Texture and terrain recognition
Sudzzz from United States  [17 posts]
17 year
Hi there,
I am currently working on a project that is somewhat similar to the DARPA urban challenge, except that I'm trying to build an autonomous radio controlled car which is capable of driving on any terrain. I've been using RoboRealm to test out the different filters that it offers to check which ones apply to my project in the best possible way. I'm trying to implement a texture recognition technique where a small sample of the drivable region is taken and is tracked over the entire image(frame by frame texture recognition) to see whether the terrain ahead is drivable. Furthermore, I'm thinking of setting up filters to carefully pick out blobs that can fall under the drivable region and eliminate the ones that aren't drivable. I was also thinking of using Cog for finding an approximate direction of drivable terrain. It would be more reasonable to apply Cog on a horizontal blob array ( How do I do this?). I was just wondering what would be the best features to use so as to accomplish this task. I've fiddled around with floor finders and color filters to possibly track certain terrain and distinguish them from others, but would like to get an expert advice on this topic.
Anonymous 17 year
The blob array output are samiler to COG output.

Steven, is that possible to get Blob orientation variable while I am using the blob replace module? Thanks

I have tried wall finder module to detect the road, lane and traffices.
I find it works fine in the cloudy day. Sunny day will blow out all the road or terrain details. Maybe it works indoor.
Sudzzz from United States  [17 posts] 17 year
Yeah, I've used the wall finder and the floor finder module to detect the road. However, the light intensities vary every so often if the sun comes out and its difficult to automatically calibrate it for the camera.
I'm trying to use the surface plot module to possibly detect rough terrain from the smooth terrain (mean module is done before this to make the surface plot smoother). However, I'm also having trouble using this plot to make a 2D representation of drivable as opposed to non drivable regions.
Anonymous 17 year

If you find you need the horizontal (and vertical) cog let us know and we can quickly add that in. It is nice for display purposes but I'm not sure in what capacity you'd want to use it for steering the robot.

As for road detection have a read of


which will probably give you some ideas. Basically the trick to removing shadows is to use some other quality other than intensity like color or saturation levels.

In terms of texture, what you can do is to run an edge detector and then add a population threshold. Changing the pop threshold module will add/remove dense texture areas and define it as a connected white blob. This helps to define drivable areas of the image and can reflex the smoothness of the images. For coarser textures scale the image prior to edge detection.

We're also working on the images seen in the above url and will hopefully post a more intensive investigation of how one could improve road following (we doubt we can fully solve it for all situations but it should be enough for basic usage).

Anonymous 17 year
Note that the MEP_COORDINATES array was just added. See the Replace Blob page for more detail on the variable.

Note that this is not really a blobs orientation but its minimal enclosed polygon. It can be used as a good approximation of its orientation .. or at least a repeatable one.

You'll have to download RR again to get that new feature.


This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index