1
$\begingroup$

I saw in this post on Self driving cars that said the cars developed by Google for the Google Self Driving Car Project, have sensors that can detect objects like humans, plastic bags within a range of two football fields. In my question, I have said "200 metres" which is rough figure because I think a football field would be anywhere between 90 metres to 120 metres.

What kind of sensors are capable of detecting objects in such a long range? If such sensors are available, are they used in any other devices or vehicles today?

$\endgroup$
1
  • 2
    $\begingroup$ Have you tried looking thru a pair of binoculars lately? $\endgroup$ – Carl Witthoft Feb 14 '16 at 17:16
3
$\begingroup$

Yes, these kind of sensors exists, they're called radars.

Military grade naval radars can detect a coin from tens of kilometers away, so that kind of performance (~1m² objects detected from ~100 meters) is not surprising. The keys are frequency (the higher the frequency, higher the accuracy) and signal processing.

However, I have doubts about the idea of dozens of these radars co-existing and overlapping each other's frequencies.

Edit : Also, it is possible to use high definition cameras and have a computer analyzing the images to detect moving objects,but it is more expensive (in terms of money and calculation capacity needed) than a radar-based technology

$\endgroup$
5
  • $\begingroup$ Hi, employee of Autoliv.com here. In the immediate future, automobiles will have 5 or more radar units on them. The difficulties with interference and so on are well known and (gosh!) there are things we're doing to ameliorate the situation. $\endgroup$ – Carl Witthoft Feb 14 '16 at 17:18
  • $\begingroup$ Hi, are you able to talk about in more detail ? I was thinking about a cognitive radio-like system where each car detects which frequency it can use, but what happens for example in a dense traffic where every frequency is already in use ? I guess you could make the cars communicating so some of them can transmit their radar data to the other, reducing the spectrum's occupation, but that implies an unique standard used by each manufacturer ... ? $\endgroup$ – MaximGi Feb 14 '16 at 17:42
  • $\begingroup$ Sorry - competitive advantage and all that. Probably once things go into mass production you will be able to find "how it works" stuff online. $\endgroup$ – Carl Witthoft Feb 14 '16 at 21:04
  • $\begingroup$ @MaximGi Navy radar would be detecting a coin at tens of km distance. But the coin would be against clear sky, and it would probably have a Doppler shift (such is the nature of the coins that the Navy trains its radars on). A pedestrian at 200m has clutter in the background, and its Doppler shift would not be too different from that of the background. Different problem. $\endgroup$ – Nick Alexeev Feb 14 '16 at 22:29
  • $\begingroup$ ...So ? You could also have mentioned that a pedestrian size is most likely to be about 1.50m instead of 1m, I was not making a detailed course about radars, just saying that the idea of "sensors detecting people from 200 m away" was not so surprising $\endgroup$ – MaximGi Feb 15 '16 at 8:31
2
$\begingroup$

The sensors which the article you linked are referring to are LIDARs. A LIDAR is a laser used in conjunction with an optical scanning system. By using sophisticated software and a LIDAR it is possible to do what is called "scene generation" which means creating a 3D model of the terrain (and objects) in front of the car.

LIDARS ranges are only limited by the power of the laser and resolution of the scanner. In theory you could map the surface of the moon with a LIDAR if you had a powerful enough laser. A scene generation system with a range of 200 yards ("two football fields") is feasible and many such systems exist.

Such systems are not commonly used, even in autonomous systems like UAVs, for several reasons: they are expensive, fragile, and suffer from motion effects. For example, for aerial UAVs one of the big problems is what is called "canyoning". As the vehicle traverses an area the parts of the scene to the forward right and left change very rapidly due to angular effects. This makes it very hard to compute the scene fast enough, except for things that are right in front of you.

The main use of 3D LIDARs today is to scan stationary objects like buildings and vehicles so they can be easily entered into a computer system as 3D models.

$\endgroup$
1
$\begingroup$

LIDAR is another possible candidate. This works on a similar principal to radar but uses laser light rather than radio waves for scanning. It is already widely use for things like high resolution mapping and creating digital models of complex objects.

It also has a few potential advantages over radar, including higher resolution and the fact that it is no impinging on an already crowded RF spectrum, similarly light doesn't pose the sort of electromagnetic interference problems associated with radar.

In practice it is likely that a practical automated car would use a variety of different sensors for different tasks. Perhaps forward looking Lidar for detailed obstacle detection, video for all round situational awareness and ultrasound for low speed maneuvering (eg parking).

$\endgroup$
1
  • $\begingroup$ Lidar is useless in rain/fog, and requires a rather more mechanically complicated device. And I fail to see why you think that multiple LIDAR sources from separate cars wouldn't cause false signal detections. Light is EM radiation and the detector doesn't care whence it came. $\endgroup$ – Carl Witthoft Feb 15 '16 at 12:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.