I am looking to create a vision based model that will identify lawn boundaries for an autonomous lawn mower project. This is just a personal project, I am not affiliated with any Uni / Company for this, nor do I intend to make it into a startup.
Among the various datasets I have found are grass and weed species identification, a very interesting audio dataset on the sounds of lawn mowers, a segmentation dataset with some images showing grass vs walking areas, etc…
With the exception of the audioset, none of them seem to have any video or image sequence data, which I believe can be helpful to the detection task.
I am wondering if I can accomplish some of this using synthesized data in a simulator? I have very basic experience with ROS’s gazebo, so the task of creating different grass and weeds, trees, shrubs, and different heights and various lighting conditions seems daunting to me.
Interestingly enough there’s an Xbox game for lawn mowing!!
My questions are:
Does anyone know of an existing lawn mower or grass mowing simulation or generated dataset that would be useful for this purpose?
Is there anyway the Xbox game can be useful? Maybe somehow extract some images/videos or maybe screen recordings?
How difficult is it for a novice sim designer to be able to achieve a useful simulation for a lawn mower? Where would one even start?