Visual Detection
I was wondering if anyone had any ideas on visual recognition of object by robots, like humans do. I'm building a robotic arm that senses an object, such as a screw, and puts it in a bin. Any help would be great. I'm not just sure where to start on this. Thanks!
Asobu
Asobu
Comments
No robotic eye robot·has that capability.
So, start by deciding what world of objects you want to sense / bin. Then, characterize those objects by where they are expected to be found, where they are expected to be put, their sizes, shapes, weights, and all other parameters that the various objects might involve in a sense/put activity.
That might give you (and anyone you ask for help) an idea as to where to start in designing a robot to perform the task.
E.g., big screw, little screw?· bin close to screw, bin far away from screw? bin with cover, bin without cover?· One screw a second, one screw a minute? Brass screws in this bin, plastic screws in that bin? Phillips head screws here, Torx head screws there? Etc.?
PAR
www.societyofrobots.com/programming_computer_vision_tutorial.shtml
Or were you looking for something more specific?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
www.intel.com/technology/computing/opencv/
If you are familiar with C, you should be able to get the library running under your preferred system.
If you have no C experience, but would like to begin learning C & OpenCV, then I suggest starting with Ch & Ch-OpenCV, which you can download from SoftIntegration.
Anyway, in searching the web to get ideas on what OS/software/hardware for my vision-capable bot I came across RoboRealm.com. They have a modular Windows PC-based software package which is excelent for trying out different filters etc. It can be used for guiding a robot, either on-board or remotely. Of if you prefer to compile your own embeded code RoboRealm is equally valuable for determining which combination of filters to use - without having to know the maths.
At worst it's an excelent machine vision training tool. Do check it out !
PAR
OK, sarcasm aside, What degree of recog. do you really expect? Is it to identify a specific shape vs a library of shapes? Etc.
Let's look at the "screw recognition" case:
Is the screw loose in a bin of screws? If so, see "Phd" comment, and let me know when you accomplish that task, it would be worth millions. I would start with ultrasonic detection, then step it up to a CMUcam ... etc. What you are trying to accomplish is something people with multiple Phd's are trying to complete and have not achieved. This is not to say you can't. Someone will, maybe you, but you should really look into the history of vision systems, and learn about the problems people are trying to overcome.
MIT's robotic lab is a great place to start (was there this summer).
My best advice, "Baby Steps" start small, then go big.....experiment, look for an unconventional solution, innovate.
Good Luck.
any luck so far?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
An hour's contemplation is better than a year's adoration.
As a side note to moderators, why does the thread show posts today between posts from 2007?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chris Savage
Parallax Engineering
http://www.roadnarrowsrobotics.com/catalog/product_info.php?pName=robosight-basic
http://www.roadnarrowsrobotics.com/catalog/product_info.php?pName=robosight-standard&cName=product-lines-rnis
sorry but both sites are not working...
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
An hour's contemplation is better than a year's adoration.
http://www.roadnarrowsrobotics.com/catalog/index.php?cName=rnis-robosight
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔