I'm doing some research into what effect the choice of vocabulary has on the efficacy of SLAM in a given environment, and trying to train my own vocabularies to see if this yields different results for a project I'm involved with.
When researching online, it seems that most users are using the default vocabulary files that come with the software (either OpenVSLAM, or a different package), and I haven't seen much discussion that gives a white-box view as to how those vocabularies have been generated - except that lotso f images have been used and so users should expect superior results to generating their own. However, I would expect that this requires the environment in which SLAM is being used to represent the environment used for the vocabulary - indoor explaration of corridors in a building, or outdoor exploration of a city, shopping centre, etc.
It would help to have some information in the readme, so I know what it is I am comparing against, and what environment it was designed for. Some ideas include:
What sort of images (and if a specific available dataset, where this is located) were used to generate the vocabulary,
How many, or what subset, of this dataset was used to create the vocabulary,
At what resolutions, or if any other transformations were applied,
Noting the k-L parameters used,
Any other settings (e.g. parameters for ORB itself)
(Also helpful for reference) how long it took to create, to help judge the computational expense of creating a vocabulary
I'm doing some research into what effect the choice of vocabulary has on the efficacy of SLAM in a given environment, and trying to train my own vocabularies to see if this yields different results for a project I'm involved with.
When researching online, it seems that most users are using the default vocabulary files that come with the software (either OpenVSLAM, or a different package), and I haven't seen much discussion that gives a white-box view as to how those vocabularies have been generated - except that lotso f images have been used and so users should expect superior results to generating their own. However, I would expect that this requires the environment in which SLAM is being used to represent the environment used for the vocabulary - indoor explaration of corridors in a building, or outdoor exploration of a city, shopping centre, etc.
It would help to have some information in the readme, so I know what it is I am comparing against, and what environment it was designed for. Some ideas include:
Thanks!!