Ceiling-View Semi-Direct Monocular Visual Odometry with Planar Constraint

Round 1

Reviewer 1 Report

This paper proposes a ceiling-view semi-direct monocular vision odometry system based on planar constraints and tries to improve the trajectory accuracy of existing VO systems in indoor ceiling scenes. Besides, the ceiling plane is exploited as an additional constraint, and the photometric error and geometric constraint are both optimized in a sliding window to improve the trajectory accuracy. The authors also provide a dataset for the ceiling-view visual odometry. But the writing can further be improved and the experiments are not sufficient. The detailed comments are as follows,

1.      This paper is poorly written and hard to understand. For example,

line 34-35, “which is an important means for robot state estimation”.

Line 57-58, “the above mentioned approaches usually decrease”, this sentence seems to end without finishing?

Line 68, “…, although it is…”, it is very strange to appear “although” here.

Line 70-71, “However, even if there is only the ceiling plane, the coplanar structured information can still be extracted to reduce the positioning error.” I'm confused about what this statement is trying to express.

Line 80, “dataset,.”.

Besides, there are also some mathematical formulas expressed irregularly, such as Pi(XiYiZi)   in line 270.

I hope the authors could revise the manuscript carefully. It is unacceptable that the current version be published.

2.      I am wonder how do you extract the coplanar points and coplanar lines and establish correspondences? I think the author should elaborate on this, as it is the key to this method.

3.      How do you select the key frame? What the criteria to set a key frame? The author should explain in detail how the key frame are selected.

Author Response

We feel great thanks to your professional review work on our article. According to your valuable suggestions, we have not only improved the writing but also seriously supplemented the experiments. The detailed modifications are listed in the attachment.

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The reviewed article deals with improving the accuracy of the SLAM algorithm by looking at the ceiling with a monocular camera. The idea itself is not new, it has been appearing in the literature for at least 17 years ago and it is gradually being added by the detection of ceiling boundaries, restrictions resulting from the relative position of marks, etc. This article brings an innovation in the form of evaluating whether the reference points used are placed in the ceiling plane or not. Therefore, I consider the given article to be relevant to the subject of CV SLAM (Ceiling View SLAM).
The method of restricting the mark points to the ceiling plane brings a reduction of the effective value of localization error in the range of several percent compared to previous published CV SLAM methods, which is the main contribution of the research described in this article. In my opinion, a further significant improvement in localization accuracy would be possible by applying not a monocular but a stereoscopic camera, which would, however, significantly increase the requirements on the computing power of the entire SLAM system.
I consider the conclusions resulting from the presented research to be correct, the authors declare the limitations of the applicability of the proposed Struct-PL-SVO method only for environments with a relatively low and single-plane ceiling with regular shapes, while declaring the deterioration of parameters in an environment with a complex multi-plane ceiling structure. The references suitably describe the topic covered in the article, but I recommend the authors to slightly expand the number of references directly related to the topic of CV-SLAM. I recommend publishing the article after minor revision. 

 

1. Please explain the abbreviations in text when they are first time used (for example AR, VR - line 31).

2. In Fig. 11, please add the second scale on x-axis with the distance covered, or specify the speed of the wheeled robot during the tests.

Author Response

We sincerely appreciate the valuable comments. These comments are all valuable and helpful for improving our article. The detailed modifications are listed in the attachment.

Please see the attachment.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The revision has addressed the comments well.