DrawingProcess
๋“œํ”„ DrawingProcess
DrawingProcess
์ „์ฒด ๋ฐฉ๋ฌธ์ž
์˜ค๋Š˜
์–ด์ œ
ยซ   2025/06   ยป
์ผ ์›” ํ™” ์ˆ˜ ๋ชฉ ๊ธˆ ํ† 
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30
  • ๋ถ„๋ฅ˜ ์ „์ฒด๋ณด๊ธฐ (967)
    • Profile & Branding (25)
      • Career (18)
    • IT Trends (254)
      • Conference, Faire (Experien.. (31)
      • News (187)
      • Youtube (19)
      • TED (8)
      • Web Page (2)
      • IT: Etc... (6)
    • Contents (97)
      • Book (66)
      • Lecture (31)
    • Project Process (94)
      • Ideation (0)
      • Study Report (34)
      • Challenge & Award (22)
      • 1Day1Process (5)
      • Making (5)
      • KRC-FTC (Team TC(5031, 5048.. (10)
      • GCP (GlobalCitizenProject) (15)
    • Study: ComputerScience(CS) (72)
      • CS: Basic (9)
      • CS: Database(SQL) (5)
      • CS: Network (14)
      • CS: OperatingSystem (3)
      • CS: Linux (39)
      • CS: Etc... (2)
    • Study: Software(SW) (95)
      • SW: Language (29)
      • SW: Algorithms (1)
      • SW: DataStructure & DesignP.. (1)
      • SW: Opensource (15)
      • SW: Error Bug Fix (43)
      • SW: Etc... (6)
    • Study: Artificial Intellige.. (149)
      • AI: Research (1)
      • AI: 2D Vision(Det, Seg, Tra.. (35)
      • AI: 3D Vision (70)
      • AI: MultiModal (3)
      • AI: SLAM (0)
      • AI: Light Weight(LW) (3)
      • AI: Data Pipeline (7)
      • AI: Machine Learning(ML) (1)
    • Study: Robotics(Robot) (33)
      • Robot: ROS(Robot Operating .. (9)
      • Robot: Positioning (8)
      • Robot: Planning & Control (7)
    • Study: DeveloperTools(DevTo.. (83)
      • DevTool: Git (12)
      • DevTool: CMake (13)
      • DevTool: NoSQL(Elastic, Mon.. (25)
      • DevTool: Container (17)
      • DevTool: IDE (11)
      • DevTool: CloudComputing (4)
    • ์ธ์ƒ์„ ์‚ด๋ฉด์„œ (64)
      • ๋‚˜์˜ ์ทจ๋ฏธ๋“ค (7)
      • ๋‚˜์˜ ์ƒ๊ฐ๋“ค (42)
      • ์—ฌํ–‰์„ ๋– ๋‚˜์ž~ (10)
      • ๋ถ„๊ธฐ๋ณ„ ํšŒ๊ณ  (5)

๊ฐœ๋ฐœ์ž ๋ช…์–ธ

โ€œ ๋งค์ฃผ ๋ชฉ์š”์ผ๋งˆ๋‹ค ๋‹น์‹ ์ด ํ•ญ์ƒ ํ•˜๋˜๋Œ€๋กœ ์‹ ๋ฐœ๋ˆ์„ ๋ฌถ์œผ๋ฉด ์‹ ๋ฐœ์ด ํญ๋ฐœํ•œ๋‹ค๊ณ  ์ƒ๊ฐํ•ด๋ณด๋ผ.
์ปดํ“จํ„ฐ๋ฅผ ์‚ฌ์šฉํ•  ๋•Œ๋Š” ์ด๋Ÿฐ ์ผ์ด ํ•ญ์ƒ ์ผ์–ด๋‚˜๋Š”๋ฐ๋„ ์•„๋ฌด๋„ ๋ถˆํ‰ํ•  ์ƒ๊ฐ์„ ์•ˆ ํ•œ๋‹ค. โ€

- Jef Raskin

๋งฅ์˜ ์•„๋ฒ„์ง€ - ์• ํ”Œ์ปดํ“จํ„ฐ์˜ ๋งคํ‚จํ† ์‹œ ํ”„๋กœ์ ํŠธ๋ฅผ ์ฃผ๋„

์ธ๊ธฐ ๊ธ€

์ตœ๊ทผ ๊ธ€

์ตœ๊ทผ ๋Œ“๊ธ€

ํ‹ฐ์Šคํ† ๋ฆฌ

hELLO ยท Designed By ์ •์ƒ์šฐ.
DrawingProcess

๋“œํ”„ DrawingProcess

Study: Artificial Intelligence(AI)

[Vision] Visual SLAM vs Visual-Inertial Odometry(VIO)

2023. 3. 19. 08:04
๋ฐ˜์‘ํ˜•
๐Ÿ’ก ๋ณธ ๋ฌธ์„œ๋Š” cv-learn ๋ธ”๋กœ๊ทธ์—์„œ 'Visual-SLAM vs VIO'์— ๋Œ€ํ•ด ํผ์˜จ ๊ธ€์ž…๋‹ˆ๋‹ค.
Visual SLAM vs Visual-Inertial Odometry(VIO) ์ฐจ์ด์ ์— ์ •๋ฆฌํ•˜์˜€์œผ๋‹ˆ ์ฐธ๊ณ ํ•˜์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค.

Visual SLAM vs Visual-Inertial Odometry(VIO)

3D pose + 3D map point๋ฅผ ๊ตฌํ•˜๊ณ  ์‹ถ์„ ๋•Œ,

  1. ์นด๋ฉ”๋ผ๋งŒ ์‚ฌ์šฉํ•˜๋ฉด Visual-SLAM
  2. ์นด๋ฉ”๋ผ + IMU๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด Visual-inertial odometry (VIO)

Visual odometry๊ฐ€ ์•„๋‹ˆ๋ผ Visual-SLAM์ด ๋˜๋ ค๋ฉด global consistency๊ฐ€ ์œ ์ง€๋˜์–ด์•ผํ•˜๋Š”๋ฐ (i.e. ๋™๋„ค ํ•œ๋ฐ”ํ€ด๋ฅผ ๋Œ๊ณ ์™€์„œ ์›์ ์œผ๋กœ ์™”์„ ๋•Œ, ์œ„์น˜ ์ •๋ณด์™€ scale์ด ์ •ํ™•ํ•˜๊ฒŒ ๋งž์•„๋–จ์–ด์ ธ์•ผํ•จ), ์ด๋Š” loop closure์™€ ๊ฐ™์€ ๊ธฐ๋Šฅ์œผ๋กœ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์–ด์•ผํ•œ๋‹ค. (Visual odometry๋งŒ์œผ๋กœ ์œ„์น˜์ •๋ณด์™€ scale์ด ์ •ํ™•ํ•˜๊ฒŒ ๋งž์•„๋–จ์–ด์ง€๋ฉด ๊ทธ๊ฒƒ๋„ SLAM์ด๋ผ๊ณ  ํ•  ์ˆ˜ ์žˆ๋‹ค. ํ•˜์ง€๋งŒ ์‹ค์ œ๋กœ๋Š” ๊ฑฐ์˜ ๋ถˆ๊ฐ€๋Šฅํ•˜๊ธฐ ๋•Œ๋ฌธ์— ํ•™๊ณ„์—์„œ๋Š” ์ž˜ ๋ฐ›์•„๋“œ๋ ค์ง€์ง€ ์•Š๋Š” ์ „์ œ์ด๋‹ค.)

Visual-SLAM๊ณผ VIO ๋ชจ๋‘ Monocular, Stereo, Multi-camera, RGB-D ํ˜•ํƒœ๋ฅผ ๊ฐ€์งˆ ์ˆ˜ ์žˆ๋‹ค.

Visual-SLAM

  • ์นด๋ฉ”๋ผ๋งŒ ์‚ฌ์šฉํ•ด์„œ 3D pose + 3D map point๋ฅผ ๊ตฌํ•˜๋Š” ๋ฐฉ๋ฒ•
  • ์ตœ๊ทผ ์—ฐ๊ตฌ๋Š” ์ฃผ๋กœ graph-optimisation ๊ธฐ๋ฐ˜์˜ SLAM ๋ฐฑ์—”๋“œ๋ฅผ ์‚ฌ์šฉํ•จ
    • Total reprojection error๋ฅผ ์ตœ์†Œํ™”ํ•˜๋Š” ๋ฐฉํ–ฅ
  • ์œ ๋ช…ํ•œ ์˜คํ”ˆ์†Œ์Šค ํ”„๋กœ์ ํŠธ
    • Monocular: MonoSLAM, PTAM, ORB-SLAM, SVO, DSO
    • Stereo: ORB-SLAM2, Stereo-DSO, ProSLAM
    • RGB-D: KinectFusion, ElasticFusion, BundleFusion

์žฅ์ 

  • VIO ๋ณด๋‹ค ์ ์€ ๊ณ„์‚ฐ๋Ÿ‰
  • ์นด๋ฉ”๋ผ๋งŒ ์žˆ์œผ๋ฉด ์ž‘๋™ํ•  ์ˆ˜ ์žˆ์Œ

๋‹จ์ 

์นด๋ฉ”๋ผ๊ฐ€ ์ž˜ ์ž‘๋™ํ•˜์ง€ ์•Š๋Š” ์ƒํ™ฉ์—์„œ๋Š” Visual-SLAM๋„ ์ž˜ ์ž‘๋™ํ•˜์ง€ ๋ชปํ•จ

  • e.g. ํ•˜์–€ ๋ฒฝ์„ ๋ฐ”๋ผ๋ณด๋ฉฐ feature๊ฐ€ ์ „ํ˜€ ๋ฝ‘ํžˆ์ง€ ์•Š์„ ๋•Œ
  • e.g. ๋ฐ˜๋ณต๋˜๋Š” ํŒจํ„ด์ด ์—ฐ์†์ ์œผ๋กœ ๋‚˜ํƒ€๋‚˜์„œ ์ œ๋Œ€๋กœ ๋œ ํ”ผ์ณ ๋งค์นญ์ด ๋˜์ง€ ์•Š์„ ๋•Œ
  • e.g. ๋„ˆ๋ฌด ์–ด๋‘ก๊ฑฐ๋‚˜ ๋„ˆ๋ฌด ๋ฐ์•„์„œ ์•„๋ฌด๊ฒƒ๋„ ๋ณด์ด์ง€ ์•Š์„ ๋•Œ

 

Visual-inertial odometry (VIO)

  • ์นด๋ฉ”๋ผ์™€ IMU๋ฅผ ์‚ฌ์šฉํ•ด์„œ 3D pose + 3D map point๋ฅผ ๊ตฌํ•˜๋Š” ๋ฐฉ๋ฒ•
  • ์ตœ๊ทผ graph-optimisation ๊ธฐ๋ฐ˜ ๋ฐฑ์—”๋“œ๊ฐ€ ์œ ํ–‰ํ•˜๊ธฐ ์‹œ์ž‘ํ–ˆ์œผ๋‚˜, ์‹œ์ค‘ ์ œํ’ˆ ์ค‘์—๋Š” ์•„์ง filter ๊ธฐ๋ฐ˜์˜ ์†Œํ”„ํŠธ์›จ์–ด๊ฐ€ ๋งŽ์Œ (e.g. Google ARCore, Apple ARKit)
    • IMU pre-integration ์—ฐ๊ตฌ ์ดํ›„ ๊ฐ€๋Šฅํ•ด์ง
      • ์ด์ „์—๋Š” IMU ๊ฐ’์„ ๋ณด์ • ์ „/ํ›„๋กœ state propagation์„ ๋‘๋ฒˆ ํ•ด์•ผํ–ˆ์ง€๋งŒ, pre-integration ์ดํ›„๋กœ ํ•œ๋ฒˆ๋งŒ ํ•ด๋„ ๋˜๊ฒŒ ๋งŒ๋“ค๋ฉด์„œ ๋™์‹œ์— ์ˆ˜๋งŽ์€ ์„ผ์„œ ๊ด€์ธก์น˜๋ฅผ ๋‹จ ํ•˜๋‚˜์˜ ๊ด€์ธก์น˜๋กœ ์ •๋ฆฌํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋จ.
    • Graph-optimisation ๊ธฐ๋ฐ˜์˜ ๋ฐฉ์‹์ด ๋ฐœ์ „ํ•˜๋ฉด์„œ loop closure ๊ธฐ๋Šฅ์„ ๊ฐ–์ถ”๊ฒŒ ๋จ
      • ์ดˆ์ฐฝ๊ธฐ์˜ loop closure๊ฐ€ ์—†๋Š” VIO์™€ ๊ตฌ๋ถ„ํ•˜๊ธฐ ์œ„ํ•ด VINS (Visual-Inertial System)๋ผ๊ณ  ํ‘œํ˜„ํ•˜๋Š” ๋ฐฉ๋ฒ•๋„ ์ƒ๊น€ (ํ•˜์ง€๋งŒ ์ •์‹ ๋‹จ์–ด๋Š” ์•„๋‹˜)
  • ์œ ๋ช…ํ•œ ์˜คํ”ˆ์†Œ์Šค ํ”„๋กœ์ ํŠธ
    • Monocular: MSCKF, OKVIS, ROVIO, VINS-Mono, ORB-SLAM3
    • Stereo: S-MSCKF, VI-DSO
    • RGB-D: ElasticFusion+IMU

์žฅ์ 

  • ์นด๋ฉ”๋ผ์™€ IMU๋Š” ์„ผ์„œ ํŠน์„ฑ์ƒ ์„œ๋กœ์˜ ๋‹จ์ ์„ ๋ณด์™„ํ•จ
    • e.g. ์–ด๋‘์šด ๊ณณ์—์„œ ์นด๋ฉ”๋ผ๊ฐ€ ์•„๋ฌด๊ฒƒ๋„ ๋ณด์ง€ ๋ชปํ•  ๋•Œ, IMU ์„ผ์„œ๋Š” ๊ณ„์† pose ์ถ”์ • ๊ฐ€๋Šฅ (VIO ์„ค๊ณ„์— ๋”ฐ๋ผ ๋‹ค๋ฆ„)
  • Monocular์˜ ๊ฒฝ์šฐ ์นด๋ฉ”๋ผ๋งŒ ์‚ฌ์šฉํ–ˆ์„ ๋•Œ๋Š” ์–ป์–ด๋‚ผ ์ˆ˜ ์—†๋Š” metric scale ์ •๋ณด๋ฅผ IMU motion์„ ํ†ตํ•ด ์ถ”์ • ๊ฐ€๋Šฅ (i.e. up-to-scale ๋งต์ด ์•„๋‹Œ ์‹ค์ œ ์„ธ์ƒ์˜ scale์„ ๊ฐ€์งˆ ์ˆ˜ ์žˆ์Œ)
  • ๋” ๋งŽ์€ ์ตœ์ ํ™” constraint๋กœ ์–ป์–ด๋‚ด๋Š” ๋†’์€ ์ •ํ™•๋„
  • ์นด๋ฉ”๋ผ ํ”ผ์ณ ํŠธ๋ž™ํ‚น์„ ํ•  ๋•Œ ๋ชจ์…˜๋ชจ๋ธ ๊ฐ€์„ค ๋Œ€์‹  IMU ์„ผ์„œ๋กœ ์–ป์€ ๋ชจ์…˜ ์‚ฌ์šฉ ๊ฐ€๋Šฅ
    • ์‹ค์ œ ์„ผ์„œ ๊ฐ’์„ ์‚ฌ์šฉํ•จ์œผ๋กœ์จ ํƒ์ƒ‰ ๋ฒ”์œ„๋ฅผ ์ถ•์†Œ ๊ฐ€๋Šฅ

๋‹จ์ 

  • ๋” ๋งŽ์€ ์ตœ์ ํ™” constraint๋กœ ์ธํ•œ ๋งŽ์€ ๊ณ„์‚ฐ๋Ÿ‰
    • ๊ณ ์„ฑ๋Šฅ CPU๋ฅผ ์š”๊ตฌํ•  ์ˆ˜ ์žˆ์Œ
  • ์นด๋ฉ”๋ผ-IMU ๊ฐ„์˜ ์ •ํ™•ํ•œ ์บ˜๋ฆฌ๋ธŒ๋ ˆ์ด์…˜์ด ํ•„์š”
    • ์–‘์‚ฐ์„ ํ•  ๊ฒฝ์šฐ ์ถ”๊ฐ€ ๊ณต์ • ํ•„์š”
  • ๋‘ ์„ผ์„œ ์ค‘ ํ•˜๋‚˜๊ฐ€ ์‹คํŒจํ•˜๋ฉด ์ „์ฒด๊ฐ€ ์‹คํŒจํ•จ (VIO ์„ค๊ณ„์— ๋”ฐ๋ผ ๋‹ค๋ฆ„)

์ฐธ๊ณ 

  • [Blog] Visual-SLAM vs VIO: https://www.cv-learn.com/20210817-vslam-vs-vio/
๋ฐ˜์‘ํ˜•
์ €์ž‘์žํ‘œ์‹œ ๋น„์˜๋ฆฌ ๋ณ€๊ฒฝ๊ธˆ์ง€ (์ƒˆ์ฐฝ์—ด๋ฆผ)

'Study: Artificial Intelligence(AI)' ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๋‹ค๋ฅธ ๊ธ€

[Vision] RADAR ๋ ˆ์ด๋” ์„ผ์„œ์˜ ๊ตฌ๋™๋ฐฉ์‹ ๋ฐ ํŠน์ง•  (0) 2023.03.21
[Vision] LiDAR ๋ผ์ด๋‹ค ์„ผ์„œ์˜ ๊ตฌ๋™๋ฐฉ์‹ ๋ฐ ํŠน์ง•  (1) 2023.03.20
[Vision] Key-Point Extraction: ์˜์ƒ ํŠน์ง•์  ์ถ”์ถœ  (0) 2023.03.16
[๋…ผ๋ฌธ ๋ฆฌ๋ทฐ] ORB-SLAM: a Versatile and Accurate Monocular SLAM System  (0) 2023.03.16
[Vision] SURF (Speed-Up Robust Feature)์˜ ์›๋ฆฌ.txt  (0) 2021.06.08
    'Study: Artificial Intelligence(AI)' ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๋‹ค๋ฅธ ๊ธ€
    • [Vision] RADAR ๋ ˆ์ด๋” ์„ผ์„œ์˜ ๊ตฌ๋™๋ฐฉ์‹ ๋ฐ ํŠน์ง•
    • [Vision] LiDAR ๋ผ์ด๋‹ค ์„ผ์„œ์˜ ๊ตฌ๋™๋ฐฉ์‹ ๋ฐ ํŠน์ง•
    • [Vision] Key-Point Extraction: ์˜์ƒ ํŠน์ง•์  ์ถ”์ถœ
    • [๋…ผ๋ฌธ ๋ฆฌ๋ทฐ] ORB-SLAM: a Versatile and Accurate Monocular SLAM System
    DrawingProcess
    DrawingProcess
    ๊ณผ์ •์„ ๊ทธ๋ฆฌ์ž!

    ํ‹ฐ์Šคํ† ๋ฆฌํˆด๋ฐ”