : you made the pictures with that drone or your data set is received from a third party ? 

This is a very good experiment and results, but I am wondering why you don't just add objects with another vehicles types for testing purposes .

The references are decent, but in order to improve the paper I would recommend to replace some of them with newer materials. I recommend adding references for : triangular mesh

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript is focused on a significant problem when performing 3D digitisation of urban areas where moving objects such as cars are inevitable and need to be removed in a "clean mesh" process for further usage . The paper describes the approach in detail and well document results are presented. I would recommend the authors to extend the literature review (related work) and include some recent works that also related with the mesh segmentation problem and car detection using similar or relevant approaches (multispectral data, machine learning, etc), . Such examples are

https://arxiv.org/abs/2106.01178

https://www.sciencedirect.com/science/article/abs/pii/S1296207421000650 

https://www.proquest.com/openview/dbe17ba9932f87ac14f3d16d42c4d218/1?pq-origsite=gscholar&cbl=2032404

 

Again, commercial software such as Agisoft Metashape do offer methods within a different framework where vertices are graded based on confidence. This confidence metric is IMHO also related to where the point in 3D space has changed its position. Maybe authors would like to research on this in their future work.

All in all, it is an interesting paper focused on the automation of removing movable objects (cars) from urban areas being 3D reconstructed using the SFM/MVS method.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear author, thank you reading and applying my feedback. Congrats for the great work.