One definition of success in driving any sort of vehicle is getting where you're going without hitting stuff. Doing that involves the coordinated use of brakes and throttle and steering
The Daimler team has devised a way to distinguish between static and moving objects, and an object's direction and speed.
The Daimler team has devised a way to distinguish between static and moving objects, and an object's direction and speed.
- but only after the eyes and the brain have done their jobs. Are they up to the task?

Actually, no, says a group of engineers at Daimler headquarters in Germany. They've been working on something quite ambitious called 6D Vision technology. The company calls it "the basis for new assistance systems and a key step along the road to accident-free driving."

Also referred to as "spatial vision," it's said to ensure instant recognition of dangerous traffic situations. It's all about augmenting human vision and speeding up an understanding of what's in front of the vehicle, then reacting to it.

Our eyes are amazing tools, but they do have their limitations, and some eyes are built better than others.

Even the best eyes have a very narrow angle of sharp vision, about two degrees, with anything outside that scope more or less blurred. We compensate for this by keeping our eyes roaming back and forth across the scene ahead. But we can only do, on average, two such sweeps per second. So even the most attentive driver might miss a critical situation. The 6D-Vision system needs only about 0.2 seconds to detect potential collisions across the full field-of-view of its camera, which is near 180 degrees. And its image is crisp and clean, not fuzzy in parts like ours.

Collecting clues

Seeing is one thing, but the bigger challenge is understanding what we're seeing quickly enough to react. This might mean anticipating the actions of a complete stranger, perhaps a pedestrian or a bicyclist, maybe another vehicle, with very few clues that we've only picked up as blurred peripheral images.

It takes a very, very alert driver to continuously collect those clues and see what's going on ahead, and then a biggish brain to decide that an obstacle might be moving into our path.

The very best amongst us might be good enough to grasp this. But in many cases it may not be not good enough to react in time.

Enter 6D Vision, with its ability to recognize danger quickly and to distinguish between static and moving obstacles. Add to that Daimler's ability to supply the electronic gizmology needed to produce an automatic evasive reaction.

'Track before detect'

The principle here, developed by Daimler researchers Uwe Franke, Stefan Gehrig and Clemens Rabe, is that moving objects such as other vehicles, cyclists or pedestrians are detected and analyzed before they cross the driving path. Potential collision hazards are defined within a split second.

The system uses a stereo camera system to perceive the surroundings in 3D, similar to the way a human eye works. But it doesn't just see potential obstacles as static objects; it sees them in motion. By continuously analyzing the object's image sequence - frame by frame, almost pixel by pixel - the moving direction and the speed of the person or vehicle can be computed as well. This enables a prediction of where the obstacle will be in, say, half a second.

6D Vision is based on what Franke and crew call a "track before detect" algorithm. That is, a motion is determined before the image details are grouped into an object. This avoids the typical errors of other stereo methods that first form objects and then try to determine motion by tracking them.

This is where the Daimler system excels. By needing to detect only parts of the whole image that may be an errant cyclist, the system isn't fooled by partially hidden objects, those that may be very close together in space, and other difficult viewing conditions. The 6D Vision system can determine the collision risk for every single image point, independent of an object detection step.

The researchers have essentially mimicked basic human perception capability on a small, energy-efficient and low-cost microchip, a kind that's used millions of times a day in TVs and digital cameras.

By the way, they get the technology's name this way: calculation of the obstacle's position in three dimensions plus motion in an additional three dimensions gives us "6D Vision."

Doing the math

So let's finish by doing some math. How fast is fast? A kid running unexpectedly into the road, for example, is perceived by 6D Vision within 200 milliseconds. Even the most alert person takes twice as long to do that, and if distracted in some way, a further 500 milliseconds can be added. Then there's an additional moment of shock to create further delay until a reaction kicks in. We're talking one second plus.

In purely mathematical terms, one second at a speed of 30 mph equates to a vehicle covering a distance of around 45 feet. The computer works twice as fast as the driver and initiates safety measures after just 20 feet. In an emergency braking situation, the vehicle comes to a standstill more than a whole car length sooner.

Another bit of math explains Daimler's motivation. In 2010, in Germany alone, about 200,000 people were injured in inner-city traffic accidents and about 1,400 died. Half those deaths were bicyclists or pedestrians.

Since 6D Vision technology can contribute significantly to accident reduction, Daimler aims to make it available to other manufacturers as well. Other researchers are already working on it, too, including a team at Carnegie Mellon University in Pittsburgh, Pa., which has put all this techno wizardry into action on a stereo-camera helmet.

The Daimler folks working on this are among three teams of researchers nominated for the Deutscher Zukunftspreis, or German Future Prize, the country's prestigious award for technical and scientific innovations.

The first car with this technology will be the next Mercedes-Benz S-Class generation, which will be launched in 2013.

When will 6D Vision show up on trucks? So far no word, but it's a given that we'll get it too.

Here's a brief video explaining 6D Vision:



From the November 2011 issue of HDT.
0 Comments