Slashdot Mirror
Google Research Leads To Automated Real-Time Pedestrian Detection
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.
So, in other words, there's a 58% chance you can hit the pedestrian?
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 testing autonomous cars for some years already without having still tackled such a secondary(?!) issue.
Idea 2: after quickly skimming through this paper, 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. Additionally, we are talking about their training dataset whose exact complexity is not too clear. For example: what about a kid suddenly crossing the street?, how good is this algorithm at differentiating between persons and similar shapes (human-like advertisement)?, how does it behave in poor-visibility conditions?, etc.
I am completely aware about the tremendous difficulty associated with accomplishing the expected goal and the outputs so far seem promising. But why are they implying that something is almost done, when quite a few basic problems haven't still been tackled?
Custom Solvers 2.0 = Alvaro Carballo Garcia = varocarbas.
This one time, I detected an Automated Real-Time Pedestrian. Now Google can do the same!
Though if they are only missing 26.2% of all pedestrians I'd call that a fail. If they are missing 74.8% that is still a fail. They should not be hitting any pedestrians. I believe eye contact between driver and pedestrian is a pretty important factor, we are pretty good at reading each other's body language.
We need to bridge the uncanny valley between driverless car and pedestrian to get this fool proof.
Look, Google cars have had a few hits, they released ONE data set showing a hit from behind, what I noticed on that metric was THERE WERE NO PEDESTRIANS being tracked. It's like it wasn't seeing pedestrians.
I also notice that other report it sits in the middle of the road far more than normal, which suggests to me the fix was to avoid the sides as much as possible and only spot pedestrians if they step out into the road.
So really what I want is a little less icky-sticky-butty-licky from the regulators and a bit more hard testing. A bit less "computer driver cars are so safe that insurance should be lower" and "people were afraid of driverless elevators" marketing and misdirection, and a bit more, throwing dummies out into the road to see how it responds, spraying water at the sensors to see how it handles it, blowing a black bin bag over the sensor, all the normal fail scenarios.
Proper f**ing testing. The kind the independent test agencies are supposed to do, in case Google engineers are self deluding themselves as to their own skills.
The last pieces of the survival of the fittest are being removed.
It just has to be equal to a human driver - and human drivers are not that good.
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.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
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.
<blink>down the rabbit hole</blink>
a 26.2% average miss rate isn't all that impressive. the caltech data set is 10 hours of 30 fps video (http://www.vision.caltech.edu/Image_Datasets/CaltechPedestrians/) but I can't tell from there (or the two papers that describe the dataset) what time of day the video was captured. Is this "daytime, middle of day" or is it "looking into the sun" and "2AM, headlights and streetlights"
The test database has also been preprocessed to stabilize the images.
Who is legally responsible if an automatic car does hit someone?
"If any question why we died, Tell them because our fathers lied."
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!
http://fusion.net/story/139703/self-parking-car-accident-no-pedestrian-detection/
I feel fantastic, and I'm still alive.
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 accomplished with a direct insinuation that self-driving cars enjoy violence in a casually indifferent manner. This so is cognitively dissonant to the entrenched idea that self-driving cars are selflessly noble creations entrusted with the protection of those they serve, it causes a jarring and novel rearrangement of thought. The presence of novelty and surprise and the idea of something horrid happening that (yet) does not directly threaten the reader mixes in a dash of relief, and is a recipe for humor. The reader is left with an adjusted impression of self-driving cars that is ludicrously disfigured by the joke, and the humorous moment will persist until novelty fades, which is also to say the period it takes for this new neuronal pathway to be established, which may be several seconds. The effect is heightened by the 'attachment' clause, which evokes a social meme of our own indifferent treatment of so-called 'lesser' forms of life. Placing the pedestrian-human as the 'lesser' form of life is another counterpoint.
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.
The audience is ever-ready for tie-in jokes that can extend the period of general hilarity. When delivered straight after they can build on the former. It does not matter whether the ploy makes sense or is likely if another indisputable fact is introduced, in this case, that Google blurs license plates. It creates a dangling question in the mind: what will be the next twist?
The follow-through for this one is weaker, for when it is finally understood it is a mere retelling of the first joke. Therefore a cheap device to stretch novelty was employed, the use of the colloquial 'em in a phrase that helps to introduce a new entity, a hypothetical 'spectator', perhaps the joke-teller, who not only approves of such things but is gung-ho enough to dispense with classical English. Presumably from excitement at the thought of this 'sport'. And as before, we are smearing Google with this dark sentiment, which is playfully dissonant to the company's desired philanthropic image.
Q: Why then is Google developing a 'real-time pedestrian detection' system?
A: To improve scoring and help populate their Deathrace 2015 leaderboard.
Clearly this trifecta has reached the end of its shelf-life. This may actually be the weakest joke of the three but some relief humor comes into play as the reader sees at a glance that it is the last and all will be over soon. The question is phrased as a wrap-up with a clear tie-in to TA, which promises novelty, albeit at a highly discounted rate. The rest is a desperate hodgepodge of cheap meme dropping (Deathrace) and a direct appeal to a presumably game-savvy audience with the idea of a 'leaderboard'. There exists the question whether the surge of hilarity ensuing from this is a result of a re-imagining of the whole series with cumulative hilarity, a particular aspect of this third joke or just relief that the series has ended.
Imagine if every joke was followed by commentary like this.
<blink>down the rabbit hole</blink>
I don't think I've ever seen someone bring up the privacy degradation that will be caused by self-driving cars. We can assume they'll eventually be linked with each other and some type of network to automatically report traffic, road usage (taxes?), crashes, mapping updates, etc... But I'd bet they'll also be data mining every person they see. These things are going to cause real-time, detailed surveillance of everything visible from a road. I've never seen a movie where the car reports on people outside it, only inside, yet all the car's cameras are pointed outwards.
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 furnace; slams door shut.
1: "WHAT did you just do? My dog is DEAD!"
2: "The fleas are dead; isn't that what you said you wanted?"
If the universe is someone's simulation -- does that mean the stars are just stuck pixels?
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 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)
It has to be much better than a human driver or the lawyers will be all over it. Human drivers get sued too (and/or hauled off to jail).
Google has deep pockets. You just have to have one death, and the lawyers will be all over it. The can smell a deep pocket from several states away.
I don't think some people posting here understand that video-based detection, the subject of this paper, is not the current object detection modality for Google's autonomous vehicles. Those primarily use LiDAR to detect people, cyclist, vehicles, and other objects. It is much easier and much more reliable to detect objects with LiDAR compared to video-based detection as you get a nice cluster of 3D points without having to worry about whether the sun is out or even in light fog or dusty conditions.
I believe Google would certainly like to use video more, as cameras are cheaper than LiDAR, but there's still a ways to go before video/image-based techniques will be as reliable as LiDAR.
I'm sure it's just a matter of time before the research is translated to autonomous weapons platforms.