Bundle Adjustment - 5 Minutes with Cyrill

แชร์
ฝัง
  • เผยแพร่เมื่อ 21 พ.ย. 2024

ความคิดเห็น • 14

  • @naingtanba6819
    @naingtanba6819 3 ปีที่แล้ว +59

    I think including pictures will provide clearer explanations.

  • @duhlmann1
    @duhlmann1 ปีที่แล้ว +1

    Great video!

  • @olamurad8345
    @olamurad8345 ปีที่แล้ว +1

    Can you help me with articles on how to bundle Adjustment in detail?

  • @jonnyjuk
    @jonnyjuk 2 ปีที่แล้ว +3

    Got a problem with Computer Vision? Better Call Cyrill ;-)

    • @CyrillStachniss
      @CyrillStachniss  2 ปีที่แล้ว +2

      😂 thanks - I hope not everyone has my number…

  • @christofhapp
    @christofhapp ปีที่แล้ว

    Hello Cyrill. Thanks for your Videos!
    If you "only" have image point correspondences and use e.g. the 8-point algorithm to estimate your 3D calibration (using the intrinsics from the camera manufacturer as an initial guess). HOW can SBA help to refine the intrinsic parameters? I would guess that you cannot get rid of the perspective ambiguity.

    • @CyrillStachniss
      @CyrillStachniss  ปีที่แล้ว

      The 8point algo determines at max 7 dofs, it the fundamental matrix and not all intrinsics.

  • @kiges
    @kiges ปีที่แล้ว

    How are the equations for SfM or Collinearty when using turntable and fixed camera? Is there a resource that explains the equations for the fixed camera and the rotating object?

  • @philippelaferriere2661
    @philippelaferriere2661 3 ปีที่แล้ว

    Great video.

  • @briancannard7335
    @briancannard7335 ปีที่แล้ว

    Why not use autoencoders instead?

  • @vikasshetty3738
    @vikasshetty3738 4 ปีที่แล้ว

    Can we estimate the visual odometry in absolute scale using bundle adjustment for monocular camera?

    • @abhishek.goudar
      @abhishek.goudar 3 ปีที่แล้ว +7

      I am not an expert in visual techniques, but I believe visual odometry and bundle adjustment are intended to solve two different problems. Bundle adjustment aims to recreate both the camera trajectory and a map of the surrounding environment and can be computationally expensive.
      Visual odometry, as the name suggests serves a source of 'odometry', that is, it provides relative motion estimates. In this case no map of the environment in generated and as such is much more computationally inexpensive. One can view VO algorithms as a subset of BA algorithms.
      Coming to the scale part both BA and VO have a hard time estimating scale in monocular settings without some additional information. In case of VO, an inertial measurement unit (IMU) can be used to estimate and track scale overtime. Hope that helps.

    • @CyrillStachniss
      @CyrillStachniss  3 ปีที่แล้ว +5

      Using only monocular images without baseline information or odometry will not allow you to recover the scale. The best you can get is a so-called "photogrammetric model", i.e., a 3D model that is only determined up to a similarity transform.