No need to wait for AVs: Vehicle cameras can already detect bicyclists, pedestrians - Mobility Lab

Another day on Pershing Drive in Arlington, Va., and another near miss for Mobility Momma on her bike.

Yes, “Motor City Maniac Dear Son,” age 14, and “Blown-the-Coup Bike-Only College Kid,” 18, did warn me to stick to roadways with bike lanes or, better yet, stay on the bike path.

I thought the vibe would be better early on a Sunday morning. And it was: there were fewer cars and the new four-way stop at Irving and Pershing did seem to slow the traffic. Nevertheless, ignoring my own maternal admonitions of the laws of greater gross tonnage, I lamented that many of the drivers just didn’t seem to see me.

Dear Son: “Or maybe the drivers are like us and they cannot predict what you are going to do or – by the way – what those putt-putt autonomous vehicles won’t be able to do either.“

Mobility Momma: “I admit that worrying about whether an AV can see a cyclist on the road may seem a little esoteric. But the challenge is indeed facing us now.”

Or maybe not.

The failure of drivers to see bicyclists and pedestrians is the “reason we are seeing a disturbing increase in bicyclist and pedestrian fatalities,“ according to Uri Tamir, senior director of Strategic Initiatives at Intel’s Mobileye.

He added, “817 bicyclists were killed in crashes with motor vehicles in 2015, a 13 percent increase from 2014 and the highest number of bicyclist deaths since 1995. 5,376 pedestrian deaths occurred in 2015, an increase of 9 percent from 2014 and the highest level since 1996.”

Mobility Momma interviewed Tamir about how camera technology can help crack this problem now and how it is different from other technologies AVs might use to detect bicyclists. I first met him at Ft. Meyer, the military base in Arlington, where we were reviewing the feasibility of an automated shuttle service connecting the base, Arlington Cemetery, and the Pentagon.

Kelley Coyner: What is Mobileye doing to help solve one of the toughest hurdles for rolling out Advanced Driver Assistance Systems (ADAS) in AVs – detecting bicycles and pedestrians? How long will it take to solve that riddle?

Uri Tamir

Uri Tamir: Mobileye has been focused on integrating cyclist-detection capabilities for our automakers’ customers since 2010. We introduced pedestrian and cyclist collision warnings to our aftermarket product in 2012. Bicycle recognition is not necessarily a tougher problem for cameras than other vulnerable road users, such as pedestrians.

Do you agree that bicycle detection is the hardest problem for autonomous vehicles?

Bicycle and pedestrian detection is usually more limited for radar sensors, mostly due to their low radar reflection (insufficient Doppler frequency and spatial resolution).

But pedestrians, like bicycles, do present specific challenges for drivers. Cluttered scenes, like urban streets and crowded bus stops, pose an attention challenge to drivers and to certain driver-assistance systems.

Occluded objects are hard to classify for all sensors, but for sensors that are not cameras, distinguishing between objects that are very close to each other, especially if one of the objects is made of metal, is a challenge. This means that, for a system that is not camera-based, distinguishing a pedestrian in close proximity to a vehicle is difficult. Since the camera looks for patterns, it can fairly easily detect such objects in cluttered scenes or an object that is partially occluded – for example, a cyclist figure is identified by a camera as independent of the environment.

Click here to read the full article: