12
$\begingroup$

I am working on Robot localization and navigation in urban environments. I want to use a camera. But I am a little bit confused about LRF (laser range finder) data or other laser data.

Why people want to use camera?

Why not LRF or other laser data?

Can anyone explain please in favor of Camera?

$\endgroup$

7 Answers 7

17
$\begingroup$

A 3D laser range finder or LIDAR such as the one on the Google Car is far more expensive than a camera. The other reason is that while in case of a LIDAR the distance of every pixel is available, the generated data to be processed is enormous. You have to transfer and process data faster which comes out again as rising cost. Finally cameras usually have a longer lifetime, so there is less maintenance required.

With relatively cheap cameras and computer vision, pretty good results can be achieved.

$\endgroup$
4
  • 3
    $\begingroup$ Cost is absolutely the answer. Quality laser scanners start (for the end user) generally around $10,000. Quality cameras are about a tenth of the cost. $\endgroup$
    – Chuck
    Oct 3, 2016 at 18:14
  • 1
    $\begingroup$ What sort of LIDAR are you thinking of? I've used one that returned about 1000 points per scan (on a 2D plane), but a typical modern camera returns millions of pixels, which is much more data. $\endgroup$
    – user253751
    Oct 3, 2016 at 21:05
  • 2
    $\begingroup$ @immibis - the Velodyne VLP-16 does about 300k points per second across 16 planes, and the SICK LMS511 does about 50k points per second across 1 plane. VLP-16 has a 360 degree field of view and is about 8k, the LMS511 has a 190 degree field of view and is about 10k, but is ruggedized for industrial use. These are distance measurements, not pictures. Cameras can of course return a higher resolution, but generally it takes such high firepower to do stereo etc. that the frames are downsampled to very low resolution B&W or the refresh rate is very low. $\endgroup$
    – Chuck
    Oct 4, 2016 at 1:12
  • 1
    $\begingroup$ So 300k points per second, versus 50-million-ish pixels per second. The camera still has more data to transfer. Of course in either case you can discard data/downsample if you can't process all of it fast enough. $\endgroup$
    – user253751
    Oct 4, 2016 at 1:31
4
$\begingroup$

In addition to those points in Bence's answer, cameras can:

  • Calculate many complex features that result in very robust matching between frames, and object recognition
  • High angular resolution (typical low->high range goes from $0.5^\circ$ -> $0.025^\circ$)
  • Lower power usage
  • Passive sensor (doesn't require 'clean' signal of a laser)
$\endgroup$
3
$\begingroup$

navigation in urban environments

Depending on the laser, there might be legal constraints on where you can use it. Running around town throwing laser rays around might require special permission/licence.

$\endgroup$
3
  • 5
    $\begingroup$ Sure, depending on the laser. But we're not talking about starship weapons, here. You don't need permission or a license to use a barcode scanner, for example. $\endgroup$ Oct 4, 2016 at 9:32
  • 1
    $\begingroup$ Most commercial LRFs (Velodyne, Hokuyo) use Class 1 lasers, and they're completely safe. Google, Uber etc. are already testing their prototypes outdoors with such LRFs installed. I really don't think their legal department is swamped with complaints from outraged parents.. $\endgroup$ Oct 4, 2016 at 16:03
  • $\begingroup$ See this related question about regulation of LiDAR in public spaces: robotics.stackexchange.com/questions/23477 $\endgroup$
    – Micropolis
    May 18, 2022 at 19:38
2
$\begingroup$

Like other already answered. Cameras typically are much cheaper than Laser Range Finders.

When you talk about camera you mean the 2D cameras isn't it? There are some 3D cameras like the ifm O3D3xx family of cameras available. Those camera may not have the accuracy of a laser scanner but they provide 3D depth data in reasonable frame rates at a price point of ~1k

$\endgroup$
1
$\begingroup$

Are there any advantages to using a LIDAR for SLAM vs a standard RGB camera?

You may check this link where I have previously answered a somewhat similar question. (advantages and disadvantages of each)

in urban environments

If you are referring to autonomous cars like the Google ones, there are lots of considerations and constraints (safety, cost etc.).

If you are interested in research & learning, I suggest that you use any hardware platform that is available.

Keep in mind:

  1. A car with a LIDAR that is extremely expensive won't be easily sold.
  2. A car moving autonmously around people, might kill in case of a mistake. Thus the considerations are different than just developing algorithms for the sake of research and learning.
$\endgroup$
1
$\begingroup$

I don't think people really "want" to use only cameras. If every researcher could afford the LiDARs they'd all put LiDARs on they robots for outdoor environment.

Cameras are pretty cheap and the only limit to range is the pixel/superpixel resolution that you can process in your algorithm/software.

Most researchers (including me) use structured light cameras (although they don't work outdoors, so we switch to RGB cameras on these sensors when the robot is outdoors). A solution to this light problem is that we also use stereo cameras (stereo vision/multi-view depth which is computationally expensive) for roughly determining depth, based on the processing capabilities of the controller/CPU. Another solution that I've yet to personally explore is to use multiple Kinects/Asus Xtions etc, where you get depth corroboration as well as multiple RGB cameras for outdoors.

LiDARs are typically very expensive (in the thousands of $$ for really good ones). Although this might change in the future with some companies coming out with $250 "LiDARs" like Sweep.

Also, LRF's/LiDARs have limited range and resolution (i.e., beyond a certain distance, they cannot resolve depth unambiguously and hence they return 0 values (I'm not sure specifically about LiDARs, but depth cameras have a maximum (above which) as well as minimum range (below which) they dont give you depth).

Hope this helps.

$\endgroup$
0
$\begingroup$

I am going to add another reason that frankly I was hoping someone else would bring up. Because why do we make robots in the first place? Emotionless machines to do our dirty work?

I think that the fact that a robot can rely purely on "vision" like we mammals do makes them more like us. So for me, lasers and sonars are cheating. What IMHO we should focus on in stead of cheating is making better cameras with higher frame-rate, higher dynamic range and less artifacts, and write software that can get the needed data from them. (Or, speaking in post 2012 terms, train our networks to get what data they need from them).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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