Saturday, November 26, 2016

AutoSens: Softkinetic Compares ToF with Other Approaches

AutoSens 2016 kindly permitted me to post a couple of slides from Softkinetic presentation "3D depth-sensing for automotive: bringing awareness to the next generation of (autonomous) vehicles" by Daniel Van Nieuwenhove. A good part of the presentation compares ToF with active and passive stereo solutions:

8 comments:

  1. Do you mean that for 100m@30fps, minium for car use, you need (100/2.5)^$1W=1600W laser?

    ReplyDelete
    Replies
    1. Assuming you're talking about the car moving at speed, then I would go for radar over lidar every time. For automated parking systems I would choose lidar imaging like ToF. There was a paper this year in Optics Express that looked at long range lidar though.

      Delete
  2. Nice overview, thanks. What is missing is the total exposure time needed in relation to object speed for a decent measurement. Most TOF solutions I am aware of have serious issues with fast moving objects.

    ReplyDelete
    Replies
    1. how fast are you talking about when it becomes a serious issue?

      Delete
    2. For TOF systems that rely on phase shift (and thus need multiple pulses for one 3D measurment) such as SwissRanger or Kinect 2, it's actually not very fast. Closing the fingers into a fist as if catching an object is sufficiently fast at about one meter distance to cause major artifacts at the finger boundaries.

      Structured light on the other hand can compute depth from a single frame, so it's only the native exposure time of the IR-enabled sensor that limits it. And even beyond that, you just get motion blur instead of deformed geometry.

      No experience with other TOF principles, would love to hear more about those.

      Delete
    3. What structured light systems are you talking about? Don't structured light systems use multiple frames? e.g http://link.springer.com/chapter/10.1007%2F978-3-319-30973-6_2

      I can think of how you'd get distance from one structured light frame, but that would require the ability to analyse the pattern (at a computational cost), and fine structure like fingers would increase difficulty because it'd be harder to see that pattern.

      I'd expect stereo to do better than both ToF and structured light for detail for fingers, but that has it's own difficulties as well. Which is the best option of the three is entirely dependent on the application.

      Delete
  3. It is not clear how the power consumption scales with range in the TOF approach. That is missing from this analysis and perhaps that is another negative for ToF compared to other approaches

    ReplyDelete
    Replies
    1. Maybe, but it is at least a square law between laser power and the distance. There is a serious questionning for the feasibility of such approach for long range applications.

      Delete

All comments are moderated to avoid spam.