OCCUPANT COUNTING:Lower Mid-Infra Red Proposed

In California video was tried in the early 1990s and abandoned as insufficiently reliable. More recently efforts in Dallas TX under the name HOVER (High Occupancy Vehicle Enforcement and Review) to demonstrate the use of lights and video in the visible light frequencies ended up a fiasco. (See TRnl#19m Sept 97 p5, TRnl#35 Jan 99 p13, TRnl#36 Feb 99 p9) The equipment supplier CRS of the UK withdrew support charging the implementors and the evaluator were a “bunch of amateurs” and the official evaluation showed an ill-designed, poorly setup system with equipment and software not functioning properly. The various contending parties still say the lighting/video technologies have promise if only they can get their act together.

Noise, noise

The HUM team say Nah! No way. “Systemic noise,” they write.

Tinted glass, fog, heavy rain and dust, and the glare and glint of the sun and reflected light will attenuate or block the view into the vehicle. Second, at least in America, faces come in many different colors exacerbating the problem of visible light imaging inside a vehicle. Imagine the civil rights issue potential of a system that fails to image the black guys and gets their cars constantly ticketed and pulled over when they’re legit-HOV.

And the challenges of night photography of the interior of vehicles with acceptable exterior illumination are too great to make video practical, say the HUM team. The only reason so many attempts have been made with video, they say is that the equipment is readily available at low cost. It just won’t hack it however.

The HUM team looked over the whole electromagnetic spectrum. At the low end of gamma, X-rays and UV there are health risks at the power levels needed. At the high end of radio, as in radar sensing, resolution is poor. In distinguishing humans from the background infrared sensing stands out as the technology of choice since humans are a rather steady 36 to 37 deg C and the surroundings are almost always at a different temperature. Very little ‘noise.’

Georgia Tech worked with a low wave length IR camera and a xenon strobe illuminator (0.6um to 0.9um). (TRnl#18 Aug 97 p3) It is just too close to the visible range, the HUM guys say. The challenge is to beat the windshield glass which severely disrupts IR sensing just as it is designed to attenutate the heat of the sun penetrating the car’s interior beyond about 1um.

Trial & error

The HUM team tested a Mitsubishi IR imager (IR-700) in the 2 to 3um range. On a static car it produced excellent side images through side windows but terrible images through the windshield. But whereas video works at a frequency of 1000Hz and can produce good pictures of a speeding vehicle from a side view, the IR imager at 30Hz produced a sideview blur. No good.

A second IR imager from Sensors Unlimited (SU-320) using the reflective near-visible or LMIR wavelength 1 to 2um shows great promise, because it penetrates windshield glass and very clearly distinguishes flesh from all the other interior items in a vehicle. The team showed their pictures at ITSA and faces of all colors stand out starkly white from everything else. A polarizing filter is used on the camera is sufficient to exclude sunlight glare effects which still have a small effect. At night and in poor sunlight conditions a LMIR illuminator will be used, just outside visible light range, and unnoticable, and therefore much less distracting to drivers than visible lights needed for video at night.

The HUM team plans to work on improving the signal discrimination. And it will be developing a neural network to manage the intelligence needed to automatically identify and count separate faces. The work is getting some money from MNDOT’s HOVL program. (Contacts: I Pavlidis Honeywell 612 951 7338 pavlidis@htc.honeywell.com, R Sfarzo 612 625 0163 sfarzo@cs.umn.edu, Kevin Schwartz MNDOT HOVL 651 582 1131 kevin.schwartz@dot.state.mn.us)