• ๋Œ€ํ•œ์ „๊ธฐํ•™ํšŒ
Mobile QR Code QR CODE : The Transactions of the Korean Institute of Electrical Engineers
  • COPE
  • kcse
  • ํ•œ๊ตญ๊ณผํ•™๊ธฐ์ˆ ๋‹จ์ฒด์ด์—ฐํ•ฉํšŒ
  • ํ•œ๊ตญํ•™์ˆ ์ง€์ธ์šฉ์ƒ‰์ธ
  • Scopus
  • crossref
  • orcid

  1. (School of Electrical and Electronic Engineering, Yonsei University, Korea.)



Local Visual Navigation, HaarยญLike Features, HOG, SURF, Feature Matching, Snapshot Model, ALV

1. ์„œ ๋ก 

๋น„๊ต์  ๋‹จ์ˆœํ•œ ๋ฉ”์ปค๋‹ˆ์ฆ˜์„ ์ง€๋‹Œ ๋™๋ฌผ์ด๋‚˜ ๊ณค์ถฉ๋“ค๋„, ๋ณด์œ ํ•˜๊ณ  ์žˆ๋Š” ๋‹ค์–‘ํ•œ ๊ฐ๊ฐ๊ธฐ๊ด€์„ ํ™œ์šฉํ•˜์—ฌ, ๊ฑฐ์ฃผ์ง€๋‚˜ ๋จน์ด๊ฐ€ ์žˆ๋Š” ์žฅ์†Œ ๋“ฑ์˜ ํŠน์ •์žฅ์†Œ๋กœ ์ด๋™ํ•  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์ด ์žˆ์Œ์€ ๋‹ค์–‘ํ•œ ์—ฐ๊ตฌ๋ฅผ ํ†ตํ•ด ํ™•์ธ๋˜์—ˆ๋‹ค. ์ด๋“ค์˜ ์ด๋Ÿฌํ•œ ๋Šฅ๋ ฅ์€ ์‹œ๊ฐ์  ์‹ ํ˜ธ๋ฟ๋งŒ์ด ์•„๋‹Œ ๋•Œ๋•Œ๋กœ ํ›„๊ฐ์ด๋‚˜ ์ฒญ๊ฐ์„ ์ด์šฉํ•˜์—ฌ ์›ํ•˜๋Š” ์œ„์น˜์— ๋Œ€ํ•œ ์ •๋ณด๋ฅผ ๊ธฐ์–ตํ•˜๋Š” ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•˜๋Š”๋ฐ [1-4], ์ด๋“ค๊ณผ ๋‹ฌ๋ฆฌ ์‚ฌ๋ง‰ ๊ฐœ๋ฏธ๋ฅผ ๋น„๋กฏํ•œ ์ผ๋ถ€ ๋™๋ฌผ๋“ค์€ ๋จน์ด๋ฅผ ์ฐพ์€ ํ›„์— ๋Œ์•„์˜ฌ ๋•Œ ์ตœ๋‹จ ๊ฒฝ๋กœ๋ฅผ ์ด์šฉํ•˜์—ฌ ๊ทธ๋“ค์˜ ๋‘ฅ์ง€๋กœ ๋ณต๊ท€ํ•˜๋Š”๋ฐ ์ด๋ฅผ ์œ„ํ•ด ๊ฒฝ๋กœ ๋ˆ„์  ๊ณ„์‚ฐ ๋ฐฉ์‹(Path Integration)์„ ์‚ฌ์šฉํ•œ๋‹ค๊ณ  ์•Œ๋ ค์ ธ ์žˆ๋‹ค [5-7]. ๊ฒฝ๋กœ ๋ˆ„์  ๊ณ„์‚ฐ๋ฐฉ์‹์€ ์ด๋™ํ•˜๋Š” ๊ฒฝ๋กœ์˜ ๋ฐฉํ–ฅ๊ณผ ๊ฑฐ๋ฆฌ๋ฅผ ๋‚ด๋ถ€์ ์œผ๋กœ ํ•ฉ์‚ฐํ•˜์—ฌ ์ถœ๋ฐœ์œ„์น˜์™€์˜ ์ƒ๋Œ€์  ์œ„์น˜๋ฅผ ๊ณ„์‚ฐ(์ถ”์ •)ํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค.

์ด๋Ÿฌํ•œ ๋ฐฉ์‹(๊ฒฝ๋กœ ๋ˆ„์  ๊ณ„์‚ฐ)์€ ๋‚ด๋น„๊ฒŒ์ด์…˜ ์ธก๋ฉด์—์„œ ๋‘ ๊ฐ€์ง€ ๋ฌธ์ œ์ ์ด ์žˆ๋Š”๋ฐ ์ด๋™ ๊ฐ„์— ์ ˆ๋Œ€์ขŒํ‘œ๊ณ„์—์„œ ์ธ์ง€๊ฐ€๋Šฅํ•œ ๋ฐฉ์œ„์ •๋ณด๊ฐ€ ์žˆ์–ด์•ผ ํ•˜๋ฉฐ, ์ด๋™์ด ์ง€์†๋ ์ˆ˜๋ก ๋ฐœ์ƒํ•˜๋Š” ์˜ค์ฐจ๊ฐ’์ด ์ถ•์ ๋œ๋‹ค๋Š” ์ ์ด๋‹ค. ๋ณธ ๋ฌธ์ œ์ ์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด ๊ฒฝ๋กœ ๋ˆ„์  ๊ณ„์‚ฐ ๋ฐฉ์‹์„ ํ™œ์šฉํ•˜๋Š” ๋™๋ฌผ(๊ณค์ถฉ)๋“ค์€ ์‹œ๊ฐ ๋žœ๋“œ๋งˆํฌ ๋“ฑ์˜ ์™ธ๋ถ€์ •๋ณด๋ฅผ ์ด์šฉํ•˜๋Š” ๊ฒƒ์œผ๋กœ ์•Œ๋ ค์ ธ ์žˆ๋‹ค.

์ผ๋ถ€ ๋ฒŒ์€ ๋จน์ด๋ฅผ ์ฐพ๊ฑฐ๋‚˜ ์ง‘์œผ๋กœ ๋Œ์•„๊ฐ€๋Š”๋ฐ ํƒœ์–‘๋น›์˜ ํŽธ๊ด‘ํŒจํ„ด์„ ์ฐธ์กฐํ•˜์—ฌ ์ด๋™ํ•œ๋‹ค [8]. ๋‘๋”์ง€๋Š” ๋•…๋ฐ‘์—์„œ๋„ ์ง€๊ตฌ์ž๊ธฐ์žฅ ์ •๋ณด๋ฅผ ๊ฒฝ๋กœ ๋ˆ„์  ๊ณ„์‚ฐ ๋ฐฉ์‹์˜ ์ •๋ณด์™€ ์กฐํ•ฉํ•˜์—ฌ ์‹œ๊ฐ ๋žœ๋“œ๋งˆํฌ๋ฅผ ํŠน์ •ํ•˜๊ธฐ ์–ด๋ ค์šด ํ™˜๊ฒฝ์—์„œ๋„ ๊ธธ์„ ์žƒ์ง€ ์•Š๊ณ  ์›ํ•˜๋Š” ์œ„์น˜๋‚˜ ์ง‘์œผ๋กœ ์ด๋™ํ•œ๋‹ค [9]. ์ผ๋ถ€ ์œ ๋Ÿฝ๊ฐœ๋ฏธ(Wood Ants)๋Š” ์‹œ๊ฐ ๋žœ๋“œ๋งˆํฌ๋ฅผ ์ง‘์œผ๋กœ ์ด๋™ํ•˜๋Š”๋ฐ ๋ณด์กฐ์ ์œผ๋กœ ํ™œ์šฉํ•˜๋ฉฐ [10], ์‹œ๊ฐ์ •๋ณด์™€ ํƒœ์–‘๋น›์˜ ํŽธ๊ด‘์ •๋ณด๋ฅผ ์กฐํ•ฉํ•œ ํ™ˆ ๋„ค๋น„๊ฒŒ์ด์…˜ ์‹œ์Šคํ…œ์„ ๊ฐ–๊ณ  ์žˆ๋Š” ๊ฐœ์ฒด๋„ ์กด์žฌํ•œ๋‹ค [11,12].

๋งŽ์€ ๋™๋ฌผ์€ ์ข…์˜ ์ƒ์กด์„ ์œ„ํ•ด ํ™˜๊ฒฝ์— ๋”ฐ๋ผ ๊ฐ€์šฉํ•œ ๋Šฅ๋ ฅ์„ ๊ฐ–์ถ”๊ณ  ์žˆ์œผ๋ฉฐ, ๋น„๊ต์  ๋‹จ์ˆœํ•œ ๊ฐ๊ฐ๊ธฐ๊ด€ ๋ฐ ์ธ์ง€์ฒด๊ณ„๋งŒ์œผ๋กœ๋„ ์ƒ์กด์„ ์œ„ํ•ด ์ตœ์ ํ™” ๋˜์–ด์žˆ์Œ์„ ๋ณด์—ฌ์ค€๋‹ค[1,3,4,6,7,13,14]. ์ƒ์กด์„ ์œ„ํ•œ ์ง„ํ™”์™€ ๋Šฅ๋ ฅ์€ ์ธ๊ฐ„์—๊ฒŒ๋„ ์ ์šฉ ๊ฐ€๋Šฅํ•˜์ง€๋งŒ ์ธ๊ฐ„์˜ ๋ณต์žกํ•œ ๊ณต๊ฐ„์ธ์ง€ ๋ฐ ๋‚ด๋น„๊ฒŒ์ด์…˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ดํ•ดํ•˜๊ธฐ๋Š” ์‰ฝ์ง€ ์•Š๋‹ค.

์ธ๊ฐ„์˜ ๋ณต์žกํ•œ ์ธ์ง€ ๋ฐ ์‚ฌ๊ณ  ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ธ๊ณต์ง€๋Šฅ์œผ๋กœ ๊ตฌํ˜„ํ•˜๊ธฐ์—๋Š” ํ•œ๊ณ„๊ฐ€ ์žˆ๊ธฐ ๋•Œ๋ฌธ์—, ๋™๋ฌผ๋“ค์˜ ํ–‰๋™ํŒจํ„ด์„ ๋ชจ๋ฐฉํ•˜๋Š” ๋‹ค์–‘ํ•œ ๋ฐฉ์‹์˜ ๋‚ด๋น„๊ฒŒ์ด์…˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋Œ€ํ•ด ๋งŽ์€ ์—ฐ๊ตฌ๊ฐ€ ์ง„ํ–‰๋˜๊ณ  ์žˆ๋‹ค. ๋„“์€ ๊ตฌ์—ญ์—์„œ ๋‚ด๋น„๊ฒŒ์ด์…˜์„ ์ˆ˜ํ–‰ํ•˜๋Š” ๊ฒฝ์šฐ์—๋Š” ์œ„์—์„œ ์–ธ๊ธ‰ํ•œ ๋ฌธ์ œ์ ์œผ๋กœ ์ธํ•ด ๋‹ค์–‘ํ•œ ์„ผ์„œ๋ฅผ ํ†ตํ•ฉํ•˜์—ฌ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹์ผ ๋•Œ, ๋งŒ์กฑํ•  ๋งŒํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ผ ์ˆ˜ ์žˆ์ง€๋งŒ, ๊ตญ์ง€์ ์œผ๋Š” ์‹œ๊ฐ์„ผ์„œ๋งŒ์œผ๋กœ๋„ ์ถฉ๋ถ„ํžˆ ์›ํ•˜๋Š” ์œ„์น˜๋กœ ๋Œ์•„๊ฐ€๊ธฐ ์œ„ํ•œ ๋ฐฉํ–ฅ (ํ™ˆ๋ฒกํ„ฐ) ์‚ฐ์ถœ์ด ๊ฐ€๋Šฅํ•˜๋‹ค.

์ด๋Š” ๋น„๊ต์  ๋‹จ์ˆœํ•œ ์›๋ฆฌ๋กœ๋„ ๊ตฌํ˜„์ด ๊ฐ€๋Šฅํ•œ๋ฐ, ๋Œ€ํ‘œ์ ์ธ ๋ฐฉ์‹์ด ์Šค๋ƒ…์ƒท ๋ชจ๋ธ(Snapshot Model)์ด๋‹ค. ์ด ๋ชจ๋ธ์€ ์ฃผ๋กœ ๊ฟ€๋ฒŒ์˜ ๋ฐฉํ–ฅ์ถ”์ • ๋ฐฉ์‹์„ ๋ชจ๋ฐฉํ•œ ๋ฐฉ์‹์œผ๋กœ ๋‘ ์ง€์ (์ฒ˜์Œ๊ณผ ๋)์—์„œ ์ดฌ์˜ํ•œ ์ด๋ฏธ์ง€๋งŒ์„ ์ด์šฉํ•˜์—ฌ ๊ตญ์ง€์˜์—ญ์—์„œ์˜ ์ƒ๋Œ€์ ์ธ ์œ„์น˜๋ฅผ ์ถ”์ •ํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค. ์Šค๋ƒ…์ƒท ๋ชจ๋ธ์„ ์‹ค์ œ๋กœ ์ ์šฉํ•˜๋Š” ๋ฐฉ์‹์€ ๋งค์šฐ ๋‹ค์–‘ํ•˜์ง€๋งŒ ๋Œ€ํ‘œ์ ์œผ๋กœ ์‹œ๊ฐ์  ๋žœ๋“œ๋งˆํฌ๋ฅผ ์ด์šฉํ•˜์—ฌ ์œ„์น˜๋ฅผ ์ถ”์ •ํ•˜๋Š” ALV (Average Landmark Vector) ๋ฐฉ์‹์„ ๋“ค ์ˆ˜ ์žˆ๋‹ค [15]. ์œ„ ๋ฐฉ์‹์„ ๋น„๋กฏํ•œ ์Šค๋ƒ…์ƒท ๋ชจ๋ธ์€ ๋งŽ์€ ๊ฒฝ์šฐ ๋ฐฉ์œ„์ •๋ณด๋ฅผ ๊ฐ–๊ณ  ์žˆ์–ด์•ผ ํ•œ๋‹ค๋Š” ์ „์ œ๊ฐ€ ์žˆ๋‹ค. ๋™๋ฌผ์ด๋‚˜ ๊ณค์ถฉ์€ ์ฃผ๋กœ ํƒœ์–‘ ๋น›์˜ ํŽธ๊ด‘์ •๋ณด๋‚˜ ์ž๊ธฐ์žฅ ์ •๋ณด๋ฅผ ์ด์šฉํ•œ๋‹ค. ์Šค๋ƒ…์ƒท ๋ชจ๋ธ์€ ์‹œ๊ฐ์„ผ์„œ ์ด์™ธ์—๋„ ๊ฑฐ๋ฆฌ ์„ผ์„œ ์ด๋ฏธ์ง€์—๋„ ํ™•์žฅ ๊ฐ€๋Šฅํ•˜๋ฉฐ, ํ•ด๋‹น ๋ชจ๋ธ์€ ์ง๊ด€์ ์ด๊ณ  ๊ณ„์‚ฐ์ด ๋‹จ์ˆœํ•˜์—ฌ ๊ตญ์ง€ ํ™ˆ ๋‚ด๋น„๊ฒŒ์ด์…˜์— ๋‹ค์–‘ํ•˜๊ฒŒ ํ™œ์šฉ๋˜์–ด ์™”๋‹ค [16].

์‹œ๊ฐ ๋žœ๋“œ๋งˆํฌ๋ฅผ ์ด์šฉํ•œ ๋‚ด๋น„๊ฒŒ์ด์…˜์€ ALV ๋ฐฉ์‹ ์™ธ์—๋„ COMALV (CenterยญOfยญMass ALV), ACV (Average Correctional Vector), DELV (Distance Estimated Landmark Vector) ๋“ฑ ๋‹ค์–‘ํ•˜๋ฉฐ [17-19], ๋žœ๋“œ๋งˆํฌ์˜ ์œ„์น˜์ •๋ณด๊ฐ€ ์•„๋‹Œ ๋ถ„ํฌ๋ฅผ ์ด์šฉํ•œ ํ™ˆ ๋ฒกํ„ฐ๋ฅผ ์ถ”์ •ํ•˜๋Š” ๋ฐฉ์‹ ๋˜ํ•œ ๊ฐ€๋Šฅํ•จ์ด ์ œ์•ˆ๋˜์–ด ์™”๋‹ค [20]. ๋˜ํ•œ ๋žœ๋“œ๋งˆํฌ ๋ฒกํ„ฐ์— ๋Œ€ํ•œ ํ•™์Šต ๋ฐฉ๋ฒ• ๋˜๋Š” ๋‚ด๋น„๊ฒŒ์ด์…˜์„ ์œ„ํ•œ ํŠน์ง• ์ถ”์ถœ์— ๋Œ€ํ•œ ์—ฐ๊ตฌ๊ฐ€ ์ง„ํ–‰๋˜์–ด ์™”๋‹ค [21,22].

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์Šค๋ƒ…์ƒท ๋ชจ๋ธ ์ค‘ ALV ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜์—ฌ, ๋‘ ์œ„์น˜๊ฐ„์˜ ์ƒ๋Œ€์  ๋ฐฉํ–ฅ์„ ์ฐพ์•„๋‚ด๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด ๊ธฐ์ˆ ํ•˜์˜€๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ์‚ฌ์ „์— ์ „ ๋ฐฉํ–ฅ ์ดฌ์˜์ด ๊ฐ€๋Šฅํ•œ ์˜ด๋‹ˆ ์นด๋ฉ”๋ผ๋กœ ํš๋“ํ•œ ์Šค๋ƒ…์ƒท์„ ํŒŒ๋…ธ๋ผ๋งˆ ํ˜•ํƒœ๋กœ ๋ณ€๊ฒฝํ•˜์˜€๋‹ค. ๋‘ ํŒŒ๋…ธ๋ผ๋งˆ ์ด๋ฏธ์ง€์—์„œ ๋‹ค์–‘ํ•œ ํ˜•ํƒœ์˜ ๋žœ๋“œ๋งˆํฌ๋ฅผ ๊ฒ€์ถœํ•˜์—ฌ, ๋‘ ์œ„์น˜์—์„œ์˜ ๋žœ๋“œ๋งˆํฌ๋“ค์˜ ์ƒ๋Œ€์  ์œ„์น˜๋ณ€ํ™”๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ์ƒ๋Œ€์ ์ธ ์œ„์น˜๋ณ€ํ™”๋ฅผ ๊ณ„์‚ฐํ•˜์˜€๋‹ค. ์ด๋•Œ ๋žœ๋“œ๋งˆํฌ๋ฅผ ๊ฒ€์ถœํ•˜๊ธฐ ์œ„ํ•ด SURF (Speeded Up Robust Features) ๋“ฑ ์ฝ”๋„ˆ ํŠน์ง•์  ์ถ”์ถœ ์•Œ๊ณ ๋ฆฌ์ฆ˜ [23]๊ณผ, ์˜์ƒ์— Mask๋ฅผ ์ ์šฉํ•˜์—ฌ ๊ฐ ๋งˆ์Šคํฌ์˜ ํ˜•ํƒœ์— ๋”ฐ๋ฅธ ๋‹ค์–‘ํ•œ ํŠน์ง• ๊ฐ’(Feature Value)์„ ์ฐพ์•„๋‚ด๋Š” HaarยญLike Features [24] ๋ฐ ์ˆ˜์ง์—ฃ์ง€ ์ถ”์ถœ ๋ฐ ๋งค์นญ์„ ์œ„ํ•œ Hough ๋ณ€ํ™˜ [25], HOG (Histogram Of Gradient) ์•Œ๊ณ ๋ฆฌ์ฆ˜ [26]์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ์ถ”๊ฐ€์ ์œผ๋กœ ์–ธ๊ธ‰ํ•œ 3๊ฐ€์ง€ ๋ฐฉ์‹์˜ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•ด ์ •ํ™•ํ•œ ๋žœ๋“œ๋งˆํฌ ์„ ์ • ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ œ์•ˆํ•˜์˜€์œผ๋ฉฐ, ๊ฒฐ๊ณผ๋ฅผ ์ƒํ˜ธ ๋น„๊ตํ•˜์—ฌ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์„ ํ™•์ธํ•˜์˜€๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ด๋Ÿฌํ•œ ๋ฐฉ์‹์œผ๋กœ ๊ตฌํ•œ ๊ฒฐ๊ณผ๋ฅผ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•ด ์œ„์—์„œ ์–ธ๊ธ‰ํ•œ ๋‹ค์–‘ํ•œ ๋ฐฉ๋ฒ•์„ ์กฐํ•ฉํ•˜์—ฌ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด ๊ธฐ์ˆ ํ•˜์˜€๋‹ค.

์–ธ๊ธ‰ํ•œ ๋ฐฉ์‹์„ ํ†ตํ•ด ๊ตฌํ•œ ๊ฒฐ๊ณผ๋“ค์—์„œ ๊ฐ ํ™˜๊ฒฝ์—์„œ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ณ , ๊ทธ ๊ฒฐ๊ณผ๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ๊ฐ ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด ๊ฐ€์ค‘์น˜๋ฅผ ๋ถ€์—ฌํ•˜์—ฌ ์ด๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ์œ„์— ์—ด๊ฑฐํ•œ ๋ฐฉ์‹์„ ์„ ๋ณ„์ ์œผ๋กœ ํ™œ์šฉํ•œ๋‹ค. ๊ฐ ๋ฐฉ๋ฒ•์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ณ  ๊ฐ€์ค‘์น˜๋ฅผ ๊ณ„์‚ฐํ•˜๊ธฐ ์œ„ํ•œ ๋ฐฉ๋ฒ•์œผ๋กœ, ์ผ๋ถ€์œ„์น˜์—์„œ ๊ตฌํ•œ ๊ฐ ๊ฒฐ๊ณผ๋ฅผ ์‹ค์ œ ์œ„์น˜์™€ ๋น„๊ตํ•˜์—ฌ, ๊ฐ๊ฐ์˜ ๋ฐฉ์‹์— ๋Œ€ํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ์‚ฌ์ „์— ๊ฒฝ์‚ฌํ•˜๊ฐ•๋ฒ• (Gradient Descent Method)์„ ์‚ฌ์šฉํ•˜์—ฌ ์ตœ์ ํ™”ํ•˜์˜€๋‹ค. ์ด๋ ‡๊ฒŒ ๊ณ„์‚ฐํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ตฌํ•œ ํ™ˆ ๋ฒกํ„ฐ๋ฅผ ์ „์ฒด์œ„์น˜์—์„œ ์ ์šฉํ•˜์˜€์„๋•Œ์˜ ๊ฒฐ๊ณผ๋ฅผ ํ™•์ธํ•˜๊ณ , ๊ธฐ์กด์˜ ๋ฐฉ์‹์˜ ๊ฒฐ๊ณผ์™€ ๋น„๊ตํ•˜์˜€๋‹ค. ๋˜ํ•œ ํŠน์ •์œ„์น˜์—์„œ ํ•™์Šตํ•œ ๊ฒฐ๊ณผ๊ฐ€ ์ด์™€ ์œ ์‚ฌํ•œ ๋‹ค๋ฅธ ํ™˜๊ฒฝ์—์„œ๋„ ์ž˜ ์ ์šฉ๋˜๋Š”์ง€์— ๋Œ€ํ•ด ์‹คํ—˜์„ ํ†ตํ•ด ํ™•์ธํ•˜์˜€๋‹ค.

2. ๋ณธ ๋ก 

2.1 ์Šค๋ƒ…์ƒท ๋ชจ๋ธ(Snapshot Model)

์Šค๋ƒ…์ƒท ๋ชจ๋ธ์€ ๋‘ ์ง€์ ์˜ ์Šค๋ƒ…์ƒท ์ด๋ฏธ์ง€๋ฅผ ํ†ตํ•ด ์ƒ๋Œ€์  ์œ„์น˜๋ฅผ ๊ณ„์‚ฐํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค [15]. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์Šค๋ƒ…์ƒท ๋ชจ๋ธ ์ค‘ ํ˜„์žฌ(Current) ์ง€์ ๊ณผ ๋ชฉํ‘œ(Home)์ง€์ ์—์„œ ์ดฌ์˜๋œ ์Šค๋ƒ…์ƒท์—์„œ ๊ด€์ธก์ด ๊ฐ€๋Šฅํ•œ ๋žœ๋“œ๋งˆํฌ์˜ ํ‰๊ท  ํ•ฉ ๋ฒกํ„ฐ๋ฅผ ์ด์šฉํ•˜์—ฌ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๋Š” ๋ฐฉ๋ฒ• (ALV Method)์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค.

2.1.1 ALV(Average Landmark Vector)

๊ทธ๋ฆผ. 1. ALV ๋ฐฉ๋ฒ• : ๋‘ ์œ„์น˜์—์„œ ๋ฐ”๋ผ๋ณธ ๋žœ๋“œ๋งˆํฌ๋“ค์˜ ํ‰๊ท  ๋ฐฉํ–ฅ๋ฒกํ„ฐ๋กœ ํ™ˆ ๋ฒกํ„ฐ ๊ณ„์‚ฐ

Fig. 1. ALV Method : Homing vector using landmark vectors at two locations

../../Resources/kiee/KIEE.2020.69.2.337/fig1.png

ALV(Average Landmark Vector) ๋ฐฉ๋ฒ•์€ ์Šค๋ƒ…์ƒท์—์„œ์˜ ๊ณตํ†ต๋œ ๋žœ๋“œ๋งˆํฌ์˜ ์œ„์น˜ ํ‰๊ท ๋ฒกํ„ฐ์˜ ์ฐจ์ด๋ฅผ ํ†ตํ•ด ๋‘ ์œ„์น˜์˜ ์ƒ๋Œ€์  ๋ฐฉํ–ฅ์„ ์ถ”์ •ํ•˜๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด๋‹ค.

(1)
$$\overline{H}=\overline{ALV_{cur}}-\overline{ALV_{tar}}=\dfrac{1}{N}\sum_{i=1}^{n}\overline{lan_{i}}^{cur}-\dfrac{1}{N}\sum_{i=1}^{n}\overline{lan_{i}}^{tar}$$

์‚ฌ์šฉ๋œ cur ํ‘œ์‹œ๋Š” ํ˜„์žฌ (current) ์œ„์น˜์—์„œ์˜ ๋žœ๋“œ๋งˆํฌ์™€ ํ‰๊ท  ๋ฒกํ„ฐ, tar ํ‘œ์‹œ๋Š” ๋ชฉํ‘œ (target) ์œ„์น˜์—์„œ์˜ ๋žœ๋“œ๋งˆํฌ์™€ ํ‰๊ท  ๋ฒกํ„ฐ๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค. ๊ทธ๋ฆผ 1๊ณผ ์‹ (1)์€ ํ˜„์žฌ(current) ์ง€์ ๊ณผ ๋ชฉํ‘œ(target) ์ง€์ ์—์„œ์˜ ๋™์ผํ•œ ๋žœ๋“œ๋งˆํฌ์˜ ํ‰๊ท ๋ฒกํ„ฐ๋ฅผ ํ†ตํ•ด ํ™ˆ ๋ฒกํ„ฐ๋ฅผ ์ถ”์ •ํ•ด๋‚ด๋Š” ๊ณผ์ •์„ ๋ณด์—ฌ์ฃผ๊ณ  ์žˆ๋‹ค. ์ž„์˜์˜ ํ•œ ์ง€์ (current)์—์„œ ์›๋ž˜์˜ ์œ„์น˜(target)์œผ๋กœ ํ–ฅํ•˜๋Š” ๋ฐฉํ–ฅ๋ฒกํ„ฐ๋Š” ๋‘ ์ง€์ ์—์„œ ๋ฐ”๋ผ๋ณธ ๋žœ๋“œ๋งˆํฌ๋“ค์˜ ๋ฐฉํ–ฅ๋ฒกํ„ฐ๋“ค์˜ ํ‰๊ท  ํ•ฉ ๋ฒกํ„ฐ์˜ ์ฐจ์ด๋กœ ๊ตฌํ•  ์ˆ˜ ์žˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ณผ์ •์—์„œ ์ค‘์š”ํ•œ ๊ฒƒ์€ ๋‘ ์œ„์น˜์—์„œ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•จ์— ์žˆ์–ด์„œ ๋™์ผํ•œ ๋žœ๋“œ๋งˆํฌ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ์•ผ ํ•œ๋‹ค๋Š” ์ ๊ณผ ๋ฐฉ์œ„๊ฐ(Compass)์ •๋ณด๋ฅผ ๊ฐ–๊ณ  ์žˆ์–ด์„œ ์–ด๋–ค ์œ„์น˜์—์„œ๋„ ๊ฐ™์€ ๋ฐฉํ–ฅ์„ ์ง€ํ–ฅํ•จ์„ ์ „์ œ๋กœ ํ•œ๋‹ค.

2.1.2 ๋žœ๋“œ๋งˆํฌ ๊ฒ€์ถœ

์Šค๋ƒ…์ƒท์—์„œ ๋žœ๋“œ๋งˆํฌ ์ถ”์ถœ์€ ์˜์ƒ์˜ ์ˆ˜์ง ์—ฃ์ง€(Vertical Edge)์™€ ์ฝ”๋„ˆ ํŠน์ง•์ ์„ ํ™œ์šฉํ•˜์—ฌ ์‚ฌ์šฉ์ด ๊ฐ€๋Šฅํ•˜๋‹ค. ํ•ด๋‹น ๊ธฐ์กด ์—ฐ๊ตฌ์˜ ๊ธฐ์ˆ  ๋ฐฉ๋ฒ• [22,25]์—์„œ๋Š” Hough Transform, HOG ๋“ฑ์˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ด์šฉํ•ด์„œ ์ˆ˜์ง ์—ฃ์ง€๋ฅผ ๊ฒ€์ถœํ•˜์—ฌ ๋žœ๋“œ๋งˆํฌ๋กœ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹ (๋ฐฉ๋ฒ•1)๊ณผ SIFT๋ฅผ ํ†ตํ•ด ์ถ”์ถœํ•œ ์ฝ”๋„ˆํŠน์ง•์ ์„ ๋žœ๋“œ๋งˆํฌ๋กœ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹ (๋ฐฉ๋ฒ•2)์œผ๋กœ ALV ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜์˜€๋‹ค. ์œ„์™€ ๊ฐ™์€ ๋ฐฉ๋ฒ•์€ ํŠน์ • ํ˜•ํƒœ์˜ ๋žœ๋“œ๋งˆํฌ๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹์œผ๋กœ ์ถฉ๋ถ„ํ•œ ์ˆ˜์˜ ๋žœ๋“œ๋งˆํฌ ๊ฒ€์ถœ์ด ๊ฐ€๋Šฅํ•œ ํ™˜๊ฒฝ์—์„œ๋Š” ์ถฉ๋ถ„ํžˆ ํ™œ์šฉ๊ฐ€๋Šฅํ•˜๋‹ค. ๋˜ํ•œ ์ด๋Ÿฌํ•œ ๋ฐฉ์‹์€ ์œ„์น˜๊ฐ€ ๋ณ€ํ™”ํ•˜๋”๋ผ๋„ ๊ฐ๊ฐ์˜ ํŠน์ง•์  ๊ฐ„์˜ ์œ ์‚ฌ์„ฑ์„ ํ†ตํ•ด ๋งค์นญ์ด ๊ฐ€๋Šฅํ•˜๋‹ค.

2.1.3 ์ˆ˜์ง์—ฃ์ง€ ๋žœ๋“œ๋งˆํฌ ๋ฐฉ์‹ (๋ฐฉ๋ฒ• 1)

์ผ๋ฐ˜์ ์ธ ์‹ค๋‚ด ํ™˜๊ฒฝ์—์„œ ํ™œ์šฉ์ด ์šฉ์ดํ•œ ๋žœ๋“œ๋งˆํฌ๋กœ ์ˆ˜์ง์—ฃ์ง€(Vertical Edge)๋ฅผ ์ƒ๊ฐํ•ด ๋ณผ ์ˆ˜ ์žˆ๋‹ค. ์ˆ˜์ง์—ฃ์ง€๋Š” ์˜ด๋‹ˆ ์ด๋ฏธ์ง€์—์„œ ํŒŒ๋…ธ๋ผ๋งˆ ์ด๋ฏธ์ง€๋กœ์˜ ๋ณ€ํ™˜์—์„œ ๋ฐœ์ƒํ•˜๋Š” ์™œ๊ณก(Warping)์— ๋น„๊ต์  ๊ฐ•์ธํ•œ ํŠน์„ฑ์„ ๊ฐ€์ง€๊ณ  ๋‹จ์ผํ•œ ์œ„์น˜(๋ฐฉ์œ„๊ฐ)๋ฅผ ๊ฐ€์ง€๋ฏ€๋กœ ๋žœ๋“œ๋งˆํฌ ์‚ฌ์šฉ์œผ๋กœ ์ ํ•ฉํ•˜๋‹ค. ์ด๋ฏธ์ง€๋กœ๋ถ€ํ„ฐ ์ˆ˜์ง์—ฃ์ง€(Vertical Edge)๋ฅผ ์ฐพ๊ธฐ ์œ„ํ•ด์„œ ์˜์ƒ์ฒ˜๋ฆฌ ๊ธฐ๋ฒ• ์ค‘ ํ—ˆํ”„๋ณ€ํ™˜(Hough Transform)์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ํ—ˆํ”„๋ณ€ํ™˜์€ ์ด์ง„ ์ด๋ฏธ์ง€(Binary Image)๋ฅผ ๊ธฐ์ค€์ ์œผ๋กœ๋ถ€ํ„ฐ์˜ ๊ฑฐ๋ฆฌ์™€ ๊ฐ๋„๋กœ ์ด๋ฃจ์–ด์ง„ ํ—ˆํ”„ํ‰๋ฉด(Hough Plane)์œผ๋กœ ๋ณ€ํ™˜ํ•˜์—ฌ ์›ํ•˜๋Š” ๊ฐ๋„์—์„œ ์ž„๊ณ„์  ์ด์ƒ์˜ ๊ฐ’์„ ๊ฐ€์ง„ ์—ฃ์ง€(Edge)์„ ๊ฒ€์ถœํ•ด์ค€๋‹ค. ๊ทธ๋ฆผ 2๋Š” ์ด๋ฏธ์ง€์—์„œ ํ—ˆํ”„๋ณ€ํ™˜์œผ๋กœ ๊ฒ€์ถœํ•œ ์ˆ˜์ง์—ฃ์ง€๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ์˜์ƒ์—์„œ ์ขŒ์šฐ์˜ ๋ฐ๊ธฐ ๋ณ€ํ™”๊ฐ€ ํฐ ์—ฃ์ง€๋ฅผ ๊ฒ€์ถœํ•ด ์ฃผ๋ฉฐ, ์—ฃ์ง€์˜ ๊ธธ์ด๋Š” ์กฐ์ •์ด ๊ฐ€๋Šฅํ•˜๋‹ค.

๊ทธ๋ฆผ. 2. ํ—ˆํ”„๋ณ€ํ™˜์„ ํ†ตํ•œ ์ˆ˜์ง์—ฃ์ง€ ๊ฒ€์ถœ

Fig. 2. Vertical edge by Hough transform

../../Resources/kiee/KIEE.2020.69.2.337/fig2.png

๊ทธ๋Ÿฌ๋‚˜ ๊ฒ€์ถœ๋œ ์ˆ˜์ง์„ ๋“ค์€ ๋งค์นญ์„ ์œ„ํ•œ ๊ณ ์œ ๊ฐ’์ด ์—†๋Š” ์ƒํƒœ์ด๊ธฐ ๋•Œ๋ฌธ์— ๋งค์นญ์„ ์œ„ํ•ด ์ˆ˜์ง์„ ์„ ์ค‘์‹ฌ์œผ๋กœ ์ผ์ •๊ตฌ์—ญ์— HOG(Histogram Of Gradient) ๊ฐ’์„ ๊ณ„์‚ฐํ•˜์˜€๋‹ค. ํ—ˆํ”„๋ณ€ํ™˜์œผ๋กœ ๊ฒ€์ถœ๋œ ๋žœ๋“œ๋งˆํฌ ํ›„๋ณด๋“ค์€ ์ˆ˜์ง์„  ์ธ๊ทผ์˜ HOG์˜ ๊ฒฐ๊ณผ๊ฐ’์„ ๋น„๊ตํ•˜์—ฌ ๋งค์นญํ•˜๋ฉฐ ์ฐจ์ด๊ฐ€ ์ž„๊ณ„๊ฐ’ ์ดํ•˜์ธ ๋งค์นญ์— ๋Œ€ํ•ด์„œ๋Š” ๋žœ๋“œ๋งˆํฌ๋กœ ์ธ์ •ํ•˜๋ฉฐ ALV ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋”ฐ๋ผ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๋Š”๋ฐ ์‚ฌ์šฉํ•œ๋‹ค (๊ทธ๋ฆผ 3).

๊ทธ๋ฆผ. 3. ์ˆ˜์ง์—ฃ์ง€ ๋žœ๋“œ๋งˆํฌ ๋ฐฉ์‹

Fig. 3. Landmark Using vertical edge method

../../Resources/kiee/KIEE.2020.69.2.337/fig3.png

2.1.4 ์ฝ”๋„ˆ ํŠน์ง•์  ๋žœ๋“œ๋งˆํฌ ๋ฐฉ์‹ (๋ฐฉ๋ฒ• 2)

SIFT (Scale-Invariant Feature Transform), SURF (Speeded Up Robust Features) ๋“ฑ์€ ๋ชจ๋‘ ์˜์ƒ์˜ ์ฝ”๋„ˆ ํŠน์ง•์  ๊ฒ€์ถœ์„ ์œ„ํ•ด ์‚ฌ์šฉํ•˜๋Š” ๋Œ€ํ‘œ์ ์ธ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด๋‹ค [23,27]. ์ฝ”๋„ˆ ํŠน์ง•์ ์„ ์ด์šฉํ•˜๋Š” ๋ฐฉ์‹์€ ๋‘ ์ง€์ ์˜ ์Šค๋ƒ…์ƒท์—์„œ ํŠน์ง•์ ์„ ๊ฒ€์ถœํ•œ ํ›„ ๊ฒ€์ถœ๋œ ํŠน์ง•์ ๋“ค์„ ๋žœ๋“œ๋งˆํฌ๋กœ ์‚ฌ์šฉํ•˜์—ฌ ๋žœ๋“œ๋งˆํฌ ๋ฒกํ„ฐ๋“ค์˜ ํ•ฉ ๋ฒกํ„ฐ๋ฅผ ์ด์šฉํ•˜์—ฌ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ตฌํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํŠน์ง•์ ๋“ค์˜ Feature Value๋กœ ํ‘œํ˜„๋˜๋Š” ์œ ์‚ฌ๋„์— ๋”ฐ๋ผ ๋งค์นญํ•˜๋Š” ๋ฐฉ์‹์€ ๊ทธ๋ฆผ 4์™€ ๊ฐ™์ด ๋‹ค์ˆ˜์˜ ๋ถ€์ •ํ™•ํ•œ ๋งค์นญ์ด ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ๋‹ค.

๊ทธ๋ฆผ. 4. SURF ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ํ†ตํ•œ ํŠน์ง•์  ๋งค์นญ

Fig. 4. Feature matching by SURF

../../Resources/kiee/KIEE.2020.69.2.337/fig4.png

์ด์ „์˜ ์—ฐ๊ตฌ [22]์—์„œ๋Š” ์ด๋Ÿฌํ•œ ๋ฌธ์ œ์ ์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด ๊ฐ€์žฅ ์œ ์‚ฌ๋„๊ฐ€ ๋†’์€ ํŠน์ง•์  ๋งค์นญ์˜ ์ผ๋ถ€๋งŒ์„ ํ™œ์šฉํ•˜์—ฌ ALV ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๋Š” ํ•œํŽธ, ์œ ์‚ฌ๋„์— ๋”ฐ๋ผ ๊ฐ€์ค‘์น˜(Weight)๋ฅผ ๋ถ€์—ฌํ•˜์—ฌ ๋ฒกํ„ฐ ๊ณ„์‚ฐ์— ๊ธฐ์—ฌ๋„๋ฅผ ์ฐจ๋“ฑ ์ ์šฉํ•˜๋Š” ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ์ด๋Ÿฌํ•œ ๋ฐฉ์‹์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์†Œ์ˆ˜์˜ ๋ถ€์ •ํ™•ํ•œ ๋งค์นญ์œผ๋กœ๋„ ์‹œ์Šคํ…œ์— ๋†’์€ ์˜ค์ฐจ๋ฅผ ๋ฐœ์ƒ์‹œํ‚ด์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์—ˆ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์œ„ ๋ฐฉ์‹์„ ๋ณด์™„ํ•˜๊ธฐ ์œ„ํ•ด์„œ ๊ฐ€๋Šฅํ•œ ๋งŽ์€ ๋žœ๋“œ๋งˆํฌ ํ›„๋ณด ๋งค์นญ์„ ์ถ”์ถœํ•œ ํ›„, ์‹ ๋ขฐ์„ฑ์ด ๋‚ฎ์€ ํ›„๋ณด๋“ค์€ ์‚ฌ์šฉํ•˜์ง€ ์•Š๋Š” ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค.

์‹ ๋ขฐ์„ฑ์ด ๋‚ฎ์€ ๋žœ๋“œ๋งˆํฌ๋ฅผ ์ฐพ๊ธฐ ์œ„ํ•ด์„œ ๋‘ ์ง€์ ๊ฐ„์˜ ๋žœ๋“œ๋งˆํฌ์˜ ์œ„์น˜์˜ ์ฐจ์ด๊ฐ€ ๋งค์šฐ ์‹ฌํ•˜๊ฑฐ๋‚˜ 90ยฐ์ด์ƒ), ํŠน์ง•์  ๊ฐ’์˜ ์ฐจ์ด๊ฐ€ ์ž„๊ณ„๊ฐ’๋ณด๋‹ค ํฐ ๋งค์นญ ํ›„๋ณด๋“ค์€ ์šฐ์„ ์ ์œผ๋กœ ์ œ๊ฑฐํ•˜๋Š” ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ์ด๋Š” ์™ธ๊ณฝ์— ์กด์žฌํ•˜๋Š” ๋žœ๋“œ๋งˆํฌ๋Š” ์ดฌ์˜์œ„์น˜๊ฐ€ 90๋„ ์ด์ƒ ๋ณ€ํ™”ํ•  ์ˆ˜ ์—†๋‹ค๋Š” ์‚ฌ์‹ค์„ ๊ทผ๊ฑฐ๋กœ ํ•˜๊ณ  ์žˆ๋‹ค. ๋งค์นญ์„ ์ •์ด ์™„๋ฃŒ๋˜๋ฉด ์ด๋“ค์„ ๋Œ€์ƒ์œผ๋กœ ALV ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๊ฒŒ ๋˜๋ฉฐ ์ •ํ™•ํ•œ ALV ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ• ์ˆ˜ ์žˆ์„ ๊ฒƒ์œผ๋กœ ๋ณด์ธ๋‹ค (๊ทธ๋ฆผ 5).

๊ทธ๋ฆผ. 5. ์ฝ”๋„ˆ ํŠน์ง•์  ๋žœ๋“œ๋งˆํฌ ๋ฐฉ์‹

Fig. 5. Landmark using corner method

../../Resources/kiee/KIEE.2020.69.2.337/fig5.png

์ด๋Ÿฌํ•œ ๋งค์นญ ์„ ์ •๋ฐฉ์‹์€ ์ˆ˜์ง์—ฃ์ง€๋ฅผ ํ™œ์šฉํ•œ ๋ฐฉ์‹(๋ฐฉ๋ฒ• 1)์—์„œ๋„ ํ™œ์šฉ์ด ๊ฐ€๋Šฅํ•˜๋ฉฐ ๋‹ค๋ฅธ ํŠน์ง•์  ์ถ”์ถœ ์•Œ๊ณ ๋ฆฌ์ฆ˜์—์„œ๋„ ์ ์šฉ์ด ๊ฐ€๋Šฅํ•˜๋‹ค.

2.1.5 ๋ฐ๊ธฐ(Brightness) ์‚ฌ์šฉ ๋ฐฉ์‹ (๋ฐฉ๋ฒ• 3)

HaarยญLikeยญFeature๋Š” ๋ฌผ์ฒด์ธ์‹์— ๋งŽ์ด ์‚ฌ์šฉ๋˜๋Š” ์˜์ƒ์ฒ˜๋ฆฌ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด๋‹ค [24]. ๋‹ค์–‘ํ•œ ํ˜•ํƒœ์˜ Haar-like ๋งˆ์Šคํฌ๋ฅผ ํ†ตํ•ด ๋Œ€์ƒ์„ ์ธ์‹ํ•˜๋Š” ๋ฐฉ์‹์„ ์ด์šฉํ•˜๋ฉฐ, ์ ์€ ๊ณ„์‚ฐ๋Ÿ‰์œผ๋กœ ์ข‹์€ ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์ฃผ๊ธฐ ๋•Œ๋ฌธ์— ๊ตญ์ง€ ๋‚ด๋น„๊ฒŒ์ด์…˜ ๋ถ„์•ผ์—์„œ๋„ ๋‹ค์–‘ํ•œ ๋ฐฉ์‹์œผ๋กœ ํ™œ์šฉ๋˜๊ณ  ์žˆ๋‹ค.

์ด์™€ ๊ฐ™์€ ๋ฐฉ๋ฒ•์„ ํ†ตํ•ด ์—ฐ๊ตฌ๋œ HFLV (HaarยญlikeยญFeature Landmark Vector)๋Š” ๋‘ ๊ฐœ์˜ ํŒŒ๋…ธ๋ผ๋งˆ ์ด๋ฏธ์ง€์˜ ๊ฐ™์€ ์œ„์น˜์— Haarยญlike Feature Mask๋ฅผ ๋ฐฐ์น˜์‹œํ‚ค๊ณ  ๋‘ ์ด๋ฏธ์ง€์˜ ์œ ์‚ฌ๋„๋ฅผ ์ธก์ •ํ•˜๊ณ  ์‹ ํ˜ธ์˜ ์ฐจ์ด๋งŒํผ ๋ฐฉํ–ฅ๋ฒกํ„ฐ๋ฅผ ์ƒ์„ฑํ•˜์—ฌ ํ˜„์žฌ์œ„์น˜์—์„œ ๋ชฉํ‘œ์ง€์ ์˜ ๋ฐฉํ–ฅ์„ ์ถ”์ธกํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค [28].

์ด๋ฒˆ ์—ฐ๊ตฌ์—์„œ๋Š” ์ด์ „์˜ ๋‹ค์–‘ํ•œ ์ข…๋ฅ˜์˜ ๋งˆ์Šคํฌ๋ฅผ ํ™œ์šฉํ•˜์—ฌ ALV ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๋Š” ๋ฐฉ์‹๊ณผ ๋‹ฌ๋ฆฌ ์˜ค์ง ํ•ด๋‹น ๊ตฌ์—ญ์˜ ๋ฐ๊ธฐ๋งŒ์„ ์‚ฌ์šฉํ•˜์—ฌ ๋žœ๋“œ๋งˆํฌ์˜ ํŠน์ง•๊ฐ’ ๋ณ€ํ™”๋Ÿ‰์„ ํ™•์ธํ•œ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ๊ฐ€์žฅ ๋‹จ์ˆœํ•œ ๋งˆ์Šคํฌ (๋งˆ์Šคํฌ ๋‚ด์˜ ํ•ด๋‹น ํ”ฝ์…€๊ฐ’์˜ ํ•ฉ)๋ฅผ ์‚ฌ์šฉํ•˜๋ฉฐ, ๊ท ๋“ฑํ•œ ๋ถ„ํฌ๋กœ ๋งˆ์Šคํฌ๋ฅผ ์ ์šฉํ•˜์—ฌ, ํ˜„์žฌ ์œ„์น˜์™€ ํ™ˆ (๋ชฉ์ ) ์œ„์น˜์˜ ๋‘ ์œ„์น˜์—์„œ์˜ ์Šค๋ƒ…์ƒท์— ๋Œ€ํ•˜์—ฌ ๊ฐ™์€ ๋งˆ์Šคํฌ (๋žœ๋“œ๋งˆํฌ ๋Œ€์ƒ)๋ฅผ ์‚ฌ์šฉํ•˜๊ณ , ๊ฐ ๋งˆ์Šคํฌ ๊ฐ’์˜ ์ฐจ์ด๋ฅผ ์‹ (2)์™€ ๊ฐ™์€ ์Šค์ฝ”์–ด๋กœ ์ •์˜ํ•˜์˜€๋‹ค.

(2)
$$Score_{ALV_{i}}=Score_{cur_{i}}-Score_{tar_{i}}$$

์ด๋ ‡๊ฒŒ ๊ณ„์‚ฐ๋œ ๊ฐ๊ฐ์˜ ์Šค์ฝ”์–ด๋ฅผ ์ด์šฉํ•˜์—ฌ ํ™ˆ ๋ฒกํ„ฐ๋ฅผ ์‹ (3), (4)๋ฅผ ํ†ตํ•ด ๊ณ„์‚ฐํ•œ๋‹ค.

(3)
$$\overline{ALV_{x}}=\sum_{i=1}^{n}Score_{ALV_{i}}\times\cos\theta$$

(4)
$$\overline{ALV_{y}}=\sum_{i=1}^{n}Score_{ALV_{i}}\times\sin\theta$$

์ด ๋ฐฉ์‹์€ ๋ณ„๋„์˜ ๋งค์นญ์ž‘์—… ์—†์ด ๋ฐ”๋กœ ๊ณ„์‚ฐ์ด ๊ฐ€๋Šฅํ•˜๊ณ , ๋‹จ์ผํ•œ ๋งˆ์Šคํฌ๋ฅผ ๊ท ์ผํ•œ ๊ณต๊ฐ„์— ์ ์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์ƒ๋Œ€์ ์œผ๋กœ ์ ์€ ์—ฐ์‚ฐ์œผ๋กœ ํ™ˆ ๋ฒกํ„ฐ๋ฅผ ์ƒ์„ฑํ•ด ๋‚ผ ์ˆ˜ ์žˆ๋Š” ์žฅ์ ์ด ์žˆ๋‹ค. ๋ฐ˜๋ฉด ์Šค๋ƒ…์ƒท ์ด๋ฏธ์ง€์˜ ๋ฐ๊ธฐ์ฐจ์ด๋ฅผ ์‚ฌ์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์ „์ฒด์ ์œผ๋กœ ์–ด๋‘์šด ํ™˜๊ฒฝ์—์„œ๋Š” ์ƒ๋Œ€์ ์œผ๋กœ ์„ฑ๋Šฅ์ด ์ €ํ•˜๋  ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ๋‹ค.

2.2 ํ•™์Šต์„ ํ†ตํ•œ ๋ฐฉ์‹์กฐํ•ฉ

๊ฐ๊ฐ์˜ ๋ฐฉ๋ฒ•์€ ๊ณตํ†ต์ ์œผ๋กœ ์Šค๋ƒ…์ƒท์„ ์‚ฌ์šฉํ•˜์—ฌ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ์ถ”์ •ํ•˜์ง€๋งŒ ๊ฐ๊ธฐ ๋‹ค๋ฅธ ๋ฐฉ์‹์˜ ํŠน์ง•์ ์„ ๋žœ๋“œ๋งˆํฌ๋กœ ์‚ฌ์šฉํ•˜๋ฏ€๋กœ, ํ™˜๊ฒฝ์— ๋”ฐ๋ผ์„œ ์ƒ๋Œ€์ ์œผ๋กœ ์ข‹์€ ์„ฑ๋Šฅ์„ ๋ณด์ด๋Š” ๋ฐฉ์‹์— ๋†’์€ ๊ฐ€์ค‘์น˜๋ฅผ ๋ถ€์—ฌํ•˜์—ฌ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹์ด ๊ฐ€๋Šฅํ•˜๋‹ค. ์ด ๋ฐฉ์‹์„ ์ ์šฉํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ํ•ด๋‹น ์žฅ์†Œ ๋‚ด ์—ฌ๋Ÿฌ ์œ„์น˜์—์„œ ์ดฌ์˜ํ•œ ์ด๋ฏธ์ง€์™€ ๊ทธ ์œ„์น˜์—์„œ ์—๋Ÿฌ๋ฅผ ๊ณ„์‚ฐํ•˜๊ธฐ ์œ„ํ•œ ์ •๋‹ต์ง€๊ฐ€ ํ•„์š”ํ•˜๋‹ค. ์ •๋‹ต์ง€๋Š” ์นด๋ฉ”๋ผ ์ด์™ธ์˜ ์„ผ์„œ๋ฅผ ์ด์šฉํ•˜์—ฌ ๊ณ„์‚ฐํ•œ ๊ฐ’์„ ์ด์šฉํ•  ์ˆ˜ ์žˆ์œผ๋‚˜, ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์‹ค์ œ ์œ„์น˜๋กœ๋ถ€ํ„ฐ ์ถ”์ถœํ•œ ์ •ํ™•ํ•œ ์ •๋‹ต์ง€๋ฅผ ๊ฐ–๊ณ  ์žˆ์Œ์„ ๊ฐ€์ •ํ•˜๊ณ  ์—ฐ๊ตฌ๋ฅผ ์ง„ํ–‰ํ•˜์˜€๋‹ค. ์‹ค์ œ ๋กœ๋ด‡์—์„œ๋Š” ๊ฒฝ๋กœ ๋ˆ„์  ๊ณ„์‚ฐ ๋ฐฉ์‹์œผ๋กœ ์ถ”์ •ํ•  ์ˆ˜ ์žˆ๋‹ค.

์ด๋ฒˆ ์—ฐ๊ตฌ์—์„œ๋Š” ๋จผ์ € 3๊ฐ€์ง€ ๋ฐฉ๋ฒ•์„ ์ด์šฉํ•˜์—ฌ ALV ๋ฒกํ„ฐ๋ฅผ ๊ตฌํ•œ๋‹ค. ์ด๋ ‡๊ฒŒ ๊ตฌํ•œ 3๊ฐ€์ง€ ALV ๋ฒกํ„ฐ๋ฅผ ์ด์šฉํ•˜์—ฌ ํ…Œ์ŠคํŠธ ์ „์ฒด ๊ตฌ์—ญ์˜ ํ‰๊ท  ์—๋Ÿฌ๊ฐ’์ด ๊ฐ€์žฅ ๋‚ฎ์€ ๊ฐ’์ด ๋˜๋„๋ก ๊ฐ ๋ฐฉ๋ฒ•๋ณ„ ALV Vector ๊ฐ’์— ๊ฐ€์ค‘์น˜๋ฅผ ๋ถ€์—ฌํ•œ๋‹ค. ์ตœ์ ์˜ ๊ฐ€์ค‘์น˜๋ฅผ ๊ณ„์‚ฐํ•˜๊ธฐ ์œ„ํ•ด ๊ฒฝ์‚ฌํ•˜๊ฐ•๋ฒ• (Gradient Descent)๋ฅผ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ๊ฒฝ์‚ฌํ•˜๊ฐ•๋ฒ•์€ Cost Function์˜ ๊ธฐ์šธ๊ธฐ๊ฐ€ ๋‚ฎ์€ ์ชฝ์œผ๋กœ ์ด๋™์‹œ์ผœ์„œ ๊ทน๊ฐ’์— ์ด๋ฅผ ๋•Œ๊นŒ์ง€ ์ด๋ฅผ ๋ฐ˜๋ณต์‹œํ‚ค๋Š” ์ตœ์ ํ™” ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด๋‹ค. ํ•ด๋‹น ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์‹ ๊ฒฝ๋ง ํšŒ๋กœ ํ•™์Šต์— ๋งŽ์ด ํ™œ์šฉ๋œ๋‹ค [29]. Cost Function์€ Desired Value ์—์„œ ๊ณ„์‚ฐ๋œ ๊ฐ’๊ณผ์˜ ์ฐจ์ด์˜ ์ œ๊ณฑ์œผ๋กœ ์ •์˜ํ•˜๋ฉฐ ์‹ (5)์™€ ๊ฐ™์ด ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ๋‹ค.

(5)
$$E(W)=\dfrac{1}{2}(\sum_{i=1}^{m}W_{i}x_{i}- d_{i})^{2}$$

์‹ (5)๋Š” ์‹ (6)์œผ๋กœ ๊ฐ„๋‹จํ•˜๊ฒŒ ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ์œผ๋ฉฐ ์ด๋ฅผ ์ด์šฉํ•˜์—ฌ ์ง€์†์ ์œผ๋กœ ๋ณธ ํ•จ์ˆ˜์˜ ๊ฒฐ๊ณผ ๊ฐ’์ด ๋ชฉํ‘œ ๊ฐ’์— ์ˆ˜๋ ดํ•˜๋„๋ก ๊ฐ€์ค‘์น˜๋ฅผ ์—…๋ฐ์ดํŠธํ•  ์ˆ˜ ์žˆ๋‹ค - ์‹ (7) ์ฐธ์กฐ.

(6)
$$\dfrac{\partial E}{\partial w}=\sum_{i=1}^{m}W_{i}- d_{i}$$

(7)
$$W_{i+1}\approx W_{i}-\eta\dfrac{\partial E}{\partial w}$$

๋ณธ ์—ฐ๊ตฌ์—์„œ ๊ฐ๊ฐ์˜ ๊ฒฐ๊ณผ๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ๊ฒฝ์‚ฌํ•˜๊ฐ•๋ฒ•์„ ์ ์šฉํ•˜์—ฌ ๊ตฌํ•˜๊ณ ์ž ํ•˜๋Š” ๊ฒƒ์€ 3๊ฐ€์ง€ ๋ฐฉ๋ฒ• (HaarยญLikeยญFeature, HOG, SURF)์œผ๋กœ ๊ตฌํ•œ ๋ฒกํ„ฐ์— ๋Œ€ํ•œ ๊ฐ€์ค‘์น˜๊ฐ’$(W)$์ด๋‹ค. 169๊ฐœ ์ง€์ ์—์„œ ๊ฐ ๋ฐฉ๋ฒ•์œผ๋กœ ๊ตฌํ•œ ๊ฐ’์— ๋Œ€ํ•œ ์˜ค์ฐจ๊ฐ’์„ ๊ตฌํ•˜๋ฉฐ, ๋ชจ๋“  ์ง€์ ์˜ ๊ฐ’์„ ๊ตฌํ•œ ๋’ค์—๋Š” ํ‰๊ท  ์˜ค์ฐจ๊ฐ’์„ ์ตœ์†Œํ™”ํ•˜๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ๊ฐ€์ค‘์น˜$W$ ๋ฅผ ๊ฐฑ์‹ ํ•œ๋‹ค. ๊ฐ ์œ„์น˜์—์„œ ์ •๋‹ต์ด ์žˆ์ง€๋งŒ ๊ฐ ๊ฐ€์ค‘์น˜๋ฅผ ๋ณ€ํ™”์‹œํ‚ค๋ฉด ๋ณด๋“  ์ง€์  ์—์„œ ์˜ํ–ฅ์„ ๋ฏธ์น˜๊ธฐ ๋•Œ๋ฌธ์— W๋Š” ์ „์ฒด (160๊ฐœ) ์ง€์ ์—์„œ์˜ ํ‰๊ท  ์˜ค์ฐจ๋ฅผ ์ค„์ด๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ๊ฐฑ์‹ ํ•˜๊ฒŒ ๋œ๋‹ค. ์ด๋ฅผ ์ˆ˜์‹์œผ๋กœ ๋‚˜ํƒ€๋‚ด๋ฉด ์‹ (8)๊ณผ ๊ฐ™๋‹ค.

(8)
$$E(W)=\dfrac{1}{2}\sum_{k=1}^{169}(\sum_{i=1}^{3}W_{i}^{k}x_{i}^{k}- d^{k})^{2}$$

์ด๋•Œ ๊ฐ€์ค‘์น˜$(W)$๋Š” $x$, $y$์ถ•์œผ๋กœ ๊ตฌ๋ถ„ํ•˜์—ฌ ๊ฐ€์ค‘์น˜๋ฅผ ๋‹ค๋ฅด๊ฒŒ ํ•  ์ˆ˜๋„ ์žˆ์ง€๋งŒ ์ด์™€ ๊ด€๋ จ ์—†์ด ๊ณตํ†ต์˜ ๊ฐ€์ค‘์น˜๋ฅผ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ตฌ๋ถ„ํ•˜์—ฌ ๊ฐ€์ค‘์น˜๋ฅผ ํ•™์Šตํ•˜๋Š” ๋ฐฉ์‹์ด ๋ณด๋‹ค ๋” ํ•™์Šต์ด ์‰ฝ๊ฒŒ ๋  ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์— ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊ฐ ๋ฐฉํ–ฅ์—์„œ ๋ณ„๋„๋กœ ๊ฐ€์ค‘์น˜๋ฅผ ๊ณ„์‚ฐํ•˜์—ฌ ์ ์šฉํ•˜๋Š” ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค.

๋ณธ ํ•™์Šต๋ฐฉ์‹์€ ํ™ˆ ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๋Š”๋ฐ ์žˆ์–ด์„œ ๊ธฐ์กด์˜ ํ•™์Šต์„ ํ†ตํ•œ ๋ฐฉ์‹ [21]๊ณผ ์ฐจ์ด๊ฐ€ ์žˆ๋‹ค. ๊ธฐ์กด ๋ฐฉ์‹ (HFLVยญGD)์˜ ๊ฒฝ์šฐ์—๋Š” ์˜์ƒ์— ์ ์šฉํ•˜๋Š” ๋งˆ์Šคํฌ์˜ ์ข…๋ฅ˜์™€ ์œ„์น˜๋ฅผ ํ•™์Šต์„ ํ†ตํ•ด ๊ฒฐ์ •ํ•˜๋Š” ๋ฐฉ์‹์ด๋ฉฐ, ๋ณธ ์—ฐ๊ตฌ์—์„œ ํ™œ์šฉํ•˜๋Š” ๋ฐฉ์‹์€ ๋žœ๋“œ๋งˆํฌ ํ˜•ํƒœ(Vertical Edge, Corner ๋“ฑ)๋ฅผ ๊ฒฐ์ •ํ•˜๊ธฐ ์œ„ํ•ด ํ•™์Šต์„ ์ˆ˜ํ–‰ํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค. ์ด๋Ÿฌํ•œ ํ•™์Šต์„ ํ†ตํ•œ ๋ฐฉ์‹์€ ์ตœ์ข…์ ์ธ ํ™ˆ ๋ฒกํ„ฐ๋ฅผ ์ƒ์„ฑํ•˜๋Š”๋ฐ ์žˆ์–ด์„œ 3๊ฐ€์ง€ ๋ฐฉ์‹์„ ๋ชจ๋‘ ์‚ฌ์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๊ณ„์‚ฐ์†๋„ ์ธก๋ฉด์—์„œ๋Š” ๋ถˆ๋ฆฌํ•˜๋‹ค๊ณ  ๋ณผ ์ˆ˜ ์žˆ๋Š” ๋ฐ˜๋ฉด์— ๊ธฐ์กด์˜ ํ•œ ๊ฐ€์ง€ ๋ฐฉ์‹์— ์˜์กดํ•˜๋Š” ๋ฐฉ๋ฒ•๋ณด๋‹ค ๋” ๋‹ค์–‘ํ•œ ํ™˜๊ฒฝ์—์„œ ํ–ฅ์ƒ๋œ ๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ๋‹ค.

2.3 ์‹คํ—˜ ๋ฐ ๊ฒฐ๊ณผ

2.3.1 ์‹คํ—˜ ํ™˜๊ฒฝ

์œ„ ๋ฐฉ๋ฒ•์œผ๋กœ ๊ตฌํ•œ ํ™ˆ๋ฒกํ„ฐ์˜ ์„ฑ๋Šฅํ™•์ธ์„ ์œ„ํ•ด ๋‘ ๊ฐ€์ง€ Vardy ๋ฐ์ดํ„ฐ ์…‹ (๊ทธ๋ฆผ 6)์„ ํ™œ์šฉํ•˜์˜€๋‹ค. ํ•ด๋‹น ๋ฐ์ดํ„ฐ ์…‹์€ ์‹œ๊ฐ ๋‚ด๋น„๊ฒŒ์ด์…˜์„ ์œ„ํ•ด ๊ตฌ์„ฑ๋˜์–ด ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์—์„œ ํ™œ์šฉ์ค‘์ด๋‹ค.

๊ทธ๋ฆผ. 6. Vardy ๋ฐ์ดํ„ฐ์…‹ : Original(์ƒ), Hall(ํ•˜)

Fig. 6. Vardy dataset and vertical edges : Original (upper), Hall (lower)

../../Resources/kiee/KIEE.2020.69.2.337/fig6.png

์‚ฌ์šฉํ•  ์‹คํ—˜ ๋ฐ์ดํ„ฐ์…‹ ์ค‘์—์„œ ์œ„์ชฝ ์ด๋ฏธ์ง€๋Š” 170๊ฐœ (17ร—10) ์œ„์น˜์—์„œ 752ร—564 ํฌ๊ธฐ์˜ ์ „ ๋ฐฉํ–ฅ ์˜ด๋‹ˆ์ด๋ฏธ์ง€๋ฅผ 720ร—120 ํŒŒ๋…ธ๋ผ๋งˆ ์ด๋ฏธ์ง€๋กœ, ์•„๋ž˜๋Š” 200๊ฐœ (20ร—10) ์œ„์น˜์—์„œ 720ร—125 ํŒŒ๋…ธ๋ผ๋งˆ ์ด๋ฏธ์ง€๋กœ ๋ณ€ํ™˜ํ•œ ๋ฐ์ดํ„ฐ์…‹์ด๋‹ค.

๋ชจ๋“  ์œ„์น˜์—์„œ ๋ชฉํ‘œ์ง€์ ๊ณผ์˜ ์Šค๋ƒ…์ƒท์„ ๋น„๊ตํ•˜์—ฌ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜๊ณ  ์ „ ์ง€์ ์—์„œ์˜ ์˜ค์ฐจ๋ฅผ ํ‰๊ท ํ•˜์—ฌ ํ‰๊ท ์˜ค์ฐจ๊ฐ (AAE: Average Angular Error)์„ ๊ณ„์‚ฐํ•˜์˜€๋‹ค. ์ด๋ ‡๊ฒŒ ๊ตฌํ•œ ํ‰๊ท  ์˜ค์ฐจ๊ฐ์€ ๋‹ค์–‘ํ•œ ๋ฐฉ์‹์„ ํ†ตํ•ด ๊ตฌํ•œ ํ™ˆ๋ฒกํ„ฐ์˜ ์ •ํ™•๋„๋ฅผ ํ‰๊ฐ€ํ•˜๋Š” ์ง€ํ‘œ๊ฐ€ ๋œ๋‹ค.

2.3.2 ์‹คํ—˜ ๊ฒฐ๊ณผ

2.3.2.1 ๊ฐ ๋ฐฉ๋ฒ•๋ณ„ ๊ฒฐ๊ณผ ๋ฐ ๋ถ„์„

ํ‘œ 1. ๊ฐ ๋ฐฉ๋ฒ•๋ณ„ ํ‰๊ท ์˜ค์ฐจ๊ฐ(AAE)

Table 1. Average angular errors (AAE) for each method

Method

Original

Hall

Using Corner Landmarks

9.76ยฐ

18.44ยฐ

Using Vertical Edge Landmarks

12.04ยฐ

11.90ยฐ

Using Brightness of Mask

12.30ยฐ

71.65ยฐ

์ฒซ๋ฒˆ์งธ ์ด๋ฏธ์ง€์…‹(Original)์—์„œ ์•ž์„œ ์–ธ๊ธ‰ํ•œ ์„ธ๊ฐ€์ง€ ๋ฐฉ๋ฒ•์„ ํ†ตํ•ด ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•œ ๊ฒฐ๊ณผ๋Š” ํ‘œ 1๊ณผ ๊ฐ™๋‹ค. ์–ธ๊ธ‰ํ•œ ์„ธ ๊ฐ€์ง€ ๋ฐฉ์‹ ๋ชจ๋‘ ์–‘ํ˜ธํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ด๊ณ  ์žˆ์Œ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ชจ๋‘ ์ ๋‹นํ•œ ์–‘์˜ ๋ฐ๊ธฐ๊ฐ€ ์œ ์ง€๋˜๋Š” ํ™˜๊ฒฝ(Original)์—์„œ๋Š” ์–‘ํ˜ธํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ด๊ณ  ์žˆ์ง€๋งŒ ์ƒ๋Œ€์ ์œผ๋กœ ์–ด๋‘์šด ํ™˜๊ฒฝ์—์„œ๋Š” ์ •ํ™•ํ•œ ํ™ˆ๋ฒกํ„ฐ ๊ณ„์‚ฐ์ด ์–ด๋ ค์šธ ์ˆ˜ ์žˆ๋‹ค. ๋ฐฉ๋ฒ• ์ค‘ ํŠนํžˆ ๋ฐ๊ธฐ๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๊ฒฝ์šฐ ํ‰๊ท ์˜ค์ฐจ๋Š” ํฌ๊ฒŒ ์ฆ๊ฐ€ํ•˜์˜€์œผ๋ฉฐ ์ด๋Š” ๋žœ๋“œ๋งˆํฌ๋กœ ์ธ์‹ํ• ๋งŒํ•œ ๋ฐ๊ธฐ์ •๋ณด๊ฐ€ ์ถฉ๋ถ„์น˜ ์•Š๊ธฐ ๋•Œ๋ฌธ์—์„œ ์˜ค๋Š” ๊ฒฐ๊ณผ๋กœ ๋ณด์ธ๋‹ค. ๋˜ํ•œ ์ฝ”๋„ˆ๋ฅผ ํŠน์ง•์ ์œผ๋กœ ์‚ฌ์šฉํ•˜๋Š” ๊ฒฝ์šฐ์—๋Š” ์ผ๋ฐ˜์ ์ธ ํ™˜๊ฒฝ์—์„œ๋Š” ๋งค์šฐ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋ณด์˜€์œผ๋‚˜, ์–ด๋‘์šด ํ™˜๊ฒฝ์—์„œ๋Š” ์„ฑ๋Šฅ์ด ์ €ํ•˜๋˜์—ˆ๊ณ  ์ˆ˜์ง ์—ฃ์ง€๋ฅผ ์ด์šฉํ•˜๋Š” ๋ฐฉ์‹์€ ๋‘ ๊ฐ€์ง€ ํ™˜๊ฒฝ (๋ฐ์ดํ„ฐ์…‹) ๋ชจ๋‘์—์„œ ์ƒ๋Œ€์ ์œผ๋กœ ์•ˆ์ •์ ์ธ ์„ฑ๋Šฅ์„ ๋ณด์˜€๋‹ค. ๋ฒกํ„ฐ ๊ณ„์‚ฐ ์†๋„๋Š” ์ค‘์š”ํ•œ ์„ฑ๋Šฅ์ง€ํ‘œ ์ค‘ ํ•˜๋‚˜์ด๋‹ค. 3๊ฐ€์ง€ ๋ฐฉ์‹์€ ์ƒํ˜ธ ๋‹ค๋ฅธ ๋žœ๋“œ๋งˆํฌ ์ˆ˜์™€ ์ข…๋ฅ˜๋ฅผ ์‚ฌ์šฉํ•˜๊ณ  ์žˆ๊ธฐ ๋•Œ๋ฌธ์—, ๊ฐ๊ธฐ ์š”๊ตฌ๋˜๋Š” ์‹œ๊ฐ„์ด ์ƒ์ดํ•˜๋‹ค. ์—ฐ์‚ฐํ™˜๊ฒฝ (i7-6700K 4.0GHz CPU)์—์„œ ์†Œ์š”๋œ ์‹œ๊ฐ„์€ ํ‘œ 2์™€ ๊ฐ™๋‹ค. ๊ฐ ๋ฐฉ๋ฒ•๋ณ„ ์—ฐ์‚ฐ์‹œ๊ฐ„์€ ๋ฐ๊ธฐ(๋ฐฉ๋ฒ•3), ํŠน์ง•์  ๋žœ๋“œ๋งˆํฌ(๋ฐฉ๋ฒ•2),์ˆ˜์ง์„  ๋žœ๋“œ๋งˆํฌ(๋ฐฉ๋ฒ•1) ๋ฐฉ์‹ ์ˆœ์œผ๋กœ ๋น ๋ฅธ ์†๋„๋ฅผ ๋ณด์ด๊ณ  ์žˆ๋‹ค.

ํ‘œ 2. ๊ฐ ๋ฐฉ๋ฒ•๋ณ„ ์—ฐ์‚ฐ์‹œ๊ฐ„

Table 2. Computing time for each method

Method

1 Point

Entire Point

Using Corner Landmark

0.0824(s)

7.8766(s)

Using Vertical Edge Landmarks

0.1812(s)

20.89(s)

Using Brightness of Mask

0.0183(s)

1.21(s)

2.3.2.2 ํ•™์Šต์„ ํ†ตํ•œ ๋ฐฉ์‹์กฐํ•ฉ

์•ž์—์„œ ์ˆ˜ํ–‰ํ•œ 3๊ฐ€์ง€ ๋ฐฉ์‹์œผ๋กœ ๊ตฌํ•œ ํ™ˆ๋ฒกํ„ฐ๋ฅผ Home์—์„œ ์ธ์ ‘ํ•œ 24๊ฐœ ์œ„์น˜์™€ ์ž„์˜๋กœ ์„ ์ •ํ•œ 24๊ฐœ ์œ„์น˜์—์„œ ๊ฐ๊ฐ ํ•™์Šตํ•œ ๊ฒฐ๊ณผ๋ฅผ ํ†ตํ•ด ๊ตฌํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ์ „์ฒด์œ„์น˜์—์„œ ์ ์šฉํ•˜์˜€๋‹ค. ๊ทธ๋ฆผ 8์€ ๋‘ ๊ฐ€์ง€ ํ•™์Šต์œ„์น˜๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค.

๊ทธ๋ฆผ. 7. ํ•™์Šต ์œ„์น˜ : ํ™ˆ ์ธ์ ‘ ์œ„์น˜(์ขŒ), ์ž„์˜ ์œ„์น˜(์šฐ)

Fig. 7. Training positions: Near home position (left), random Positions (right)

../../Resources/kiee/KIEE.2020.69.2.337/fig7.png

๊ทธ๋ฆผ. 8. ๊ฐ€๊นŒ์šด ์œ„์น˜(์ขŒ), ์ž„์˜์˜ ์œ„์น˜(์šฐ)์—์„œ์˜ ํ•™์Šต ์„ฑ๋Šฅ

Fig. 8. Training performances at close (left) / random (right) locations

../../Resources/kiee/KIEE.2020.69.2.337/fig8.png

๊ทธ๋ฆผ 7์˜ ๊ฐ ์œ„์น˜์—์„œ ๊ณ„์‚ฐํ•œ 3๊ฐ€์ง€ ๋ฐฉ์‹์˜ ๊ฒฐ๊ณผ๋ฅผ ์•ž์„œ ์–ธ๊ธ‰ํ•œ ๊ฒฝ์‚ฌํ•˜๊ฐ•๋ฒ•์„ ์ด์šฉํ•˜์—ฌ ์˜ค์ฐจ๋ฅผ ์ตœ์†Œ๋กœ ํ•˜๊ธฐ ์œ„ํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ๊ณ„์‚ฐํ•˜์˜€๋‹ค. ๊ทธ๋ฆผ 8์€ ์•ž์„œ ์–ธ๊ธ‰ํ•œ ๋‘ ๊ฐ€์ง€ ์œ„์น˜์—์„œ๊ตฌํ•œ ๊ฒฐ๊ณผ๋ฅผ ํ•™์Šตํ•˜๋Š” ๊ณผ์ •์—์„œ์˜ ์˜ค์ฐจ์˜ ๋ณ€ํ™”๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค. ํ•™์Šต์„ ๊ฑฐ๋“ญํ• ์ˆ˜๋ก ์ตœ์ข…์ ์œผ๋กœ ์—๋Ÿฌ๊ฐ€ ๊ฐ์†Œํ•˜๋‹ค๊ฐ€ ์ˆ˜๋ ดํ•จ์„ ์•Œ ์ˆ˜ ์žˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๊ฐ€์žฅ ๋‚ฎ์€ ์—๋Ÿฌ๊นŒ์ง€ ์ˆ˜๋ ดํ•œ ๊ฒฝ์šฐ๋Š” ์ธ์ ‘ํ•œ ์œ„์น˜์—์„œ ํ•™์Šตํ•œ ๊ฒฝ์šฐ์ด๋ฉฐ, ์ด๋Š” ๊ฐ€๊นŒ์šด ์œ„์น˜์—์„œ๋Š” ๊ฑฐ๋ฆฌ๊ฐ€ ๋จผ ์œ„์น˜๋ณด๋‹ค ์—๋Ÿฌ๊ฐ€ ์ ๊ฒŒ ๋ฐœ์ƒํ•˜๊ธฐ ๋•Œ๋ฌธ์ด๋ฉฐ, ํ•™์Šต์— ํ•„์š”ํ•œ ๊ณ„์‚ฐ (Iteration)๋„ ๋งค์šฐ ์ ๊ฒŒ ๋ฐœ์ƒํ•˜์˜€๋‹ค. ๋งŒ์•ฝ ์ธ์ ‘ํ•œ ๊ฐ๋„์—์„œ ํ•™์Šตํ•œ ๋ฐ์ดํ„ฐ๊ฐ€ ์ถฉ๋ถ„ํžˆ ์‹ ๋ขฐ๋„๊ฐ€ ์žˆ๋‹ค๋ฉด ๊ฐ€๋Šฅํ•œ ์ธ์ ‘ํ•œ ์œ„์น˜์—์„œ ํ•˜๋Š” ๋ฐฉ์‹์ด ๊ฐ€์žฅ ํšจ์œจ์ ์ผ ๊ฒƒ์ด๋‹ค. ๊ทธ๋ฆผ 9์™€ ํ‘œ 3์€ ๊ฐ๊ฐ์˜ ๋ฐฉ์‹์—์„œ ํ•™์Šตํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ๋ฐ์ดํ„ฐ์…‹์˜ ์ „์ฒด ์œ„์น˜์—์„œ ์ ์šฉํ•œ ๊ฒฐ๊ณผ์ด๋‹ค.

ํ‘œ 3. ๊ฐ ๋ฐฉ๋ฒ•๋ณ„ ์—ฐ์‚ฐ์‹œ๊ฐ„

Table 3. Computing time for each method

Method

AAE (Average Angular Error)

Test

Training

Close Positions

Min : 9.76ยฐ

Aver : 11.37ยฐ

6.27ยฐ

Random Positions

6.68ยฐ

๊ทธ๋ฆผ. 9. ํ•™์Šตํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ์ด์šฉํ•˜์—ฌ ์ „์ฒด ์œ„์น˜์— ์ ์šฉํ•œ ๊ฒฐ๊ณผ; ์ธ์ ‘ ์œ„์น˜ (์ขŒ), ๋žœ๋ค ์œ„์น˜ (์šฐ)์—์„œ ๊ฐ๊ฐ ํ•™์Šต

Fig. 9. Vector map using weight learning (training at close (left) and random (right) locations)

../../Resources/kiee/KIEE.2020.69.2.337/fig9.png

3๊ฐ€์ง€ ์œ„์น˜์—์„œ ํ•™์Šตํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ์ ์šฉ๊ฒฐ๊ณผ ํ•™์Šตํ•˜์ง€ ์•Š๋Š” ๊ฒฝ์šฐ์—์„œ ๋ฐœ์ƒํ•œ ๊ฐ ๋ฐฉ์‹๋“ค์˜ ์˜ค์ฐจ (AAE) ์ตœ์†Œ๊ฐ’๋ณด๋‹ค ์„ฑ๋Šฅ์ด ํ–ฅ์ƒ๋˜์—ˆ์Œ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์—ˆ๋‹ค. ๋˜ํ•œ ์ผ๋ถ€์œ„์น˜์—์„œ ํ•™์Šต ํ•œ ๊ฒฐ๊ณผ๋„ ์ถฉ๋ถ„ํžˆ ์ „์ฒด์œ„์น˜์—์„œ ๋™์ผํ•œ ๊ณผ์ •์„ ์ˆ˜ํ–‰ํ•œ ๋ฐฉ์‹๊ณผ ํฐ ์„ฑ๋Šฅ์ฐจ์ด๋ฅผ ๋ณด์ด์ง€ ์•Š์Œ์„ ์•Œ ์ˆ˜ ์žˆ์—ˆ๋‹ค.

3. ๊ฒฐ ๋ก 

๋ณธ ์—ฐ๊ตฌ๋ฅผ ํ†ตํ•ด ๋จผ์ € ์Šค๋ƒ…์ƒท์˜ ๋žœ๋“œ๋งˆํฌ์˜ ์œ„์น˜๋ฅผ ์ด์šฉํ•˜์—ฌ ๋‘ ์ง€์ ๊ฐ„์˜ ์ƒ๋Œ€์œ„์น˜๋ฅผ ๊ณ„์‚ฐํ•˜์˜€๋‹ค. ๋žœ๋“œ๋งˆํฌ๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹์€ ํฌ๊ฒŒ 3๊ฐ€์ง€๋กœ ํ—ˆํ”„๋ณ€ํ™˜์„ ์ด์šฉํ•˜์—ฌ ์ˆ˜์ง์„ ์„ ๊ฒ€์ถœํ•œ ํ›„ HOG๋ฅผ ํ†ตํ•ด ํŠน์ง•์  ๋งค์นญ์„ ํ•˜๋Š” ๋ฐฉ์‹ (๋ฐฉ๋ฒ• 1)๊ณผ SURF ๋“ฑ์„ ์ด์šฉํ•˜์—ฌ ํŠน์ง•์ ์„ ๋žœ๋“œ๋งˆํฌ๋กœ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹ (๋ฐฉ๋ฒ•2), ๊ทธ๋ฆฌ๊ณ  HaarยญLike Features๋ฅผ ํ™œ์šฉํ•˜์—ฌ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ์ถ”์ •ํ•˜๋Š” ๋ฐฉ์‹ (๋ฐฉ๋ฒ•3)์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ทธ ๊ฒฐ๊ณผ๋ฅผ ํ™•์ธํ•˜์˜€๋‹ค. ์‹คํ—˜๊ฒฐ๊ณผ ์ผ๋ฐ˜์ ์ธ ํ™˜๊ฒฝ์—์„œ๋Š” ๋Œ€๋ถ€๋ถ„ ์œ„์น˜์—์„œ ์ข‹์€ ์„ฑ๋Šฅ์„ ๋ณด์ž„์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์—ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋น›์ด ๋ถ€์กฑํ•œ ํ™˜๊ฒฝ์—์„œ๋Š” ๋ฐ๊ธฐ๋ณด๋‹ค๋Š” ๋žœ๋“œ๋งˆํฌ ํ˜•ํƒœ๋ฅผ ์ด์šฉํ•œ ๋ฐฉ์‹์ด, ์ˆ˜์ง์„ ์„ ์ฐพ๋Š” ๋ฐฉ์‹๋ณด๋‹ค ๋ชจ์„œ๋ฆฌ์™€ ๊ฐ™์€ ํŠน์ง•์ ์„ ์ฐพ๋Š” ๋ฐฉ์‹์ด ์ข‹์€ ์„ฑ๋Šฅ์„ ๋ณด์ž„์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์—ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๊ฐ๊ฐ์˜ ๋ฐฉ์‹์œผ๋กœ ์‚ฌ์šฉํ•œ ๊ฒฐ๊ณผ๋ฅผ ๊ณ„์‚ฐํ•œ ํ›„ ํ•™์Šต์„ ํ†ตํ•ด ์‹คํ—˜ํ™˜๊ฒฝ์—์„œ ์ตœ์ ์˜ ๊ฐ€์ค‘์น˜๋ฅผ ๊ตฌํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค.

ํ•™์Šต์„ ์œ„ํ•ด์„œ ์ผ๋ถ€์œ„์น˜์—์„œ ๊ณ„์‚ฐํ•œ ๊ฐ ๋ฐฉ์‹์˜ ALV ๋ฒกํ„ฐ์™€ ์ •๋‹ต์ง€๋ฅผ ํ™œ์šฉํ•˜์—ฌ ํ•ด๋‹น์œ„์น˜์—์„œ์˜ ์—๋Ÿฌ๋ฅผ ๊ณ„์‚ฐํ•œ ํ›„ ์ „์ฒด ์—๋Ÿฌ๊ฐ€ ์ตœ์†Œ๊ฐ€ ๋˜๋Š” ๊ฐ€์ค‘์น˜๋ฅผ ๊ณ„์‚ฐํ•˜๊ธฐ ์œ„ํ•ด ๊ฒฝ์‚ฌํ•˜๊ฐ•๋ฒ•์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ๊ณ„์‚ฐ๋œ ๊ฐ ๋ฐฉ์‹์˜ ๊ฐ€์ค‘์น˜๋ฅผ ๋ถ€์—ฌํ•˜์—ฌ ์ „์ฒด์ง€์ ์—์„œ ์ตœ์ข…์ ์ธ ํ™ˆ๋ฒกํ„ฐ๋ฅผ ๊ณ„์‚ฐํ•˜์˜€๋‹ค. ์‹คํ—˜๊ฒฐ๊ณผ ๊ฐ๊ฐ์˜ ์œ„์น˜์—์„œ ๊ณ„์‚ฐ๋œ ๊ฐ€์ค‘์น˜๋ฅผ ์‚ฌ์šฉํ•œ ๋ฐฉ์‹์ด ๊ทธ๋ ‡์ง€ ์•Š๋Š” ๊ฒฝ์šฐ๋ณด๋‹ค ์ข‹์€ ์„ฑ๋Šฅ์„ ๋ณด์ž„์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์—ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฐฉ์‹์€ ๊ฒฐ๊ณผ์ ์œผ๋กœ ํ™˜๊ฒฝ์— ๋”ฐ๋ผ ๋” ์ ํ•ฉํ•œ ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด ๋†’์€ ๊ฐ€์ค‘์น˜๋ฅผ ์‚ฌ์šฉํ•จ์œผ๋กœ์จ ์ •ํ™•๋„ ์ธก๋ฉด์—์„œ ๋” ๋‚˜์€ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค€๋‹ค.

๊ฐ ๋ฐฉ์‹์„ ์กฐํ•ฉํ•˜๋Š” ๋ฐฉ์‹์€ ํ–ฅ์ƒ๋œ ์„ฑ๋Šฅ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ๊ธฐ๋ณธ์ ์œผ๋กœ 3๊ฐ€์ง€ ๋ฐฉ์‹์˜ ALV ๋ฒกํ„ฐ๋ฅผ ๋ชจ๋‘ ๊ณ„์‚ฐํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๋†’์€ computing time์„ ํ•„์š”๋กœ ํ•œ๋‹ค๋Š” ๋‹จ์ ์ด ์žˆ๋‹ค. ๋˜ํ•œ ๊ธฐ๋ณธ์ ์œผ๋กœ ์‹ค์ œ ๋ณต๊ท€๋ฅผ ์œ„ํ•œ ๋‚ด๋น„๊ฒŒ์ด์…˜ ์‹œ์Šคํ…œ์„ ๋™์ž‘ํ•˜๊ธฐ ์ด์ „์— ์ผ์ •์ˆ˜ ์ด์ƒ์˜ ์ด๋™์„ ํ†ตํ•ด ํ™˜๊ฒฝ์— ๋Œ€ํ•œ ํ‰๊ฐ€๊ฐ€ ์š”๊ตฌ๋œ๋‹ค๋Š” ์ ๋„ ๋ณธ ์‹œ์Šคํ…œ์˜ ๋‹จ์ ์ด๋ผ ํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํ˜„์žฌ ํ™˜๊ฒฝ์— ๋Œ€ํ•œ ํ‰๊ฐ€ ๋ฐ ๊ฐ€์ค‘์น˜ ์ ์šฉ์„ ํ†ตํ•œ ์„ฑ๋Šฅ๊ฐœ์„ ์€ ์–ธ๊ธ‰ํ•œ ๋ฐฉ์‹ ์™ธ์—๋„ ๋‹ค๋ฅธ ๋งŽ์€ ๋ฐฉ์‹์„ ์ ์šฉํ•  ์ˆ˜ ์žˆ๊ณ , ์‹ค์ œ๋กœ ์–ป์€ ์„ฑ๋Šฅํ–ฅ์ƒ์ด ์ž‘์ง€ ์•Š๋‹ค๋Š” ์ ์—์„œ ๊ทธ ์˜์˜๊ฐ€ ์žˆ๋‹ค๊ณ  ํ•  ์ˆ˜ ์žˆ๋‹ค.

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๋‚˜์นจ๋ฐ˜์ด ์žˆ๋Š” ์‹œ์Šคํ…œ์„ ๊ฐ€์ •ํ•˜์—ฌ ๋ฌธ์ œ ์ ‘๊ทผ์ด ์ด๋ฃจ์–ด์ง€๊ณ  ์žˆ๋‹ค. ๋ฐฉํ–ฅ ์ •๋ณด๋ฅผ ์œ„ํ•ด์„œ๋Š” ๊ณค์ถฉ์˜ ํŽธ๊ด‘ ๋ฐฉ์‹์„ ๋ชจ๋ธ๋กœ ํ•˜๋Š” ์‹œ์Šคํ…œ์„ ์ด์šฉํ•  ์ˆ˜ ์žˆ๋‹ค [8,30]. ๊ฒฝ๋กœ ๋ˆ„์ ์„ ํ†ตํ•œ ์ฃผํ–‰ ์ •๋ณด, ๋‚˜์นจ๋ฐ˜ ์ •๋ณด, ํ†ต์‹  ์ •๋ณด, ๊ฑฐ๋ฆฌ ์ •๋ณด, ๋น„์ ผ ์ •๋ณด์˜ ๊ฒฐํ•ฉ์„ ํ†ตํ•ด ๋” ํšจ์œจ์ ์ธ ๋‚ด๋น„๊ฒŒ์ด์…˜ ์‹œ์Šคํ…œ์„ ๊ตฌํ˜„ํ•  ์ˆ˜ ์žˆ๊ณ  [6-8,31,32], ์ด์— ๋Œ€ํ•œ ์ถ”ํ›„ ์—ฐ๊ตฌ๋ฅผ ์ง„ํ–‰ํ•  ์˜ˆ์ •์ด๋‹ค.

Acknowledgements

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea Government (MSIT) (No. 2017R1A2B4011455).

References

1 
T. S. Collett, 1996, Insect navigation en route to the goal: multiple strategies for the use of landmarks, Journal of Experimental Biology, pp. 227-235Google Search
2 
T. S. Collett, 1992, Landmark learning and guidance in insects, Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, pp. 295-303DOI
3 
F. Papi, 1990, Olfactory navigation in birds, Experientia, Vol. 46, pp. 352-363DOI
4 
J. Rossier, C. Haeberli, F. Schenk, 2000, Auditory cues support place navigation in rats when associated with a visual cue, Behavioural brain research, Vol. 117, No. 1-2, pp. 209-214DOI
5 
M. Mรผller, R. Wehner, 1988, Path integration in desert ants, Cataglyphis fortis, Proceedings of the National Academy of Sciences, Vol. 85, No. 14, pp. 5287-5290DOI
6 
T. S. Collett, M. Collett, 2000, Path integration in insects, Current opinion in neurobiology, Vol. 10, No. 6, pp. 757-762DOI
7 
D. Kim, J. Lee, 2011, Path integration mechanism with coarse coding of neurons, Neural Processing Letters, Vol. 34, No. 3, pp. 277-291DOI
8 
S. Rossel, R. Wehner, 1984, How bees analyse the polarization patterns in the sky, Journal of Comparative Physiology A, Vol. 154, No. 5, pp. 607-615DOI
9 
T. Kimchi, A. S. Etienne, J. Terkel, 2004, A subterranean mammal uses the magnetic compass for path integration, Proceedings of the National Academy of Sciences, Vol. 101, No. 4, pp. 1105-1109DOI
10 
S. Judd, T. S. Collett, 1998, Multiple stored views and landmark guidance in ants, Nature, Vol. 392, pp. 710DOI
11 
S. ร…kesson, R. Wehner, 2002, Visual navigation in desert ants Cataglyphis fortis: are snapshots coupled to a celestial system of reference?, Journal of Experimental Biology, Vol. 205, No. 14, pp. 1971-1978Google Search
12 
B. A. Cartwright, T. S. Collett, 1987, Landmark maps for honeybees, Biological cybernetics, Vol. 57, No. 1-2, pp. 85-93DOI
13 
D. Kim, 2006, Neural network mechanism for the orientation behavior of sand scorpions towards prey, IEEE Transactions on Neural Networks, Vol. 17, No. 4, pp. 1070-1076DOI
14 
D. Kim, 2004, A spiking neuron model for synchronous flashing of fireflies, BioSystems, Vol. 76, No. 1-3, pp. 7-20DOI
15 
D. Lambrinos, R. Mรถller, T. Labhart, R. Pfeifer, R. Wehner, 2000, A mobile robot employing insect strategies for navigation, Robotics and Autonomous Systems, Vol. 30, pp. 39-64DOI
16 
C. Lee, S. Yu, D. Kim, 2017, Landmark-based homing navigation using omnidirectional depth information, Sensors, Vol. 17, No. 8, pp. 1928DOI
17 
V. V. Hafner, 2001, Adaptive homing - robotic exploration tours, Adaptive Behavior, Vol. 9, No. 3-4, pp. 131-141DOI
18 
K. Weber, S. Venkatesh, M. Srinivasan, 1999, Insect-inspired robotic homing, Adaptive Behavior, Vol. 7, pp. 65-97DOI
19 
S. Yu, D. Kim, 2011, Landmark vectors with quantized distance information for homing navigation, Adaptive Behavior, Vol. 19, No. 2, pp. 121-141DOI
20 
C. Lee, D. Kim, 2017, Local homing navigation based on the moment model for landmark distribution and features, Sensors, Vol. 17, No. 11, pp. 2658DOI
21 
M. Kim, D. Kim, 2019, Local Visual Homing Navigation Using Gradient-Descent Learning of Haar-like Features, Transactions of the Korean Institute of Electrical Engineers, Vol. 68, pp. 1244-1251DOI
22 
S. Baek, 2015, Local visual navigation using features in images, Master dissertation, Yonsei University, Seoul, KoreaGoogle Search
23 
H. Bay, T. Tuytelaars, L. van Gool, 2006, SURF: Speeded up robust features, in Proc. of European Conference on Computer Vision, pp. 404-417DOI
24 
P. Viola, M. Jones, 2001, Rapid object detection using a boosted cascade of simple features, in Proc. of Int. Conf. on Computer Vision & Pattern Recognition, Vol. 1, pp. 511-518DOI
25 
S. Baek, D. Kim, 2014, Snapshot homing navigation based on edge features, Proc. of Int. Conf. on Simulation of Adaptive Behavior, pp. 98-107DOI
26 
N. Dalal, B. Triggs, 2005, Histograms of oriented gradients for human detection, in Proc. of Int. Conf. on Computer Vision & Pattern Recognition, IEEE Computer Society, pp. 886-893DOI
27 
D. G. Lowe, 2004, Distinctive image features from scale-invariant keypoints, International Journal of Computer Vision, Vol. 60, No. 2, pp. 91-110DOI
28 
C. Lee, D. Kim, 2018, Visual Homing Navigation with Haar- Like Features in the Snapshot, IEEE Access, Vol. 6, pp. 33666-33681DOI
29 
D. Kim, J. Lee, 2000, Handling continuous-valued attributes in decision tree with neural network modeling, in Proc. of European Conf. on Machine Learning, pp. 211-219DOI
30 
E. Baek, D. Kim, 2020, Estimating Linear Polarization with Multi-Point Optic Sensing, Transactions of the Korean Institute of Electrical Engineers, Vol. 69, No. 1, pp. 99-106Google Search
31 
C. Lee, D. Kim, 2018, High-Order Moment Models of Landmark Distribution for Local Homing Navigation, IEEE Access, Vol. 6, pp. 72137-72152DOI
32 
S. L. Kim, W. Burgard, D. Kim, 2009, Wireless Communications in Networked Robotics: Editorial, IEEE Wireless Communications, Vol. 16, No. 1, pp. 4-5DOI

์ €์ž์†Œ๊ฐœ

์ตœ์ข…ํ•˜ (Jong-Ha Choi)
../../Resources/kiee/KIEE.2020.69.2.337/au1.png

Jong-Ha Choi received M.S. in the department of Electrical and Electronic Engineering, Yonsei University in 2020.

His research interests are biorobotics, robot navigation, and artificial intelligence.

๊น€๋Œ€์€ (DaeEun Kim)
../../Resources/kiee/KIEE.2020.69.2.337/au2.png

DaeEun Kim received his B.E. and M.S. in Seoul National University and the University of Michigan at Ann Arbor, respectively.

He received his Ph.D. degree from the University of Edinburgh in 2002.

Currently he is a professor at Yonsei University in South Korea.

His research interests are in the area of biorobotics, autonomous robots, artificial life, neural networks and neuroethology.