Control Engineering

Paul Grayson

The Five Senses - Machine Vision
December 5, 2007

The first of the five or more senses available to automatic guided vehicles is machine vision.  In the world of industrial arms it is used to guide welding arcs along the seams that need to be joined or identify parts to be picked out of a bin or off an assemble line belt.  Machine vision has gone beyond that now and has left the assembly line to roam the shop floor and the highways.

Each of the vehicles in the 2007 DARPA Urban Challenge used machine vision as part of its navigation and collision avoidance system.  How much each relied on this feature varied from team to team but machine vision is something all of the vehicles had in common. 

Yesterday I spoke on the phone with Dr. Petko Dinev, Ph.D. - President, Chief Executive Officer of Imperx Inc. about his view of where machine vision is and where it is headed.  He has worked for both the Bulgarian Space Agency and NASA.  His connections throughout the world and his company’s wide customer base give him a unique perspective on what is going on.  For the past 20 years his customers have asked him to develop cameras, support hardware, and software to meet to meet their cutting edge needs. 

An interesting fact Dr. Dinev pointed out is that the higher the resolution of the camera, the lighter the lens can be.  In the case of tiny autonomous Unmanned Aerial Vehicles (UAVs), which fly under the radar yet are too small to see from the ground, weight is critical.  The high-powered precision lenses used for aerial reconnaissance are the heaviest item on the aircraft (think solid glass blocks).  Saving every bit of weight possible in these tiny aircraft is essential to their success.  High-resolution cameras have made a big difference.  

In digital cameras each frame is a digital file that software can analyze. This opens up all kinds of possibilities.  We are at a point in technology development where high-resolution digital cameras are finding their way into most every product.  Cell phones with cameras built in and toys that recognize their owners face are becoming more common.  This is just the start of every appliance eventually having a camera built into it for the things digital files and clever software can do.  Imagine how things would be different if your refrigrator could inventory its own contents and let you know what you are low or out of.  As communications bandwidth improves, bandwidth intensive applications are becoming practical.  Detecting the first traces of fire in California and dispatching fire crews while the blazes are still very tiny was one suggested use of this new capability.

Industrial digital cameras can cost anywhere from $1,000 to $56,000 depending on what you need them to do.  Part of their value comes from the things you can cut back on or leave out such using the less expensive lens and taking over some of the image processing tasks.

High mega-pixel intelligent cameras with some image processing being done in the camera itself are the new cogwheels for engineers and inventors to work with.  Companies like Sarnoff are making use of the advances in digital cameras to create visual navigation systems for intelligent vehicles.  They sponsored Team #51 on my list: Team Autonomous Solutions in the 2007 DARPA Urban Challenge and their vehicle "Ted" to demonstrate visual navigation products in action.

What could you build with a 16 megapixel progressive scan CCD camera with a resolution of 4872 x 3248 pixels, a variable frame rate from  3 fps to 29 fps, a selectable data depth 8-bit, 10-bit or 12-bit, video gain of  6-40db, 1024 levels, single or dual video output and an base camera link output interface?  While not exactly Gilbert Erector Set parts, these are the new building blocks from which to make the things we want or need.




Team #65 on my list is Team MIT.  Below is a picture of their 2007 DARPA Urban Challenge vehicle as it passed one of the photo-op corners in the 55 mile urban course.  As you can see at this prototype stage of driverless vehicle development it is packed with a variety of sensors.  Most people believe that these will eventually all be reduced to just a few key, high performing optical devices.  The massive amount of electronics on board required a roof top air conditioner in addition to the vehicles cabin air conditioning.  To be useful, the amount of electronics on board needs to be reduced to just the bare essentials in later versions.  Some other teams had already produced more compact and aerodynamic designs.  Their pictures should appear later in this column as we work through the stack of 2,500 pictures that AIM Team members Linda and Tom Graham brought back to share with us here.



GO ROBOTS !

Posted by Paul Grayson on December 5, 2007 | Comments (0)