{"id":1634,"date":"2023-11-01T11:14:25","date_gmt":"2023-11-01T04:14:25","guid":{"rendered":"https:\/\/www.gdsonline.tech\/?p=1634"},"modified":"2024-04-04T08:19:49","modified_gmt":"2024-04-04T08:19:49","slug":"lts-gds-chu-thich-du-lieu-la-gi","status":"publish","type":"post","link":"https:\/\/www.gdsonline.tech\/vi\/lts-gds-chu-thich-du-lieu-la-gi\/","title":{"rendered":"Ch\u00fa th\u00edch d\u1eef li\u1ec7u l\u00e0 g\u00ec?"},"content":{"rendered":"

\"Ch\u00fa<\/h2>\n

<\/span>Ch\u00fa th\u00edch d\u1eef li\u1ec7u l\u00e0 g\u00ec?<\/span><\/h2>\n

Q\u00faa tr\u00ecnh g\u1eafn nh\u00e3n c\u00e1c lo\u1ea1i d\u1eef li\u1ec7u nh\u01b0 h\u00ecnh \u1ea3nh, video, v\u0103n b\u1ea3n, \u00e2m thanh \u0111\u01b0\u1ee3c g\u1ecdi l\u00e0 ch\u00fa th\u00edch d\u1eef li\u1ec7u, qu\u00e1 tr\u00ecnh n\u00e0y gi\u00fap m\u00e1y h\u1ecdc nh\u1eadn d\u1ea1ng c\u00e1c v\u1eadt th\u1ec3 v\u00e0 t\u1eeb \u0111\u00f3 \u0111\u01b0a ra c\u00e1c d\u1ef1 \u0111o\u00e1n trong t\u01b0\u01a1ng lai.<\/p>\n

<\/span>T\u1ea1i sao c\u1ea7n ch\u00fa th\u00edch d\u1eef li\u1ec7u?<\/b><\/span><\/h2>\n

Nh\u1eefng d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c ch\u00fa th\u00edch s\u1ebd l\u00e0 ph\u1ea7n quan tr\u1ecdng c\u1ee7a\u00a0m\u00e1y h\u1ecdc c\u00f3 gi\u00e1m s\u00e1t b\u1edfi c\u00e1c m\u00f4 h\u00ecnh m\u00e1y h\u1ecdc n\u00e0y c\u1ea7n m\u1ed9t s\u1ed1 l\u01b0\u1ee3ng l\u1edbn\u00a0d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c ch\u00fa th\u00edch \u0111\u1ec3 ph\u00e1t hi\u1ec7n v\u1eadt th\u1ec3 v\u00e0 \u0111\u01b0a ra quy\u1ebft \u0111\u1ecbnh. \u0110\u00f3 c\u0169ng l\u00e0 l\u00fd do t\u1ea1i sao ch\u1ea5t l\u01b0\u1ee3ng ch\u00fa th\u00edch d\u1eef li\u1ec7u l\u00e0 nh\u00e2n t\u1ed1 h\u00e0ng \u0111\u1ea7u t\u00e1c \u0111\u1ed9ng \u0111\u1ebfn \u0111\u1ed9 ch\u00ednh x\u00e1c v\u00e0 m\u1ee9c \u0111\u1ed9 t\u01b0\u01a1ng quan\u00a0c\u1ee7a c\u00e1c d\u1ef1 \u0111o\u00e1n.<\/p>\n

<\/span>M\u1ed9t s\u1ed1 lo\u1ea1i d\u1eef li\u1ec7u ph\u1ed5 bi\u1ebfn<\/b><\/span><\/h2>\n

Trong th\u1ef1c t\u1ebf, c\u00f3 \u0111a d\u1ea1ng c\u00e1c lo\u1ea1i d\u1eef li\u1ec7u c\u1ea7n \u0111\u01b0\u1ee3c ch\u00fa th\u00edch nh\u01b0 h\u00ecnh \u1ea3nh, video, \u00e2m thanh, v\u0103n b\u1ea3n v\u00e0 d\u1eef li\u1ec7u c\u1ea3m bi\u1ebfn 3D:<\/p>\n

<\/span>H\u00ecnh \u1ea3nh<\/span><\/h3>\n

C\u00e1c h\u00ecnh \u1ea3nh c\u1ea7n \u0111\u01b0\u1ee3c ch\u00fa th\u00edch nh\u01b0 con ng\u01b0\u1eddi, \u0111\u1ed3 v\u1eadt, \u0111\u1ed9ng v\u1eadt…Lo\u1ea1i d\u1eef li\u1ec7u n\u00e0y l\u00e0 ph\u1ed5 bi\u1ebfn nh\u1ea5t v\u00e0 chi\u1ebfm t\u1ef7 l\u1ec7 l\u1edbn trong c\u00e1c d\u1ef1 \u00e1n. \u0110\u1eb7c bi\u1ec7t trong m\u1ed9t s\u1ed1 ng\u00e0nh nh\u01b0 \u00f4 t\u00f4 t\u1ef1 l\u00e1i, c\u00f4ng ngh\u1ec7 t\u1ef1 \u0111\u1ed9ng h\u00f3a trang thi\u1ebft b\u1ecb y t\u1ebf y\u00eau c\u1ea7u m\u1ed9t s\u1ed1 l\u01b0\u1ee3ng kh\u1ed5ng l\u1ed3 d\u1eef li\u1ec7u d\u1ea1ng \u1ea3nh c\u1ea7n ch\u00fa th\u00edch. V\u1edbi d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c ch\u00fa th\u00edch ch\u00ednh x\u00e1c v\u00e0 hi\u1ec7u qu\u1ea3, h\u1ec7 th\u1ed1ng AI c\u00f3 th\u1ec3 t\u1ef1 h\u1ecdc t\u1eeb h\u00ecnh \u1ea3nh ch\u00fa th\u00edch \u0111\u1ec3 v\u1eadn h\u00e0nh m\u1ed9t c\u00e1ch ho\u00e0n ch\u1ec9nh m\u00e0 kh\u00f4ng c\u1ea7n s\u1ef1 h\u1ed7 tr\u1ee3 t\u1eeb con ng\u01b0\u1eddi.<\/p>\n

<\/span>Video<\/span><\/h3>\n

D\u1ea1ng d\u1eef li\u1ec7u n\u00e0y l\u00e0 nh\u1eefng \u0111o\u1ea1n ghi h\u00ecnh t\u1eeb camera quan s\u00e1t \u0111\u01b0\u1ee3c chia nh\u1ecf th\u00e0nh c\u00e1c khung h\u00ecnh. Kh\u00e1c v\u1edbi h\u00ecnh \u1ea3nh, video l\u00e0 m\u1ed9t chu\u1ed7i c\u00e1c h\u00ecnh \u1ea3nh li\u00ean ti\u1ebfp nhau, m\u1ed7i h\u00ecnh \u1ea3nh s\u1ebd c\u00f3 s\u1ef1 thay \u0111\u1ed5i r\u1ea5t \u00edt so v\u1edbi h\u00ecnh \u1ea3nh tr\u01b0\u1edbc \u0111\u00f3. \u0110i\u1ec1u n\u00e0y \u0111\u00f2i h\u1ecfi nh\u00e2n vi\u00ean x\u1eed l\u00fd d\u1eef li\u1ec7u\u00a0ph\u1ea3i ch\u00fa th\u00edch\u00a0c\u1ea9n th\u1eadn v\u00e0 ch\u00ednh x\u00e1c \u0111\u1ec3 th\u1ec3 hi\u1ec7n \u0111\u01b0\u1ee3c t\u00ednh li\u00ean t\u1ee5c c\u1ee7a chuy\u1ec3n \u0111\u1ed9ng trong c\u00e1c khung h\u00ecnh video.<\/p>\n

<\/span>V\u0103n b\u1ea3n<\/span><\/span><\/h3>\n

V\u1edbi d\u1ea1ng d\u1eef li\u1ec7u v\u0103n b\u1ea3n, ch\u1eef hay con s\u1ed1 \u1edf \u0111a d\u1ea1ng c\u00e1c ng\u00f4n ng\u1eef khi \u0111\u01b0\u1ee3c ch\u00fa th\u00edch s\u1ebd gi\u00fap m\u00e1y h\u1ecdc hi\u1ec3u \u0111\u01b0\u1ee3c ng\u1eef c\u1ea3nh c\u1ee7a n\u1ed9i dung qua t\u1eebng ch\u1eef, t\u1eebng c\u00e2u, t\u1eebng \u0111o\u1ea1n nh\u01b0 t\u00ean qu\u1ed1c gia, t\u00ean ng\u01b0\u1eddi, c\u1ea3m x\u00fac hay th\u00e1i \u0111\u1ed9. D\u1ea1ng n\u00e0y s\u1ebd \u0111\u01b0\u1ee3c \u00e1p d\u1ee5ng ph\u1ed5 bi\u1ebfn nh\u1ea5t khi x\u1eed l\u00fd ng\u00f4n ng\u1eef t\u1ef1 nhi\u00ean (NLP).<\/p>\n

<\/span>\u00c2m thanh<\/span><\/h3>\n

Nh\u1eefng \u0111o\u1ea1n \u00e2m thanh \u0111\u01b0\u1ee3c ghi \u00e2m v\u1edbi ng\u00f4n ng\u1eef \u1edf m\u1ed7i \u0111\u1ea5t n\u01b0\u1edbc, v\u00f9ng mi\u1ec1n t\u1ef1 nhi\u00ean kh\u00e1c nhau s\u1ebd \u0111\u01b0\u1ee3c ch\u00fa th\u00edch c\u00e1c th\u00f4ng tin li\u00ean quan. \u00c2m thanh \u0111\u00f3 c\u00f3 th\u1ec3 l\u00e0 b\u1ea5t k\u1ef3 lo\u1ea1i ti\u1ebfng \u1ed3n, \u00e2m thanh n\u00e0o nh\u01b0 ti\u1ebfng kh\u00f3c, ti\u1ebfng la h\u00e9t,\u2026 t\u1ea5t c\u1ea3 \u0111\u00e3 \u0111\u01b0\u1ee3c nghi\u00ean c\u1ee9u v\u00e0 \u1ee9ng d\u1ee5ng r\u1ea5t nhi\u1ec1u trong l\u0129nh v\u1ef1c gi\u1ecdng n\u00f3i AI.<\/p>\n

<\/span>D\u1eef li\u1ec7u c\u1ea3m bi\u1ebfn 3D<\/span><\/h3>\n

Lo\u1ea1i d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c thu th\u1eadp t\u1eeb c\u1ea3m bi\u1ebfn 3D th\u01b0\u1eddng ph\u1ee9c t\u1ea1p h\u01a1n d\u1eef li\u1ec7u t\u1eeb camera th\u00f4ng th\u01b0\u1eddng. C\u1ea3m bi\u1ebfn 3D c\u00f3 th\u1ec3 ho\u1ea1t \u0111\u1ed9ng ngay c\u1ea3 trong b\u00f3ng t\u1ed1i v\u00ec ch\u00fang s\u1eed d\u1ee5ng ngu\u1ed3n \u00e1nh s\u00e1ng h\u1ed3ng ngo\u1ea1i. B\u00ean c\u1ea1nh \u0111\u00f3, c\u1ea3m bi\u1ebfn 3D c\u00f3 th\u1ec3 \u01b0\u1edbc t\u00ednh k\u00edch th\u01b0\u1edbc v\u1eadt th\u1ec3 c\u00f3 \u0111\u1ed9 ch\u00ednh x\u00e1c cao h\u01a1n camera.<\/p>\n

<\/span>C\u00e1c ki\u1ec3u ch\u00fa th\u00edch d\u1eef li\u1ec7u<\/b><\/span><\/h2>\n

C\u00f3 r\u1ea5t nhi\u1ec1u ki\u1ec3u ch\u00fa th\u00edch d\u1eef li\u1ec7u kh\u00e1c nhau trong v\u00f4 v\u00e0n c\u00e1c d\u1ef1 \u00e1n, t\u00f9y thu\u1ed9c v\u00e0o y\u00eau c\u1ea7u c\u1ee7a kh\u00e1ch h\u00e0ng v\u00e0 v\u1eadt th\u1ec3 c\u1ea7n d\u00f2 t\u00ecm s\u1ebd quy\u1ebft \u0111\u1ecbnh lo\u1ea1i ch\u00fa th\u00edch d\u1eef li\u1ec7u \u0111\u01b0\u1ee3c s\u1eed d\u1ee5ng. D\u1ef1a v\u00e0o lo\u1ea1i d\u1eef li\u1ec7u c\u1ea7n ch\u00fa th\u00edch, c\u00f4ng c\u1ee5 ch\u00fa th\u00edch v\u00e0 k\u1ef9 thu\u1eadt s\u1ebd x\u00e1c \u0111\u1ecbnh lo\u1ea1i ch\u00fa th\u00edch d\u1eef li\u1ec7u.<\/p>\n

\"\"<\/p>\n

<\/span>Bounding box<\/span><\/h3>\n

L\u00e0 m\u1ed9t khung h\u00ecnh ch\u1eef nh\u1eadt bao quanh \u0111\u1ed1i t\u01b0\u1ee3ng trong \u1ea3nh. N\u00f3 \u0111\u01b0\u1ee3c d\u00f9ng \u0111\u1ec3 x\u00e1c \u0111\u1ecbnh v\u1ecb tr\u00ed \u0111\u1ed1i t\u01b0\u1ee3ng, ph\u00e2n lo\u1ea1i \u0111\u1ed1i t\u01b0\u1ee3ng nh\u01b0 \u00f4 t\u00f4, con ng\u01b0\u1eddi ho\u1eb7c tr\u1ea1ng th\u00e1i. \u0110\u00e2y l\u00e0 lo\u1ea1i ch\u00fa th\u00edch \u0111\u01a1n gi\u1ea3n v\u00e0 d\u1ec5 th\u1ef1c hi\u1ec7n nh\u1ea5t.<\/p>\n

<\/span>Polygon<\/span><\/h3>\n

V\u1edbi c\u00e1c lo\u1ea1i v\u1eadt th\u1ec3 c\u1ea7n ch\u00fa th\u00edch y\u00eau c\u1ea7u \u0111\u1ed9 ch\u00ednh x\u00e1c cao nh\u01b0 con ng\u01b0\u1eddi, logo th\u01b0\u01a1ng hi\u1ec7u, bi\u1ec3n b\u00e1o \u0111\u01b0\u1eddng ph\u1ed1 th\u00ec m\u1ed9t \u0111\u01b0\u1eddng \u0111a gi\u00e1c bao quanh v\u1eadt th\u1ec3 s\u1ebd ph\u00f9 h\u1ee3p. \u0110\u00e2y l\u00e0 lo\u1ea1i ch\u00fa th\u00edch d\u1eef li\u1ec7u linh ho\u1ea1t nh\u1ea5t \u0111\u1ec3 ch\u00fa th\u00edch \u0111a d\u1ea1ng c\u00e1c lo\u1ea1i d\u1eef li\u1ec7u, cung c\u1ea5p m\u1ed9t b\u1ee9c tranh ch\u00ednh x\u00e1c v\u1ec1 h\u00ecnh d\u1ea1ng v\u00e0 k\u00edch th\u01b0\u1edbc c\u1ee7a v\u1eadt th\u1ec3, t\u1eeb \u0111\u00f3 cho ph\u00e9p m\u00e1y h\u1ecdc \u0111\u01b0a ra d\u1ef1 \u0111o\u00e1n k\u1ebft qu\u1ea3 ch\u00ednh x\u00e1c h\u01a1n.<\/p>\n

<\/span>Polyline & Spline<\/span><\/h3>\n

Lo\u1ea1i ch\u00fa th\u00edch n\u00e0y l\u00e0 \u0111\u01b0\u1eddng k\u1ebb th\u1eb3ng c\u00f3 \u0111i\u1ec3m b\u1eaft \u0111\u1ea7u v\u00e0 k\u1ebft th\u00fac, th\u01b0\u1eddng \u0111\u01b0\u1ee3c s\u1eed d\u1ee5ng \u0111\u1ec3 x\u00e1c \u0111\u1ecbnh c\u1ea5u tr\u00fac tuy\u1ebfn t\u00ednh chi ti\u1ebft trong h\u00ecnh \u1ea3nh, video nh\u01b0 l\u00e0n \u0111\u01b0\u1eddng, \u0111\u01b0\u1eddng ray v\u00e0 \u0111\u01b0\u1eddng \u1ed1ng. Lo\u1ea1i ch\u00fa th\u00edch n\u00e0y s\u1ebd gi\u00fap m\u00e1y h\u1ecdc trong ng\u00e0nh c\u00f4ng ngh\u1ec7 \u00f4 t\u00f4 t\u1ef1 l\u00e1i c\u00f3 th\u1ec3 ph\u00e1t hi\u1ec7n l\u00e0n \u0111\u01b0\u1eddng v\u00e0 x\u00e1c \u0111\u1ecbnh l\u00e0n \u0111\u01b0\u1eddng ph\u00f9 h\u1ee3p trong qu\u00e1 tr\u00ecnh di chuy\u1ec3n.<\/p>\n

<\/span>3D Cuboid<\/span><\/h3>\n

Kh\u1ed1i l\u1eadp ph\u01b0\u01a1ng bao quanh m\u1ed9t \u0111\u1ed1i t\u01b0\u1ee3ng trong h\u00ecnh \u1ea3nh 2D, nh\u1eb1m ch\u00fa th\u00edch v\u1eadt th\u1ec3 trong kh\u00f4ng gian 3 chi\u1ec1u v\u1edbi \u0111\u1ed9 s\u00e2u c\u1ee7a \u0111\u1ed1i t\u01b0\u1ee3ng. Lo\u1ea1i n\u00e0y th\u01b0\u1eddng \u0111\u01b0\u1ee3c s\u1eed d\u1ee5ng nhi\u1ec1u nh\u1ea5t trong c\u00f4ng ngh\u1ec7 xe t\u1ef1 l\u00e1i, c\u00f3 th\u1ec3 gi\u00fap \u0111o kho\u1ea3ng c\u00e1ch \u0111\u1ebfn ch\u01b0\u1edbng ng\u1ea1i v\u1eadt.<\/p>\n

<\/span>Semantic segmentation<\/span><\/h3>\n

L\u00e0 nhi\u1ec7m v\u1ee5 ph\u00e2n v\u00f9ng t\u1eebng lo\u1ea1i \u0111\u1ed1i t\u01b0\u1ee3ng nh\u1eb1m nh\u1eadn d\u1ea1ng t\u1ea5t c\u1ea3 c\u00e1c pixel thu\u1ed9c c\u00f9ng m\u1ed9t l\u1edbp. M\u1ed7i pixel trong \u1ea3nh s\u1ebd \u0111\u01b0\u1ee3c g\u00e1n nh\u00e3n t\u01b0\u01a1ng \u1ee9ng b\u1eb1ng c\u00e1c m\u00e0u s\u1eafc bi\u1ec3u th\u1ecb kh\u00e1c nhau. S\u1eed d\u1ee5ng lo\u1ea1i ch\u00fa th\u00edch d\u1eef li\u1ec7u n\u00e0y gi\u00fap thu\u1eadt to\u00e1n m\u00e1y h\u1ecdc ph\u00e2n lo\u1ea1i c\u00e1c t\u00ednh n\u0103ng c\u1ee5 th\u1ec3 c\u1ee7a \u0111\u1ed1i t\u01b0\u1ee3ng. M\u1ed9t v\u00ed d\u1ee5 ti\u00eau bi\u1ec3u l\u00e0 lo\u1ea1i ch\u00fa th\u00edch d\u1eef li\u1ec7u n\u00e0y \u0111\u01b0\u1ee3c \u00e1p d\u1ee5ng cho qu\u00e1 tr\u00ecnh ki\u1ec3m so\u00e1t quy tr\u00ecnh c\u00f4ng nghi\u1ec7p nh\u1eb1m ph\u00e1t hi\u1ec7n s\u1ef1 thi\u1ebft h\u1ee5t hay l\u00e3ng ph\u00ed.<\/p>\n

<\/span>Landmark<\/span><\/h3>\n

V\u1edbi landmark (keypoint), c\u00e1c \u0111\u1ed1i t\u01b0\u1ee3ng trong \u1ea3nh s\u1ebd \u0111\u01b0\u1ee3c \u0111\u00e1nh d\u1ea5u c\u00e1c \u0111i\u1ec3m quan tr\u1ecdng nh\u1ea5t. Ph\u01b0\u01a1ng ph\u00e1p n\u00e0y r\u1ea5t h\u1eefu \u00edch \u0111\u1ec3 nh\u1eadn d\u1ea1ng khu\u00f4n m\u1eb7t, c\u1ea3m x\u00fac, c\u1eed ch\u1ec9 c\u1ee7a con ng\u01b0\u1eddi. Nh\u00e2n vi\u00ean x\u1eed l\u00fd d\u1eef li\u1ec7u c\u00f3 th\u1ec3 s\u1eed d\u1ee5ng ki\u1ec3u ch\u00fa th\u00edch n\u00e0y k\u1ebft h\u1ee3p v\u1edbi c\u00e1c lo\u1ea1i ch\u00fa th\u00edch d\u1eef li\u1ec7u kh\u00e1c \u0111\u1ec3 t\u1ea1o b\u1ea3n \u0111\u1ed3 \u0111i\u1ec3m x\u00e1c \u0111\u1ecbnh t\u01b0 th\u1ebf c\u1ee7a con ng\u01b0\u1eddi.<\/p>\n

<\/span>LiDAR point cloud<\/span><\/h3>\n

L\u00e0 c\u00f4ng ngh\u1ec7 s\u1eed d\u1ee5ng \u00e1nh s\u00e1ng \u1edf d\u1ea1ng ch\u00f9m tia laser \u0111\u1ec3 \u0111o kho\u1ea3ng c\u00e1ch gi\u1eefa c\u1ea3m bi\u1ebfn v\u00e0 \u0111\u1ed1i t\u01b0\u1ee3ng. Qu\u00e1 tr\u00ecnh n\u00e0y s\u1ebd l\u1eb7p l\u1ea1i h\u00e0ng tri\u1ec7u l\u1ea7n b\u1edfi c\u1ea3m bi\u1ebfn LiDAR, cu\u1ed1i c\u00f9ng s\u1ebd \u0111\u01b0\u1ee3c th\u1ec3 hi\u1ec7n d\u01b0\u1edbi d\u1ea1ng \u0111\u00e1m m\u00e2y \u0111i\u1ec3m.\u00a0C\u00e1c \u0111\u1ed1i t\u01b0\u1ee3ng trong \u0111\u00e1m m\u00e2y \u0111i\u1ec3m c\u00f3 th\u1ec3 \u0111\u01b0\u1ee3c ch\u00fa th\u00edch b\u1eb1ng bounding box, kh\u1ed1i 3D cuboid…t\u00f9y thu\u1ed9c v\u00e0o y\u00eau c\u1ea7u c\u1ee7a t\u1eebng d\u1ef1 \u00e1n. Lo\u1ea1i ch\u00fa th\u00edch \u0111\u01b0\u1ee3c \u1ee9ng d\u1ee5ng nhi\u1ec1u nh\u1ea5t trong c\u00f4ng ngh\u1ec7 xe t\u1ef1 l\u00e1i.<\/p>\n

<\/span>C\u00e1c c\u00f4ng c\u1ee5 ch\u00fa th\u00edch d\u1eef li\u1ec7u ph\u1ed5 bi\u1ebfn<\/b><\/span><\/h2>\n

Trong c\u00e1c d\u1ef1 \u00e1n ch\u00fa th\u00edch d\u1eef li\u1ec7u, m\u1ed9t s\u1ed1 c\u00f4ng c\u1ee5 ph\u1ed5 bi\u1ebfn gi\u00fap nh\u00e2n vi\u00ean x\u1eed l\u00fd d\u1eef li\u1ec7u th\u1ef1c hi\u1ec7n qu\u00e1 tr\u00ecnh l\u00e0m vi\u1ec7c hi\u1ec7u qu\u1ea3 h\u01a1n nh\u01b0:<\/p>\n