October 2006

"Seeing in the Dark" — Safe Night Driving

By Joseph V. Borruso, Michael J. Flannagan, David R. McLellan and Joseph F. Ziomek

In an effort to reduce the number of deaths and serious injuries on Europe's roads, a team of carmakers, automotive suppliers and university researchers joined forces to develop future onboard night vision systems to help alert drivers to unexpected obstacles and improve driver visibility. Consider the following excerpt from the European Union's Information Society Program's June 2005 article "Driving improvements to night vision":

The 30 percent of road accidents that happen at night involve half of the people killed on the roads. Darkness is a major risk factor: while drivers travel just 28 percent of their miles at night, 55 percent of all motor fatalities occur after sunset. Ninety percent of a driver's reaction depends on vision, which is severely limited at night. Depth perception and color recognition are also compromised after sunset. Other dangers besides reduced visibility include fatigue, drowsiness, blurring of peripheral vision and impairment in judgment of distances and movements. The numbers behind these statistics do not reveal how many accidents occurred because of lack of visibility at night, but introducing an effective and easy-to-use system to enhance the driver's perception would help prevent accidents and ultimately reduce fatalities and injuries for drivers, cyclists and pedestrians. For example, a pedestrian wearing dark clothes is only visible at a distance of 100 feet to a driver using low beams. [Source: IST Results]

Our U.S. experience is similar to the European perspective on night driving, and these sobering conclusions are leading researchers to solutions that involve infrared (IR) cameras to augment the drivers ability to see, and displays — both heads-up (HUD) and head-down flat panels — to let the driver "see" what the sensor sees.

Unfortunately, any display not exactly superimposed on the driving scene is impractical. Driving is first a visual task. All decisions about accelerating, steering and braking come from what the driver sees and responds to. The driver can use a flat panel map display because he chooses to look away from the driving scene when he's decided that it's safe to do so. To ask him to look away at a moment of danger is not sensible. Even a HUD at the base of the windshield requires the driver to look away from driving scene and refocus to the front of the car to read the display. And how does the driver know when to look down?

For a few years, Cadillac sold a HUD displaying the output of a passive IR camera. But even when the driver watched the display, a deer had to be within the narrow field of view of the camera before it appeared on the screen, probably too late for the driver to take evasive action. After the novelty wore off, sales went south. Users were giving us their sense of the system's usefulness.

Critiquing the approaches already tried doesn't make the problem go away, but it can help us learn from our past attempts — even if they were less than successful.

Let's back up, restate the problem, and ask ourselves what we want the driver to see, and then try to connect the dots in a better fashion.

Night driving involves background lighting conditions that range from pitch-black night, to wet asphalt where, even with the best passive high beams, visibility is almost zero, to the Las Vegas strip, where there are so many lights that it's hard to discriminate what's important. Passive lighting's brightness and elevation are limited because it blinds drivers ahead in their rearview mirrors and the oncoming drivers directly. Lighting patterns have been developed over the years to be as good as they can be, given that they are fixed with respect to the car.

What do we want to see that we're not seeing now? Pedestrians dressed in black, deer darting across the road, poorly or unlit road signs, the road turning ahead, an unlit rural crossroad where missing the corner could mean dropping into a drainage ditch, to name a few. We also want to illuminate other cars, particularly at intersections, that could be on a collision path.

With sensors we have the ability to see over a range of wavelengths much broader than that of the human eye. For example, we can identify live objects from their heat signature.

The unanswered question is how to communicate this broad spectrum of information to the driver, without overloading him with too much extraneous information. In aircraft, pilots are already using HUDs that superimpose needed information on the flight horizon. We could do the same with the automobile, but why bother? Why not super illuminate the objects of interest on the roads path ahead? If we knew the path ahead in 3-D, and where leading and oncoming cars were on that path, could we design a light system that would selectively illuminate what the driver needed to see?

The Tucker Torpedo of 1948 had a third, center-mounted, steerable headlamp. Unfortunately, it illuminated into the turn only after the driver turned the steering wheel. The driver turns the steering wheel as the last act in the decision making process when driving the car. What's really needed is a light that anticipates the drivers need and looks into the turn in the seconds before the driver turns the wheel. We also need to be able to put more light selectively on the road surface when it exhibits poor reflectivity, as with wet asphalt.

Could we imagine a very bright light that scanned the road space ahead, moved by a mirror that is driven electronically, as the mirrors in a digital light processing (DLP) television are moved? Such a mirror would turn the light down or off when it is momentarily aiming at the car ahead or at an oncoming car. When objects are detected that represent danger, the light would illuminate them at a higher level than its surroundings, alerting the driver to potential danger. The light would follow the danger and could even warn the driver of its projected path. The driver might receive audible alerts as well. And in the most sophisticated systems, the car could take action on its own, braking and even steering out of danger.

The key to an effective and safe night driving system is developing the ability to electronically move a very bright light, and the logic that tells the light where to scan and at what brightness. Objects, signs or other vehicles, identified by our car-mounted sensors, are already referenced to car position. The road itself can sometimes be identified with the sensors, but too often its location is as indeterminate to the sensors as it is to the human eye. The crucial element to finding the road ahead under all conditions of darkness is precision GPS and a precision road map. Given the elevation changes of the real world, the precision map will need to be a 3-D road map, which is now becoming available.

Joseph V. Borruso is a former President and CEO of Hella North America; Michael J. Flannagan is a research associate professor in the University of Michigan'Transportation Research Institute's Human Factors Division; David R. McLellan is a consultant and retired director of Corvette Engineering; Joseph F. Ziomek is director emeritus of the Convergence Transportation Electronics Association. Comments may be submitted to todaysengineer@ieee.org.