for Robot Artificial Inteligence

5. Stereo & RGB-D camera model and Image Information

|

Stereo Camera Model

Recap : Pinhole camera model

Intro

Stereo camera principle

  • b is Baseline
  • d is distance between left and right camera that looking at point of object.

RGB-D camera Model

  • 2 Types of RGB-D Camera
    • Structured Light RGB-D Camera
    • Time of Flight(TOF) RGB-D Camera

  • Structured light를 쏘고 Encoding pattern 으로 받고(Acquisition)
    • Encoding patters: 시간추로 봤을때 라인마다 코드가 있다 (Stripe ID)
  • Decoding하여서(stripe ID 매칭) 3D 구조화 한다.

  • 카메라 안에있는 셀(Emitor)을 통해서 방출과 받음을 반복하는 Frequency를 이용하여 3D 구현화 (Kinetic 2 )
  • Lidar와 다른점
    1. 적외선 사용
    2. Lidar Force처럼 촘촘하게 만드는것이 어려움

Limitation of RGB-D Camera

  • Active Emission & reception
  • Limited Range for depth
  • Not available for outside usage(interference of natural light and other sensor)
  • Cannot handle transparent materials
  • How to overcome? Deep Learning Based Approach

Image information

  • Def: Data obtained by converting 3D space information into 2D digital format through camera equipment

  • Useful information can be obtained by digital image processing

  • Intensity value is 0~255. (Expressed using 8-bit int)
    • Grayscale image -> 0 (dark) to 255 (light)
      • 1 pixel occupies 8 bits of storage space
    • RGB image -> 0,0,0 (dark) to 255,255,255 (bright)
      • 1 pixel occupies 8x3 = 24 bits of storage
  • In case of RGB-D image, it contains depth data (expressed using 16bit int)
    • 0~65536 (up to 65m can be expressed when displayed in meters)

Reference

SLAM KR

Comments