SC3D Single Camera 3D

• What is SC3D?
• How SC3D Works
• How I Chose this Technology
• Why I Chose this Technology
• Resources

What is SC3D?

	In familiarizing myself with SC3D and preparing for my in-class presentation, I read the
	article summarized here:
	
Revolutionary breakthrough in VGR technology
Unlike traditional stereoscopic or laser-triangulation 2D systems which often use multiple 
cameras
Uses only one conventional CDD camera to gather orientation for all six degrees of movement: 
x, y, z, pitch, roll and yaw
Information from the camera is sent to a standard PC loaded with Braintech software
Braintech software, acting much like a human brain, delivers "location coordinates" to the
robot controller
The controller uses this information to guide the robot's end-effector and intercept each part
correctly for grasping or performing other robotic processes
Calibration of this system is fully automated
The entire system boasts tolerances of 0.1mm with a processing time of less than one second

VGR technology offers tremendous new advantages
Most important being greater flexibility for loosely fixtured parts
Robots equipped with sensory vision can now perform traditional, often dangerous jobs previously 
only suitable for human labor
Potential for enormous cost savings by eliminating or reducing custom designed fixturing devices

Without Vision, a traditional robot follows a pre-determined path, which has to be "trained" or 
"taught"
During routine operation, the robot plays back the memorized path at the specified speeds and 
accuracies
This approach works so long as the part remains in the same location and orientation in space with 
respect to the robotic arm
If this relation changes, the robot will simply miss the target or worse, crash into the part
The result is damage to the part, the robot and very costly "downtime" on the factory line

With Braintech 3D, the industrial robot can "see" the changes in the location and orientation of 
parts in 3D Cartesian space and react accordingly to adjust its path
Robots do not have to be trained to learn a specific path and there are less risks of a crash
Further advantages are the reduced cost of retooling, and readjusting the production lines due to 
design changes

How SC3D Works

	All of the information about SC3D on the Braintech website is presented	from a marketing 
	perspective. There are no real details of how full 3D information is derived from still
	camera images. It was interesting that during my presentation there was a lot of interest
	in SC3D and speculation	about how it works. A reasonable guess offered by several people
	was that multiple images from different angles are required. It is not clear from the 
	SC3D video that the camera is acquiring or processing multiple images. Furthermore,
	the processing takes less than a second. This suggests, to me at least, that the image
	processing is spatial and not spectral (frequency domain). I think Roger Matar's comment
	was the most insightful. He said the algorithm probably uses preprogrammed templates to
	match against distinct features on the part. Phil Thoren and I did an experiment where we
	looked at an object (drawn on a piece of paper) and noted changes in the 2D image due to
	changes in the position	of the camera. This approach wasn't terribly helpful so we made 
	some drawings of simple objects on the whiteboard. Being able to identify three points on an
	object would get you 2D rotation and translation easily. And depth would come from the size
	of a known feature. Of course there needs to be some mapping from pixel size (pixel area or
	number of pixels) to a useable z-coordinate. The final piece of the puzzle comes when one
	realizes that the object features have known dimension. When an object is tilted in 3D space,
	the "segmentable" features (in 2D space) will have smaller or greater pixel size depending on
	where the feature is relative to the known, fixed camera position. I'm sure there are calibration
	routines running so that the software always knows the 3D position of the robot arm. And, of 
	course,	the spatial relationship between object features is known. Therefore, one can imagine
	how full 3D information from a single camera is possible.

How I Chose this Technology

	I searched Google for new robotics technology.

	One of the top hits was:
	http://www.robotics-technology.com/tekcentre.html

	There I discovered this article:
	Braintech’s new "Vision-Guided Robotics" SC3D: A promise of revolution for the robotics industry 

	Local copy:
	robotics-technology_com - TEK-Centre.htm

Why I Chose this Technology

	Well, using a single camera to derive full 3D information for a part (object) sounded
	intriguing all by itself. I went to the Braintech website to investigate SC3D and was
	hooked by the application -- manipulating huge, heavy V8 cylinder heads. There were some
	beautiful, color PDF files and a cool video of a robot arm slinging massive cylinder heads
	around like hash browns. So SC3D combined two things that interest me -- computer vision
	and brawny, American V8 engines. As it turned out, my search for a technology to present 
	was brief indeed.

Resources

sc3d.pdf
engine_head_decking.pdf
engine_head_transfer.pdf
	
	Action video of SC3D in an automobile assembly plant -- notice the speed and 
	precision of the robotic arm -- it handles the heavy cylinder head as if it 
	were made of plastic!

sc3d.wmv (Windows Media Video)