Object recognition
Community Forums/General Help/Object recognition
| ||
| I have gotten my camera successfully sending blitz max a live video feed which is basically just constantly updating a pixmap/image. I can get basic recognition like if I put brightly colored dots on objects the camera can easily track them. But at some point I would like to learn how to do object tracking and recognition that doesnt involve putting flourescent stickers on everything that I need tracked. Eventually I plan on mounting my laptop to a robot I am building and using the camera as the "eye" and the computer as the middle man doing most of the image processing and calculating. That may be a little farther in the future though! Anyway, does anyone around here have any experience with object recognition/tracking. If so what is the best method to use and do you know of any good tutorials on the subject? Even a good conceptual idea of how to go about object recognition would be super helpful! |
| ||
| For "general" computer vision tasks (haha, no such thing) the place to start is with OpenCV (+doc). However, this will be slow. If it's at all possible to narrow down what you want to do beyond "stuff", you'll do much better to find a library optimised for the sort of tracking you want to perform, on what sort of objects. You should also consider how bad it really is to have to attach stickers to things: to date tracking markers remain far and away the best (both performance and reliability) way to handle this sort of thing, in which case you might want to investigate a full-on augmented reality toolset instead. |
| ||
| Admittedly tracking markers are not that bad and are fairly quick to track. Thanks for the link, I haven't seen that before but I will look into it! For now I may do some simple "texture" recognition algorithm to determine what type of ground is safe to drive over. I really wanted this robot to be able to drive around anywhere with a decent and consistent ground texture without hitting obstacles and it seems a simpler texture recognition might be more reliable than a 3d polygon interpretation of the scene based on multiple view angles would be. Thanks for the help! |
   |