Post

Room Segmentation Survey ๋…ผ๋ฌธ ์š”์•ฝ

๐Ÿ“‘ Room Segmentation Survey ๋…ผ๋ฌธ์„ ์š”์•ฝํ•œ ๊ธ€์ž…๋‹ˆ๋‹ค.

Room Segmentation Survey ๋…ผ๋ฌธ ์š”์•ฝ

R. Bormann, F. Jordan, W. Li, J. Hampp, and M. Hรคgele.
Room Segmentation: Survey, Implementation, and Analysis.
In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA) https://ieeexplore.ieee.org/abstract/document/7487234

Room Segmentation: Survey, Implementation, and Analysis

์†Œ๊ฐœ

Room Segmentation์€ ๋กœ๋ด‡ ๊ณตํ•™ ์‘์šฉ์—์„œ ๋งค์šฐ ์ค‘์š”ํ•œ ์ž‘์—…์œผ๋กœ, ์œ„์ƒ ๋งตํ•‘, ์˜๋ฏธ๋ก ์  ๋งตํ•‘, ๋‚ด๋น„๊ฒŒ์ด์…˜, ๊ทธ๋ฆฌ๊ณ  ํšจ์œจ์ ์ธ ์ž‘์—… ๊ณ„ํš ์ˆ˜๋ฆฝ์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ํ™˜๊ฒฝ์—์„œ ์˜์—ญ(๋ฐฉ)์˜ ๊ฒฝ๊ณ„๋ฅผ ์ •ํ™•ํžˆ ์ •์˜ํ•˜๋Š” ๊ฒƒ์€ ๋กœ๋ด‡์˜ ์„ฑ๋Šฅ์— ์ค‘์š”ํ•œ ์˜ํ–ฅ์„ ๋ฏธ์น˜๋ฉฐ, ์ด๋Š” ์ฃผ๋กœ ๋ฐ์ดํ„ฐ ํšจ์œจ์„ฑ๊ณผ ๋‚ด๋น„๊ฒŒ์ด์…˜ ๊ฒฝ๋กœ ์ตœ์ ํ™” ์ธก๋ฉด์—์„œ ๋‘๋“œ๋Ÿฌ์ง‘๋‹ˆ๋‹ค. ์ด ๋…ผ๋ฌธ์€ ํ˜„์žฌ๊นŒ์ง€ ์ œ์•ˆ๋œ ์—ฌ๋Ÿฌ Room Segmentation ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๋น„๊ต ๋ถ„์„ํ•˜๋ฉฐ, ๋„ค ๊ฐ€์ง€ ์ธ๊ธฐ ์žˆ๋Š” ๋ฐฉ๋ฒ•์˜ ๊ตฌํ˜„๊ณผ ์ด๋“ค์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ณ  ํŠน์ • ์‘์šฉ ํ™˜๊ฒฝ์—์„œ ์ตœ์ ์˜ ์„ ํƒ์„ ์•ˆ๋‚ดํ•ฉ๋‹ˆ๋‹ค.


์ฃผ์š” ๊ธฐ์—ฌ

  1. ๋‹ค์–‘ํ•œ Room Segmentation ์ ‘๊ทผ๋ฒ•์— ๋Œ€ํ•œ ํฌ๊ด„์ ์ด๊ณ  ์‹ฌ์ธต์ ์ธ ์กฐ์‚ฌ.
  2. ๋„๋ฆฌ ์‚ฌ์šฉ๋˜๋Š” ๋„ค ๊ฐ€์ง€ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์˜คํ”ˆ ์†Œ์Šค ๊ตฌํ˜„ ์ œ๊ณต.
  3. ์ •์„ฑ์  ๋ฐ ์ •๋Ÿ‰์  ๊ธฐ์ค€์— ๋”ฐ๋ฅธ ๋น„๊ต ๋ถ„์„ ์ˆ˜ํ–‰.
  4. ํ‰๊ฐ€๋ฅผ ์œ„ํ•œ 20๊ฐœ์˜ ๋ณต์žกํ•œ ํ‰๋ฉด๋„ ๋ฐ์ดํ„ฐ์…‹๊ณผ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ™˜๊ฒฝ ์ œ๊ณต.
  5. ๋กœ๋ด‡ ์ฒญ์†Œ ์ž‘์—… ๋“ฑ ์‹ค์ œ ์‘์šฉ ์‚ฌ๋ก€๋ฅผ ํ†ตํ•ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์‹ค์งˆ์ ์ธ ์„ฑ๋Šฅ ํ‰๊ฐ€.

๋ถ„ํ•  ์•Œ๊ณ ๋ฆฌ์ฆ˜

1. ํ˜•ํƒœํ•™์  ๋ถ„ํ•  (Morphological Segmentation)

์ž‘๋™ ์›๋ฆฌ
  • ๊ธฐ๋ณธ ์•„์ด๋””์–ด: ๋งต์—์„œ ์ ‘๊ทผ ๊ฐ€๋Šฅํ•œ ์˜์—ญ(ํฐ์ƒ‰ ํ”ฝ์…€)์„ ๋ฐ˜๋ณต์ ์œผ๋กœ ์นจ์‹์‹œ์ผœ, ์—ฐ๊ฒฐ๋œ ์˜์—ญ์„ ๋ถ„๋ฆฌํ•ฉ๋‹ˆ๋‹ค.
๋‹จ๊ณ„:
  1. ์ดˆ๊ธฐ ์ƒํƒœ: ์žฅ์• ๋ฌผ(๋ฒฝ)์€ ๊ฒ€์€์ƒ‰, ์ ‘๊ทผ ๊ฐ€๋Šฅํ•œ ๊ณต๊ฐ„์€ ํฐ์ƒ‰์œผ๋กœ ํ‘œ์‹œ๋œ ์ด์ง„ํ™”๋œ ๋งต ์ƒ์„ฑ.
  2. ์นจ์‹ ์—ฐ์‚ฐ: ์ ‘๊ทผ ๊ฐ€๋Šฅํ•œ ๊ณต๊ฐ„์„ ์ ์ง„์ ์œผ๋กœ ์ค„์ด๋ฉฐ(ํ”ฝ์…€ ์ œ๊ฑฐ), ์—ฐ๊ฒฐ๋œ ๊ณต๊ฐ„์„ ๋ถ„๋ฆฌ.
  3. ์˜์—ญ ๋ถ„์„: ํŠน์ • ํฌ๊ธฐ์˜ ๋ถ„๋ฆฌ๋œ ์˜์—ญ์„ ๊ฐœ๋ณ„ ๋ฐฉ์œผ๋กœ ๋ผ๋ฒจ๋ง.
  4. ํŒŒํ˜• ์ „ํŒŒ: ๋ผ๋ฒจ์ด ์—†๋Š” ์˜์—ญ์„ ํ™•์žฅํ•˜์—ฌ ๋ชจ๋“  ๊ณต๊ฐ„์— ๋ผ๋ฒจ ํ• ๋‹น.
ํŠน์ง•
  • ์žฅ์ : ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๊ฐ„๋‹จํ•˜๊ณ  ๊ณ„์‚ฐ ์†๋„๊ฐ€ ๋น ๋ฆ„๋ฆ„.
  • ๋‹จ์ : ์žฅ์• ๋ฌผ(๊ฐ€๊ตฌ ๋“ฑ)์ด ๋งŽ์„ ๊ฒฝ์šฐ ๋ฐฉ ๊ฒฝ๊ณ„๊ฐ€ ์™œ๊ณก๋˜๊ฑฐ๋‚˜ ์ •ํ™•๋„๊ฐ€ ๋‚ฎ์•„์งˆ ์ˆ˜ ์žˆ์Œ.

2. ๊ฑฐ๋ฆฌ ๋ณ€ํ™˜ ๊ธฐ๋ฐ˜ ๋ถ„ํ•  (Distance Transform-based Segmentation)

์ž‘๋™ ์›๋ฆฌ
  • ๊ธฐ๋ณธ ์•„์ด๋””์–ด: ์ ‘๊ทผ ๊ฐ€๋Šฅํ•œ ๊ฐ ํ”ฝ์…€์—์„œ ๊ฐ€์žฅ ๊ฐ€๊นŒ์šด ์žฅ์• ๋ฌผ๊นŒ์ง€์˜ ๊ฑฐ๋ฆฌ๋ฅผ ๊ณ„์‚ฐํ•ด, ๋ฐฉ์˜ ์ค‘์‹ฌ์„ ์ฐพ์Šต๋‹ˆ๋‹ค.
๋‹จ๊ณ„:
  1. ๊ฑฐ๋ฆฌ ๋ณ€ํ™˜ ๋งต ์ƒ์„ฑ: ๋งต์˜ ๊ฐ ํ”ฝ์…€์— ๊ฐ€์žฅ ๊ฐ€๊นŒ์šด ์žฅ์• ๋ฌผ๊นŒ์ง€์˜ ๊ฑฐ๋ฆฌ๋ฅผ ํ‘œ์‹œ.
  2. ์ง€์—ญ ์ตœ๋Œ€๊ฐ’ ๊ฒ€์ƒ‰: ๊ฑฐ๋ฆฌ ๋ณ€ํ™˜ ๋งต์—์„œ ๊ฐ€์žฅ ๋†’์€ ๊ฐ’(๋ฐฉ ์ค‘์‹ฌ) ํƒ์ง€.
  3. ์ตœ์  ์ž„๊ณ„๊ฐ’ ์„ ํƒ: ๋ฐฉ ์ค‘์‹ฌ์˜ ๊ฐœ์ˆ˜๋ฅผ ์ตœ๋Œ€ํ™”ํ•˜๋Š” ์ž„๊ณ„๊ฐ’ ๊ฒฐ์ •.
  4. ํŒŒํ˜• ์ „ํŒŒ: ์ค‘์‹ฌ์—์„œ ์ฃผ๋ณ€์œผ๋กœ ํ™•์žฅํ•˜๋ฉฐ ์˜์—ญ์„ ๋ผ๋ฒจ๋ง.
ํŠน์ง•
  • ์žฅ์ : ํ˜•ํƒœํ•™์  ๋ถ„ํ• ๊ณผ ์œ ์‚ฌํ•˜์ง€๋งŒ ๋ฐฉ ์ค‘์‹ฌ์„ ํšจ๊ณผ์ ์œผ๋กœ ํƒ์ง€.
  • ๋‹จ์ : ์žฅ์• ๋ฌผ๋กœ ์ธํ•ด ๋ฐฉ ์ค‘์‹ฌ์˜ ํƒ์ง€๊ฐ€ ์™œ๊ณก๋  ์ˆ˜ ์žˆ์Œ.

3. ๋ณด๋กœ๋…ธ์ด ๊ทธ๋ž˜ํ”„ ๊ธฐ๋ฐ˜ ๋ถ„ํ•  (Voronoi Graph-based Segmentation)

์ž‘๋™ ์›๋ฆฌ
  • ๊ธฐ๋ณธ ์•„์ด๋””์–ด: ๋งต์—์„œ ๋ณด๋กœ๋…ธ์ด ๊ทธ๋ž˜ํ”„๋ฅผ ์ƒ์„ฑํ•ด ๋ฐฉ ๊ฒฝ๊ณ„๋ฅผ ์ •์˜ํ•ฉ๋‹ˆ๋‹ค.
๋‹จ๊ณ„:
  1. ๋ณด๋กœ๋…ธ์ด ๊ทธ๋ž˜ํ”„ ์ƒ์„ฑ: ์žฅ์• ๋ฌผ๋กœ๋ถ€ํ„ฐ์˜ ๋“ฑ๊ฑฐ๋ฆฌ์„ ์„ ๋”ฐ๋ผ ๊ทธ๋ž˜ํ”„๋ฅผ ์ƒ์„ฑ.
  2. Critical Points ํƒ์ง€: ๋ฌธ์ฒ˜๋Ÿผ ์ข์€ ํ†ต๋กœ๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š” ์ ์„ ํƒ์ง€.
  3. Critical Lines ์ƒ์„ฑ: ์ค‘์š” ์ ์„ ์—ฐ๊ฒฐํ•ด ์ดˆ๊ธฐ ๋ฐฉ ๊ฒฝ๊ณ„ ์ƒ์„ฑ.
  4. ์˜์—ญ ๋ณ‘ํ•ฉ: ์ž‘์€ ์˜์—ญ์ด๋‚˜ ๋น„์ •ํ˜•์ ์ธ ์˜์—ญ์„ ์—ฌ๋Ÿฌ ๊ธฐ์ค€(๋ฉด์ , ๊ฒฝ๊ณ„ ๊ธธ์ด ๋“ฑ)์— ๋”ฐ๋ผ ๋ณ‘ํ•ฉ.
ํŠน์ง•
  • ์žฅ์ : ๋ฐฉ์˜ ๊ตฌ์กฐ๋ฅผ ์ž˜ ๋ณด์กดํ•˜๋ฉฐ, ๋†’์€ ์ •๋ฐ€๋„๋กœ ๋ถ„ํ• .
  • ๋‹จ์ : ๊ณ„์‚ฐ๋Ÿ‰์ด ๋‹ค๋ฅธ ์•Œ๊ณ ๋ฆฌ์ฆ˜๋ณด๋‹ค ๋งŽ์œผ๋ฉฐ, ์„ธ๋ถ€ ๋ณ‘ํ•ฉ ๊ทœ์น™์— ๋”ฐ๋ผ ๊ฒฐ๊ณผ๊ฐ€ ๋‹ฌ๋ผ์งˆ ์ˆ˜ ์žˆ์Œ.

4. ํŠน์ง• ๊ธฐ๋ฐ˜ ๋ถ„ํ•  (Feature-based Segmentation)

์ž‘๋™ ์›๋ฆฌ
  • ๊ธฐ๋ณธ ์•„์ด๋””์–ด: ๋ ˆ์ด์ € ์Šค์บ๋„ˆ๋กœ ์–ป์€ ๋ฐ์ดํ„ฐ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ, ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ(AdaBoost)์„ ์‚ฌ์šฉํ•ด ๊ณต๊ฐ„์˜ ์ข…๋ฅ˜๋ฅผ ๋ผ๋ฒจ๋งํ•ฉ๋‹ˆ๋‹ค.
๋‹จ๊ณ„:
  1. ๋ ˆ์ด์ € ์Šค์บ” ๋ฐ์ดํ„ฐ ์ƒ์„ฑ: 360ยฐ ๋ ˆ์ด์ € ์Šค์บ๋„ˆ๋กœ ๋งต์˜ ๊ฐ ํ”ฝ์…€์—์„œ ๊ฑฐ๋ฆฌ ์ •๋ณด๋ฅผ ์ถ”์ถœ.
  2. ํŠน์ง• ์ถ”์ถœ: ๊ฑฐ๋ฆฌ ๋ฐ์ดํ„ฐ๋กœ๋ถ€ํ„ฐ ๊ฐ„๋‹จํ•œ ๊ธฐํ•˜ํ•™์  ํŠน์ง•(์˜ˆ: ํ‰๊ท  ๊ฑฐ๋ฆฌ, ๊ฐ€์žฅ ๊ธด ๋ ˆ์ด์ € ๊ฑฐ๋ฆฌ ๋“ฑ) 33๊ฐ€์ง€๋ฅผ ๊ณ„์‚ฐ.
  3. AdaBoost ๋ถ„๋ฅ˜: ๊ฐ ํ”ฝ์…€์„ ๋ฐฉ, ๋ณต๋„, ๋ฌธ ๋“ฑ์œผ๋กœ ๋ถ„๋ฅ˜.
  4. ์˜์—ญ ๋ณ‘ํ•ฉ: ๊ฐ™์€ ๋ผ๋ฒจ์„ ๊ฐ€์ง„ ์ด์›ƒ ํ”ฝ์…€ ๋ณ‘ํ•ฉ.
  5. ๋ผ๋ฒจ๋ง ์กฐ์ •: ๋งˆ์ฝ”ํ”„ ๋„คํŠธ์›Œํฌ๋ฅผ ์‚ฌ์šฉํ•ด ๊ฒฐ๊ณผ๋ฅผ ๋ถ€๋“œ๋Ÿฝ๊ฒŒ ์กฐ์ •.
ํŠน์ง•
  • ์žฅ์ : ํ™˜๊ฒฝ ๋ณ€ํ™”(๊ฐ€๊ตฌ ๋“ฑ)์— ๊ฐ•ํ•˜๊ณ  ์•ˆ์ •์ ์ž„.
  • ๋‹จ์ : ์ดˆ๊ธฐ ํ•™์Šต ๋ฐ์ดํ„ฐ๊ฐ€ ํ•„์š”ํ•˜๋ฉฐ, ๊ณ„์‚ฐ ์‹œ๊ฐ„์ด ๋งค์šฐ ๊ธธ์–ด์งˆ ์ˆ˜ ์žˆ์Œ.

ํ‰๊ฐ€ ์ง€ํ‘œ

1. ์ผ๋ฐ˜ ์ˆ˜์น˜ ์†์„ฑ

๋‹ค์–‘ํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์ „๋ฐ˜์ ์ธ ํŠน์„ฑ์„ ์ˆ˜์น˜์ ์œผ๋กœ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•œ ์ง€ํ‘œ์ž…๋‹ˆ๋‹ค.

1) ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์‹คํ–‰ ์‹œ๊ฐ„ (Runtime)
  • ์ •์˜: ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ์˜์—ญ(๋ฐฉ) ๋ถ„ํ• ์„ ์™„๋ฃŒํ•˜๋Š” ๋ฐ ๊ฑธ๋ฆฌ๋Š” ์‹œ๊ฐ„(์ดˆ).
  • ์˜๋ฏธ: ๊ณ„์‚ฐ ์†๋„๊ฐ€ ๋น ๋ฅผ์ˆ˜๋ก ์‹ค์‹œ๊ฐ„ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์— ์œ ๋ฆฌํ•จ.
2) ์„ธ๊ทธ๋จผํŠธ ์ˆ˜ (Number of Segments)
  • ์ •์˜: ๋ถ„ํ•  ํ›„ ์ƒ์„ฑ๋œ ์˜์—ญ(๋ฐฉ)์˜ ์ด ๊ฐœ์ˆ˜.
  • ์˜๋ฏธ: ๋„ˆ๋ฌด ๋งŽ์€ ์„ธ๊ทธ๋จผํŠธ๋Š” ๊ณผ๋ถ„ํ• (over-segmentation)์„ ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ๊ณ , ๋„ˆ๋ฌด ์ ์œผ๋ฉด ๊ณผ์†Œ๋ถ„ํ• (under-segmentation)์„ ์˜๋ฏธํ•  ์ˆ˜ ์žˆ์Œ.
3) ์„ธ๊ทธ๋จผํŠธ ๋ฉด์  (Segment Area)
  • ์ •์˜: ๊ฐ ์˜์—ญ(๋ฐฉ)์˜ ํ‰๊ท  ๋ฉด์ (์ œ๊ณฑ๋ฏธํ„ฐ): \(A_i\)
  • ์˜๋ฏธ: ๋ฐฉ์˜ ํฌ๊ธฐ์™€ ๊ท ์ผ์„ฑ์„ ํ‰๊ฐ€.
4) ์„ธ๊ทธ๋จผํŠธ ๋‘˜๋ ˆ (Segment Perimeter)
  • ์ •์˜: ๊ฐ ์˜์—ญ(๋ฐฉ)์˜ ํ‰๊ท  ๋‘˜๋ ˆ(๋ฏธํ„ฐ): \(u_i\)
  • ์˜๋ฏธ: ๋ฐฉ ๊ฒฝ๊ณ„์˜ ๋ณต์žก์„ฑ์„ ๋‚˜ํƒ€๋ƒ„.
5) A-์ปดํŒฉํŠธ๋‹ˆ์Šค (A-Compactness)
  • ์ •์˜: ์˜์—ญ(๋ฐฉ)์˜ ๋ฉด์ ๊ณผ ๋‘˜๋ ˆ์˜ ๋น„์œจ๋กœ ๊ณ„์‚ฐ:

    \[\mathrm{\text{A-Compactness}} = {A_i \over {u_i}^2}\]
  • ์˜๋ฏธ: ๊ฐ’์ด ํด์ˆ˜๋ก ์˜์—ญ(๋ฐฉ)์ด ๋” ์ •ํ˜•์ ์ด๊ณ  ์ปดํŒฉํŠธํ•จ.

6) B-์ปดํŒฉํŠธ๋‹ˆ์Šค (B-Compactness)
  • ์ •์˜: ์˜์—ญ(๋ฐฉ) ๋ฉด์ ์„ ์ตœ์†Œ ์™ธ์ ‘ ์ง์‚ฌ๊ฐํ˜•์˜ ๋ฉด์ ์œผ๋กœ ๋‚˜๋ˆˆ ๊ฐ’:

    \[\mathrm{\text{B-Compactness}} = {A_i \over A_{bb,i}}\]
  • ์˜๋ฏธ: ๊ฐ’์ด ํด์ˆ˜๋ก ์˜์—ญ(๋ฐฉ)์ด ์ง์‚ฌ๊ฐํ˜•์— ๊ฐ€๊นŒ์›€์„ ๋‚˜ํƒ€๋ƒ„.

7) ํ˜•ํƒœ ์ง€ํ‘œ (Shape)
  • ์ •์˜: ์ฃผ์„ฑ๋ถ„๋ถ„์„(PCA)์„ ์ด์šฉํ•ด ์˜์—ญ(๋ฐฉ)์˜ ์ฃผ์ถ•๊ณผ ๋ถ€์ถ•์˜ ๊ณ ์œ ๊ฐ’ ๋น„์œจ๋กœ ๊ณ„์‚ฐ:

    \[\mathrm{\text{Shape}} = {e_{i,1} \over e_{i,2}}\]
  • ์˜๋ฏธ: ๊ฐ’์ด 1์— ๊ฐ€๊นŒ์šฐ๋ฉด ์ •์‚ฌ๊ฐํ˜•์ด๋‚˜ ์›ํ˜•์— ๊ฐ€๊นŒ์›€์„ ์˜๋ฏธํ•˜๊ณ , ๊ฐ’์ด ํฌ๋ฉด ๊ธธ์ญ‰ํ•œ ํ˜•ํƒœ๋ฅผ ๋‚˜ํƒ€๋ƒ„.

2. ํ’ˆ์งˆ ํ‰๊ฐ€ (Quality of Room Segmentation)

๊ฐ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ์‚ฌ๋žŒ์ด ๋ผ๋ฒจ๋งํ•œ Grouhd truth์™€ ์–ผ๋งˆ๋‚˜ ์ž˜ ์ผ์น˜ํ•˜๋Š”์ง€ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค.

1) ์žฌํ˜„์œจ (Recall)
  • ์ •์˜: Ground truth ์˜์—ญ(๋ฐฉ)์˜ ํ”ฝ์…€ ์ค‘ ๋ถ„ํ•  ๊ฒฐ๊ณผ์™€ ๊ฒน์น˜๋Š” ํ”ฝ์…€ ๋น„์œจ:

    \[\mathrm{Recall} = {\text{segmented room pixel}\,\cap\,\text{ground truth room pixel} \over \text{ground truth room pixel}}\]
  • ์˜๋ฏธ: Ground truth ์˜์—ญ(๋ฐฉ)์ด ๊ฒฐ๊ณผ ์˜์—ญ(๋ฐฉ)์— ์ž˜ ํฌํ•จ๋˜์–ด ์žˆ๋Š”์ง€ ํ‰๊ฐ€. ๋†’์„์ˆ˜๋ก Ground truth ์˜์—ญ(๋ฐฉ)์ด ๋ˆ„๋ฝ๋˜์ง€ ์•Š์Œ์„ ์˜๋ฏธ.

2) ์ •๋ฐ€๋„ (Precision)
  • ์ •์˜: ๋ถ„ํ• ๋œ ์˜์—ญ(๋ฐฉ)์˜ ํ”ฝ์…€ ์ค‘ Ground truth ์˜์—ญ(๋ฐฉ)๊ณผ ๊ฒน์น˜๋Š” ํ”ฝ์…€ ๋น„์œจ:

    \[\mathrm{Precision} = {\text{segmented room pixel}\,\cap\,\text{ground truth room pixel} \over \text{segmented room pixel}}\]
  • ์˜๋ฏธ: ๊ฒฐ๊ณผ ์˜์—ญ(๋ฐฉ)์ด Ground truth ์˜์—ญ(๋ฐฉ) ์•ˆ์— ์–ผ๋งˆ๋‚˜ ์ •ํ™•ํžˆ ๋“ค์–ด๋งž๋Š”์ง€ ํ‰๊ฐ€. ๋†’์„์ˆ˜๋ก ๊ณผ๋ถ„ํ• ์ด ์ ์Œ์„ ์˜๋ฏธ.

3) F1 Score
  • ์ •์˜: ์žฌํ˜„์œจ๊ณผ ์ •๋ฐ€๋„์˜ ์กฐํ™” ํ‰๊ท :

    \[\mathrm{\text{F1 Score}} = 2 \times {\text{Precison}\,\times\,\text{Recall} \over \text{Precsion}\,+\,\text{Recall}}\]
  • ์˜๋ฏธ: ์ •๋ฐ€๋„์™€ ์žฌํ˜„์œจ ๊ฐ„์˜ ๊ท ํ˜•์„ ํ‰๊ฐ€.

3. ์‘์šฉ ์„ฑ๋Šฅ ํ‰๊ฐ€ (Performance in a Cleaning Robot)

๋กœ๋ด‡์˜ ์ž‘์—… ํšจ์œจ์„ฑ์„ ์ธก์ •ํ•˜๊ธฐ ์œ„ํ•ด Room Segmentation ๊ฒฐ๊ณผ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋กœ๋ด‡ ์ฒญ์†Œ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์„ ์ง„ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค.

1) ์ฒญ์†Œ ์‹œ๊ฐ„ (Cleaning Time)
  • ์ •์˜: ๋กœ๋ด‡์ด ๋ชจ๋“  ๋ฐฉ์„ ๋ฐฉ๋ฌธํ•˜๊ณ  ์ž‘์—…์„ ์™„๋ฃŒํ•˜๋Š” ๋ฐ ๊ฑธ๋ฆฌ๋Š” ์ด ์‹œ๊ฐ„.
  • ์˜๋ฏธ: ๋” ์งง์€ ์ฒญ์†Œ ์‹œ๊ฐ„์ด ๋” ํšจ์œจ์ ์ธ ๋ฐฉ ๋ถ„ํ• ์„ ๋‚˜ํƒ€๋ƒ„.
2) ๊ฒฝ๋กœ ์ตœ์ ํ™” (Path Optimization)
  • ์ •์˜: ์˜์—ญ(๋ฐฉ) ๋ถ„ํ•  ๊ฒฐ๊ณผ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ์ตœ์  ๊ฒฝ๋กœ์˜ ๊ณ„์‚ฐ ์—ฌ๋ถ€(์˜ˆ: ์—ฌํ–‰ ํŒ๋งค์› ๋ฌธ์ œ ํ•ด๊ฒฐ).
  • ์˜๋ฏธ: ๋” ์ž˜ ๋ถ„ํ• ๋œ ๋งต์€ ๊ฒฝ๋กœ ์ตœ์ ํ™”๋ฅผ ๋” ์‰ฝ๊ฒŒ ๋งŒ๋“ฆ.
3) ๋„๊ตฌ ์ด๋™ ํšŸ์ˆ˜ (Tool Placement Moves)
  • ์ •์˜: ๋กœ๋ด‡์ด ์ฒญ์†Œ ๋„๊ตฌ๋ฅผ ์ด๋™์‹œํ‚จ ํšŸ์ˆ˜.
  • ์˜๋ฏธ: ๋„๊ตฌ ์ด๋™์ด ์ ์„์ˆ˜๋ก ์˜์—ญ(๋ฐฉ) ๋ถ„ํ• ์ด ์ž˜ ์ด๋ฃจ์–ด์กŒ์Œ์„ ์˜๋ฏธ.

๊ฒฐ๊ณผ ์š”์•ฝ

์•Œ๊ณ ๋ฆฌ์ฆ˜์†๋„์ •ํ™•๋„์žฅ์• ๋ฌผ์— ๋Œ€ํ•œ ์•ˆ์ •์„ฑ๊ณ„์‚ฐ ๋ณต์žก๋„
ํ˜•ํƒœํ•™์  ๋ถ„ํ• ๋น ๋ฆ„๋ณดํ†ต๋‚ฎ์Œ๋‚ฎ์Œ
๊ฑฐ๋ฆฌ ๋ณ€ํ™˜ ๊ธฐ๋ฐ˜ ๋ถ„ํ• ๋น ๋ฆ„๋ณดํ†ต๋‚ฎ์Œ๋‚ฎ์Œ
๋ณด๋กœ๋…ธ์ด ๊ธฐ๋ฐ˜ ๋ถ„ํ• ์ค‘๊ฐ„๋†’์Œ๋ณดํ†ต์ค‘๊ฐ„
ํŠน์ง• ๊ธฐ๋ฐ˜ ๋ถ„ํ• ๋Š๋ฆผ๋†’์Œ๋†’์Œ๋†’์Œ

1. ์ผ๋ฐ˜ ์†์„ฑ

  • ํ˜•ํƒœํ•™์  ๋ฐ ๊ฑฐ๋ฆฌ ๋ณ€ํ™˜ ๊ธฐ๋ฐ˜: ๋น ๋ฅด์ง€๋งŒ ์žฅ์• ๋ฌผ์— ๋ฏผ๊ฐํ•˜์—ฌ ๋ณต์žกํ•œ ํ™˜๊ฒฝ์—์„œ๋Š” ์ •ํ™•๋„๊ฐ€ ๋‚ฎ์•„์งˆ ์ˆ˜ ์žˆ์Œ.
  • ๋ณด๋กœ๋…ธ์ด ๊ธฐ๋ฐ˜: ์ ๋‹นํ•œ ์‹คํ–‰ ์‹œ๊ฐ„๊ณผ ๋†’์€ ์ •ํ™•๋„๋กœ ์‹ค์šฉ์ ์ธ ์„ ํƒ.
  • ํŠน์ง• ๊ธฐ๋ฐ˜: ๊ฐ€์žฅ ์•ˆ์ •์ ์ด์ง€๋งŒ ๊ณ„์‚ฐ ๋น„์šฉ๊ณผ ์‹คํ–‰ ์‹œ๊ฐ„์ด ๋†’์Œ.

2. ํ’ˆ์งˆ ์ง€ํ‘œ

  • ์ตœ๊ณ  ์žฌํ˜„์œจ: ํ˜•ํƒœํ•™์  ๋ฐ ๊ฑฐ๋ฆฌ ๋ณ€ํ™˜ ๊ธฐ๋ฐ˜ (๋‹จ์ˆœํ•œ ํ™˜๊ฒฝ์—์„œ ์šฐ์ˆ˜).
  • ์ตœ๊ณ  ์ •๋ฐ€๋„: ๋ณด๋กœ๋…ธ์ด ๊ธฐ๋ฐ˜ ๋ถ„ํ•  (ํ˜„์‹ค์ ์ธ ํ™˜๊ฒฝ์—์„œ๋„ ๋†’์€ ์ •ํ™•๋„).
  • ๊ท ํ˜• ์žกํžŒ ์„ฑ๋Šฅ: ๋ณด๋กœ๋…ธ์ด ๊ธฐ๋ฐ˜ (์žฌํ˜„์œจ๊ณผ ์ •๋ฐ€๋„ ๋ชจ๋‘์—์„œ ์•ˆ์ •์ ).

3. ์ฒญ์†Œ ๋กœ๋ด‡ ์„ฑ๋Šฅ

  • ์ตœ๋‹จ ์ฒญ์†Œ ์‹œ๊ฐ„: ๋ณด๋กœ๋…ธ์ด ๊ธฐ๋ฐ˜ ๋ถ„ํ•  (์••์ถ•๋œ ํด๋Ÿฌ์Šคํ„ฐ ๋•๋ถ„์— ํšจ์œจ์ ).
  • ์ตœ์žฅ ์ฒญ์†Œ ์‹œ๊ฐ„: ํŠน์ง• ๊ธฐ๋ฐ˜ ๋ถ„ํ•  (๋” ๊ธด ๊ฒฝ๋กœ์™€ ๋†’์€ ๊ณ„์‚ฐ ์‹œ๊ฐ„).

๊ฒฐ๋ก 

์ด ๋…ผ๋ฌธ์€ Room Segmentation ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์„ ์ข…ํ•ฉ์ ์œผ๋กœ ๋ถ„์„ํ•˜๋ฉฐ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ๊ฒฐ๋ก ์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.

  • ํ˜•ํƒœํ•™์  ๋ฐ ๊ฑฐ๋ฆฌ ๋ณ€ํ™˜ ๊ธฐ๋ฐ˜: ๊ณ„์‚ฐ ์†๋„๋Š” ๋น ๋ฅด์ง€๋งŒ, ์žฅ์• ๋ฌผ์ด ๋งŽ์„ ๋•Œ ์˜์—ญ(๋ฐฉ) ๊ฒฝ๊ณ„๋ฅผ ์ •ํ™•ํžˆ ๊ตฌ๋ถ„ํ•˜์ง€ ๋ชปํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์žˆ์Œ. ์ •๋ฐ€๋„๋Š” ๋‚ฎ์ง€๋งŒ ์žฌํ˜„์œจ์ด ๋†’์•„ ๊ณผ์†Œ๋ถ„ํ• ๋ณด๋‹ค๋Š” ๊ณผ๋ถ„ํ•  ๊ฒฝํ–ฅ. ๋‹จ์ˆœํ•œ ํ™˜๊ฒฝ์— ์ ํ•ฉ.
  • ๋ณด๋กœ๋…ธ์ด ๊ธฐ๋ฐ˜: ์žฌํ˜„์œจ๊ณผ ์ •๋ฐ€๋„ ๋ชจ๋‘ ๋†’์œผ๋ฉฐ, ์˜์—ญ(๋ฐฉ) ๊ตฌ์กฐ๋ฅผ ์ž˜ ๋ฐ˜์˜. ํ˜„์‹ค์ ์ธ ์‘์šฉ์—์„œ ์ตœ์ ์˜ ์„ฑ๋Šฅ์„ ์ œ๊ณตํ•˜๋ฉฐ, ์ •ํ™•๋„์™€ ์‹คํ–‰ ์‹œ๊ฐ„์˜ ๊ท ํ˜•์ด ์šฐ์ˆ˜ํ•จ.
  • ํŠน์ง• ๊ธฐ๋ฐ˜: ๊ฐ€์žฅ ์•ˆ์ •์ ์ด๋ฉฐ, ์žฅ์• ๋ฌผ์— ๋œ ๋ฏผ๊ฐ. ์ ์ ˆํ•œ ํ•™์Šต ๋ฐ์ดํ„ฐ์™€ ์ถฉ๋ถ„ํ•œ ๊ณ„์‚ฐ ์ž์›์ด ์žˆ์„ ๋•Œ ๋ณต์žกํ•˜๊ณ  ๋™์ ์ธ ํ™˜๊ฒฝ์—์„œ ์ด์ƒ์ .

์ถ”๊ฐ€ ์—ฐ๊ตฌ ๊ณ„ํš

์ €์ž๋“ค์€ ์•ž์œผ๋กœ ๋ณด๋กœ๋…ธ์ด ๊ธฐ๋ฐ˜๊ณผ ํŠน์ง• ๊ธฐ๋ฐ˜ ์ ‘๊ทผ๋ฒ•์„ ํ†ตํ•ฉํ•˜์—ฌ ์•ˆ์ •์„ฑ๊ณผ ์ •ํ™•์„ฑ์„ ๋”์šฑ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๋ฐฉ๋ฒ•์„ ์—ฐ๊ตฌํ•  ๊ณ„ํš์ด๋ผ๊ณ  ํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๋‹ค์–‘ํ•œ ์‘์šฉ ํ™˜๊ฒฝ์—์„œ ๋กœ๋ด‡์˜ ์„ฑ๋Šฅ์„ ๊ทน๋Œ€ํ™”ํ•  ์ˆ˜ ์žˆ์„ ๊ฒƒ์œผ๋กœ ๊ธฐ๋Œ€๋œ๋‹ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค.

This post is licensed under CC BY 4.0 by the author.

ยฉ sirius-mhlee. Some rights reserved.

Using the Chirpy theme for Jekyll.