Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!


Forgot your password?
Google AI Transportation

Google Research Leads To Automated Real-Time Pedestrian Detection 57

An anonymous reader writes with a link to a story about one of the unexciting but vital bits of technology that will need to be even further developed as autonomous cars' presence grows: making sure that those cars don't hit people. Google researchers have recently presented findings about a method that tops previous ones for real-time pedestrian detection using neural nets "that is both extremely fast and extremely accurate." From the article: There are other approaches that provide a real-time solution on the GPU but in doing so, have not achieved accuracy targets (in this real-time approach there was a miss rate of 42% on the Caltech pedestrian detection benchmark). Another approach called the VeryFast method can run at 100 frames per second (compared to the Google team's 15) but the miss rate is even greater. Others that emphasize accuracy, even with GPU acceleration, are up to 195 times slower.
This discussion has been archived. No new comments can be posted.

Google Research Leads To Automated Real-Time Pedestrian Detection

Comments Filter:
  • Idea 1: the fact that accounting for the eventuality of not hitting pedestrians (or any other being/thing) is "one of the unexciting but vital bits of technology" when talking about an autonomous car provides a quite accurate summary about what a big proportion of AI-focused approaches are about. Lots buzz (= exciting technological break-troughs) and not actually-working results (= unexciting technical bits avoiding the big idea to work at all). And this is not just what the OP thinks; Google has been testi
    • by Gravis Zero ( 934156 ) on Saturday August 15, 2015 @06:05AM (#50321333)

      it seems that the new much-more-accurate algorithm still misses 30% of cases. For me, hurting (even killing) 3 out of 10 pedestrians still sounds quite bad.

      missing 30% isn't killing 3 out of 10 people, it's killing 7 out of 10 people which is a solid 70 points or 210 points if you are drifting. #Carmageddon

    • by WoLpH ( 699064 )

      Well, the question here is. Given that 26.2% at 15 fps, does that make the probability of a detection within a second (1-.262)^15 = .010491689? So less than one percent?

      Or is it far larger because the results are not actually independent?

      • I took the information from the paper itself, more specifically from the following part:
        "For example, when training on the KITTI pedestrian dataset [18], the best known average miss rate is 61.2%, whereas when training on INRIA [10], the average miss rate is 50.2% [6]. Both miss rates are much higher than 31.1% of our method"

        As I understand it, a miss rate of 31.1% on a given dataset means that 31.1% of the tested attempts failed.
      • PS: I am not sure about where your reasoning is coming from. What 26.2% at 15 fps means is that 26.2% of the occurrences of the target behaviour (= not necessarily matching a given frame) succeeded. You cannot extrapolate this information to different conditions, like higher/lower number of frames per second.
    • For me, hurting (even killing) 3 out of 10 pedestrians still sounds quite bad.

      Unless we know what the video feed is we can't make that statement. Are these pedestrians crossing the road or on the sidewalk? If the algorithm is missing 3 out of 10 sidewalk pedestrains that's much less serious than 3 out of 10 crossing the road. I suspect the idea behind the visual search is to identify people who could potentially cross the road so the car can slow down in anticipation. People actually on the road, in front of the car, can be spotted in other ways using other sensors.

      • When analysing the reliability of a given approach, you have to consider the worst scenario conditions. In this case, not being able to adequately recognise a pedestrian might be irrelevant or a tremendous problem. It is impossible to know where people or objects would be located with respect to the car and thus you cannot count on "perhaps I fail to recognise a person who is far away enough" (and another algorithm will take care of this determination; the recognition algorithm just has to worry about havin
    • Re: (Score:2, Interesting)

      by Anonymous Coward

      But why are they implying that something is almost done, when quite a few basic problems haven't still been tackled?

      This isn't about not hitting pedestrians in the roadway. This is about categorizing objects outside the roadway so that you know if they are pedestrians, who may enter the roadway at any time, or stationary objects, which may be presumed to stay stationary. Once categorized as a pedestrian, additional algorithms can be used to guess whether the pedestrians are going to enter the roadway and take defensive driving precautions to avoid hitting them.

      Note that even if this algorithm were wrong 100% of the tim

      • Your first paragraph explains what I have said in some other comments: this is a (pedestrian-)recognition algorithm and thus it has only to take care of performing proper recognitions (not what the proposed figures indicate). Properly recognising a pedestrian is the pre-step to perform many further actions (e.g., calling the whattodowithhumans() method); if a so basic feature fails, lots of further issues would also fail. In your own words "Once categorized as a pedestrian", represents the starting point fo
  • by SuricouRaven ( 1897204 ) on Saturday August 15, 2015 @06:41AM (#50321405)

    It just has to be equal to a human driver - and human drivers are not that good.

    • by pellik ( 193063 )
      No, it has to be way better then a human driver. There is a completely different scale of liability for a self driving car then for a human. The blame will hit much harder when someone gets hurt.
      • This notion that human drivers aren't that good needs to die. How many rides occur each day? How many pedestrians are hit? Yeah, a lot and very few. Computers have a very high bar to reach just to be on par with humans.


    • It just has to be equal to a human driver - and human drivers are not that good.

      In a completely rational world, perhaps. We don't live in a rational world. We live in a world where unusual accidents are governed by media hysteria and lawyers.

      And what happens with liability for such an accident with an autonomous car? Who is responsible? The driver? The manufacturer? The individual programmers who created the recognition and behavioral subroutines?

      Here's the reality -- early adopters of autonomous cars are probably going to be wealthy folks, because like any new technology the

      • by dougmc ( 70836 )

        There's another confounding factor to this ... every autonomous car collision will be documented in exquisite detail, but in a format that few are familiar with.

        So ... if the logs say that the car was at fault, people will use that to crucify those responsible for the car. And if the logs say that the pedestrian was at fault ... people will say that the logs were altered, incomplete, etc. and use those claims (accurate or not) to crucify those responsible for the car. And if something went wrong and there

  • One of the recent models of Mazda I drove (I'm pretty sure all manufacturers have that, Ford at the very least) had "active city stop" feature, active at speeds up to, 30km/h, if I remember correctly.

    Car would emergency break ON ITS OWN if it would spot a pedestrian.
    To my knowledge, they use some "radar like" technology for it.
    I guess it's not far sighted enough for a self driving car.

  • They will just follow some simple trial-and-error hit-or-miss approach. No harm done.
  • Q: Why does Google blur out pedestrians' faces in Street View?
    A: So self-driving cars won't develop an attachment to them.

    Q: Why does Google blur out license plates?
    A: To protect the identity of self-driving cars that mow 'em down for sport and points.

    Q: Why then is Google developing a 'real-time pedestrian detection' system?
    A: To improve scoring and help populate their Deathrace 2015 leaderboard.

    • Q: Why does Google blur out pedestrians' faces in Street View?
      A: So self-driving cars won't develop an attachment to them.

      It is indisputable that Google blurs faces in Street View, and the same company is also developing self-driving cars. Though different teams are assigned these projects and Street View images are not used by self-driving cars, the fact that Google is responsible for both is mentally noted, providing enough connection to lay a comedic foundation. Such a foundation is tenuous however and successful delivery of a joke requires follow-through that is quick and emotionally jarring.

      The follow-though is accompli

  • Who is legally responsible if an automatic car does hit someone?

  • The car being tested didn't have the "pedestrian detection" option so the car hesitated, then plowed into the journalists recording the self-parking event!


  • There'll eventually be automatic functions to stop a car when a pedestrian is detected in the car's path and to stop the car or pull it over if there is a collision. 1. Step in front of self-driving car on lonely road: car stops 2. Jump on hood of self-driving car: car is immobilized 3. ??? 4. Profit!!
  • Automated Real-Time Pedestrian Detection?

    Simply add face recognition to the muffler, problem solved. " Yep , that was a pedestrian -- and we even know who."

    But since it's Google, they'll probably do something higher-tech like measuring the reaction of the shock absorbers. And unlike their WiFi scanning, they've got two chances to get it right!

    What? Why are you looking at me funny?

    1: "My dog here has fleas and I'd like to kill them."
    2: Opens the door and tosses the dog into the roaring fu
  • I don't think it's as bad as it sounds. It doesn't say it can't detect objects, just that it can't always determine if that object is a human. So it's not going to just run people over. If you had to decide between hitting a cone or a person, most people would prefer to hit the cone. Autonomous cars strive to make the same decisions. Another thought, we already know Google's autonomous cars try to predict what an object might do next. A cone will likely act differently than a human, which may affect how the

    • which may affect how the car chooses to act when it gets close to the object/person. (Slowing down when it gets close to the human, even if it's not in the car's direct path)

      This is why the robot car will suck. It will have to be slow to be over-cautious, and this will make it unappealing, if not the laughing stock of other car owners. If you've ever been a passenger of on old person you will know this is not an experience anyone will pay money for.

In seeking the unattainable, simplicity only gets in the way. -- Epigrams in Programming, ACM SIGPLAN Sept. 1982